Language of document : ECLI:EU:T:2022:19

JUDGMENT OF THE GENERAL COURT (Fourth Chamber, Extended Composition)

26 January 2022 (*)

(Competition – Abuse of dominant position – Microprocessors market – Decision finding an infringement of Article 102 TFEU and of Article 54 of the EEA Agreement – Loyalty rebates – ‘Naked’ restrictions – Characterisation as abuse – As-efficient-competitor analysis – Overall strategy – Single and continuous infringement)

In Case T‑286/09 RENV,

Intel Corporation, Inc., established in Wilmington, Delaware (United States), represented by A. Parr, Solicitor, D. Beard QC and J. Williams, Barrister,

applicant,

supported by

Association for Competitive Technology, Inc., established in Washington, DC (United States), represented by J.-F. Bellis and K. Van Hove, lawyers,

intervener,

v

European Commission, represented by T. Christoforou, V. Di Bucci, N. Khan and M. Kellerbauer, acting as Agents,

defendant,

supported by

Union fédérale des consommateurs – Que choisir (UFC – Que choisir), established in Paris (France), represented by E. Nasry, lawyer,

intervener,

APPLICATION under Article 263 TFEU seeking, primarily, the annulment of Commission Decision C(2009) 3726 final of 13 May 2009 relating to a proceeding under Article [102 TFEU] and Article 54 of the EEA Agreement (Case COMP/C‑3/37.990 – Intel) or, in the alternative, annulment or reduction of the fine imposed on the applicant,

THE GENERAL COURT (Fourth Chamber, Extended Composition),

composed of H. Kanninen, President, J. Schwarcz (Rapporteur), C. Iliopoulos, I. Reine and B. Berke, Judges,

Registrar: E. Artemiou, Administrator,

having regard to the written part of the procedure and further to the hearing on 10 to 12 March 2020,

gives the following

Judgment

 Background to the dispute

1        Intel Corporation, Inc., (‘the applicant’ or ‘Intel’) is a company established under United States law which designs, develops, manufactures and markets central processing units (‘CPUs’), chipsets and other semiconductor components, as well as platform solutions for data processing and communications devices.

2        At the end of 2008, Intel employed around 94 100 people worldwide. In 2007, Intel had net revenues of 38 334 million United States dollars (USD) and a net income of USD 6 976 million. In 2008, Intel had net revenues of USD 37 586 million and a net income of USD 5 292 million.

 Administrative procedure

3        On 18 October 2000, Advanced Micro Devices, Inc. (‘AMD’) filed a formal complaint with the Commission of the European Communities under Article 3 of Council Regulation (EEC) No 17 of 6 February 1962, First Regulation implementing Articles [101] and [102 TFEU] (OJ, English Special Edition, 1959-1962 (I), p. 87), which it further supplemented by putting forward new facts and allegations, in the context of a supplementary complaint of 26 November 2003.

4        In May 2004, the Commission launched a series of investigations relating to elements in AMD’s supplementary complaint. Within the framework of that investigation, in July 2005, the Commission, assisted by several national competition authorities, carried out on-the-spot inspections pursuant to Article 20(4) of Council Regulation (EC) No 1/2003 of 16 December 2002 on the implementation of the rules on competition laid down in Articles [101] and [102 TFEU] (OJ 2003 L 1, p. 1) at four Intel locations in Germany, Spain, Italy and the United Kingdom, as well as at the locations of several Intel customers in Germany, Spain, France, Italy and the United Kingdom.

5        On 17 July 2006, AMD filed a complaint with the Bundeskartellamt (Federal Cartel Office, Germany), in which it claimed that Intel had engaged in, inter alia, exclusionary marketing arrangements with Media-Saturn-Holding GmbH (‘MSH’), a European retailer of microelectronic devices and the largest desktop computer distributor in Europe. The Federal Cartel Office exchanged information with the Commission on that subject, pursuant to Article 12 of Regulation No 1/2003.

6        On 23 August 2006, the Commission held a meeting with D1 [confidential], (1) a customer of Intel. The Commission did not place the indicative list of topics for that meeting in the case file and did not take minutes of it. A member of the team responsible for the file at the Commission drafted a note concerning that meeting, which was described as internal by the Commission. On 19 December 2008, the Commission provided the applicant with a non-confidential version of that note.

7        On 26 July 2007, the Commission notified to the applicant a statement of objections (‘the Statement of Objections of 2007’) concerning its conduct vis-à-vis five major original equipment manufacturers (‘OEMs’), namely Dell, Hewlett-Packard Company (HP), Acer Inc., NEC Corp. and International Business Machines Corp. (IBM). Intel replied to that statement of objections on 7 January 2008 and an oral hearing was held on 11 and 12 March 2008. Intel was granted access to the file on three occasions, namely on 31 July 2007, 23 July and 19 December 2008.

8        The Commission undertook several investigative measures relating to AMD’s allegations, including on-the-spot inspections at the sites of several computer retailers and Intel locations in February 2008. In addition, it addressed several written requests for information, pursuant to Article 18 of Regulation No 1/2003, to a number of major OEMs.

9        On 17 July 2008, the Commission issued to the applicant a supplementary statement of objections concerning its conduct vis-à-vis MSH. That statement of objections (‘the Supplementary Statement of Objections of 2008’) also covered Intel’s conduct vis-à-vis Lenovo Group Ltd (‘Lenovo’) and included new evidence relating to Intel’s conduct vis-à-vis some of the OEMs covered by the Statement of Objections of 2007, which had been obtained by the Commission following the publication of the latter.

10      The Commission originally set Intel a deadline of eight weeks to submit its reply to the Supplementary Statement of Objections of 2008. On 15 September 2008, that deadline was extended to 17 October 2008 by the Hearing Officer.

11      Intel did not reply to the Supplementary Statement of Objections of 2008 within the prescribed period. By contrast, on 10 October 2008, Intel lodged an application with the General Court, registered under reference T‑457/08, requesting the Court, first, to annul two decisions of the Commission relating to the setting of the period for replying to the Supplementary Statement of Objections of 2008 and to the Commission’s refusal to obtain several categories of documents emanating from, inter alia, the file of the private litigation between Intel and AMD in the State of Delaware (United States) and, second, to extend the deadline for lodging its reply to the Supplementary Statement of Objections of 2008 in order to have a period of 30 days from the day on which it obtained access to the relevant documents.

12      Intel also lodged an application for interim measures, registered under reference T‑457/08 R, by which it sought to obtain the suspension of the Commission’s procedure pending the judgment on its substantive application, as well as the suspension of the period set for lodging its reply to the Supplementary Statement of Objections of 2008 and, in the alternative, that it be allowed a period of 30 days from the date of that judgment in order to reply to the Supplementary Statement of Objections of 2008.

13      On 19 December 2008, the Commission sent Intel a letter drawing its attention to a number of specific items of evidence which the Commission intended to use in a potential final decision (‘the Letter of Facts’). Intel did not reply to that letter by the deadline of 23 January 2009.

14      On 27 January 2009, the President of the General Court dismissed the application for interim measures by order of 27 January 2009, Intel v Commission (T‑457/08 R, not published, EU:T:2009:18). Following that order, on 29 January 2009, Intel proposed to file its reply to the Supplementary Statement of Objections of 2008 and to the Letter of Facts within 30 days of the order of the President of the Court.

15      On 2 February 2009, the Commission informed Intel by letter that its services had decided not to grant it an extension of the deadlines for a reply to the Supplementary Statement of Objections of 2008 or to the Letter of Facts. That letter also stated that the Commission services were nevertheless willing to consider the possible relevance of belated written submissions, provided that Intel served such submissions by 5 February 2009. Finally, the Commission expressed the view that it was not obliged to grant a request for a hearing lodged out of time, and that its services considered that the proper conduct of the administrative procedure did not make the preparation of an oral hearing necessary.

16      On 3 February 2009, Intel withdrew its action in the main proceedings in Case T‑457/08 and the case was removed from the Register by order of the President of the Fifth Chamber of the General Court of 24 March 2009.

17      On 5 February 2009, Intel served a written submission including observations relating to the Supplementary Statement of Objections of 2008 and the Letter of Facts, which it classed as a ‘reply to the Supplementary Statement of Objections [of 2008]’ and a ‘reply to the [Letter of Facts]’.

18      On 10 February 2009, Intel wrote to the Hearing Officer and asked to be granted an oral hearing in relation to the Supplementary Statement of Objections of 2008. The Hearing Officer rejected that request by letter of 17 February 2009.

19      On 13 May 2009, the Commission adopted Decision C(2009) 3726 final relating to a proceeding under Article [102 TFEU] and Article 54 of the EEA Agreement (Case COMP/C‑3/37.990 – Intel) (‘the contested decision’), a summary of which appears in the Official Journal of the European Union (OJ 2009 C 227, p. 13).

 The contested decision

20      According to the contested decision, Intel committed a single and continuous infringement of Article 102 TFEU and of Article 54 of the Agreement on the European Economic Area (EEA), from October 2002 until December 2007, by implementing a strategy aimed at foreclosing a competitor, AMD, from the market for x86 CPU microprocessors (‘x86 CPUs’).

 Relevant market

21      The goods at issue in the contested decision are CPUs, which are key components of any computer, both in terms of overall performance and cost of the system. CPUs are often referred to as a computer’s ‘brain’. The process for the manufacture of CPUs requires expensive high-tech facilities.

22      CPUs used in computers can be subdivided into two categories, namely x86 CPUs and CPUs based on another architecture. The x86 architecture is a standard designed by Intel for its CPUs. It can run both the Windows and Linux operating systems. Windows is primarily linked to the instruction set corresponding to the x86 architecture. Prior to 2000, a number of manufacturers of x86 CPUs were present on the market. However, most of those manufacturers have since exited the market. The contested decision states that, since then, Intel and AMD have been essentially the only two companies still manufacturing x86 CPUs.

23      The Commission’s inquiry led to the conclusion that the relevant product market was not wider than the market of x86 CPUs. The contested decision leaves open the question of whether there is a single x86 CPU market for all computers or whether it is necessary to draw a distinction between three x86 CPU markets, namely the market for desktop computers, the market for notebook computers and the market for servers. According to the contested decision, given Intel’s market shares for each segment, there is no difference to the conclusion on dominance.

24      The geographical market was defined as worldwide.

 Dominance

25      In the contested decision, the Commission finds that, in the 10-year period which was examined (1997 to 2007), Intel consistently held market shares in excess of or around 70%. Furthermore, according to the contested decision, there are significant barriers to entry and expansion in the x86 CPU market. Those barriers arise from sunk investments in research and development, intellectual property and the production facilities which are necessary to produce x86 CPUs. In consequence, all of Intel’s competitors, except AMD, have exited the market or are left with an insignificant share.

26      On the basis of Intel’s market shares and the barriers to entry and expansion in the relevant market, the contested decision concludes that, at least in the period covered by that decision, that is to say, from October 2002 to December 2007, Intel held a dominant position in that market.

 Abuse and fine

27      The contested decision describes two types of Intel conduct vis-à-vis its trading partners, namely conditional rebates and naked restrictions.

28      First, according to the contested decision, Intel granted to four OEMs, namely Dell, Lenovo, HP and NEC, rebates which were conditional on those OEMs purchasing all or almost all of their x86 CPUs from Intel. Similarly, Intel awarded payments to MSH which were conditional on MSH selling exclusively computers containing Intel’s x86 CPUs.

29      The contested decision concludes that the conditional rebates granted by Intel constitute fidelity rebates. With regard to Intel’s conditional payments to MSH, the contested decision establishes that the economic mechanism of those payments is equivalent to that of the conditional rebates granted to the OEMs.

30      The contested decision also conducts an economic analysis of the capability of the rebates to foreclose a hypothetical competitor as efficient as Intel (as-efficient-competitor test; ‘the AEC test’), albeit not dominant. More precisely, the analysis establishes at what price a competitor as efficient as Intel would have had to offer CPUs in order to compensate an OEM for the loss of an Intel rebate. The same kind of analysis was conducted for the Intel payments to MSH.

31      The evidence gathered by the Commission led it to conclude that Intel’s conditional rebates and payments induced the loyalty of the key OEMs and of MSH. The effects of these practices were complementary, in that they significantly diminished competitors’ ability to compete on the merits of their x86 CPUs. Intel’s anticompetitive conduct thereby resulted in a reduction of consumer choice and in lower incentives to innovate.

32      Second, with regard to naked restrictions, the Commission states that Intel awarded three OEMs, namely HP, Acer and Lenovo, payments which were conditional on those OEMs postponing or cancelling the launch of products with AMD x86 CPUs or placing restrictions on the distribution of those products. The contested decision concludes that Intel’s conduct also directly harmed competition, and did not constitute normal competition on the merits.

33      The Commission concludes in the contested decision that, in each instance, Intel’s conduct vis-à-vis the OEMs mentioned above and MSH constitutes an abuse under Article 102 TFEU, but that each of those individual abuses is also part of a single strategy aimed at foreclosing AMD, Intel’s only significant competitor, from the market for x86 CPUs. Those individual abuses, it found, are therefore part of a single infringement of Article 102 TFEU.

34      By applying the Guidelines on the method of setting fines imposed pursuant to Article 23(2)(a) of Regulation No 1/2003 (OJ 2006 C 210, p. 2), the Commission imposed on the applicant a fine of EUR 1.06 billion.

 Operative Part

35      The operative part of the contested decision reads as follows:

Article 1

Intel … has committed a single and continuous infringement of Article [102 TFEU] and Article 54 of the EEA Agreement from October 2002 until December 2007 by implementing a strategy aimed at foreclosing competitors from the market of x86 CPUs which consisted of the following elements:

(a)      Granting rebates to Dell between December 2002 and December 2005 at a level that was conditional on Dell obtaining all of its x86 CPU supplies from Intel;

(b)      Granting rebates to HP between November 2002 and May 2005 at a level that was conditional on HP obtaining at least 95% of its corporate desktop x86 CPU supplies from Intel;

(c)      Granting rebates to NEC between October 2002 and November 2005 at a level that was conditional on NEC obtaining at least 80% of its client PC x86 CPU supplies from Intel;

(d)      Granting rebates to Lenovo between January 2007 and December 2007 at a level that was conditional on Lenovo obtaining all of its notebook x86 CPU supplies from Intel;

(e)      Granting payments to [MSH] between October 2002 and December 2007 at a level that was conditional on [MSH] selling only computers incorporating Intel x86 CPUs;

(f)      Granting payments to HP between November 2002 and May 2005 conditional on: (i) HP directing HP’s AMD-based x86 CPU business desktops to Small and Medium Business and Government, and Educational and Medical customers rather than to enterprise business customers; (ii) precluding HP’s channel partners from stocking HP’s AMD-based x86 CPU business desktops such that such desktops would only be available to customers by ordering them from HP (either directly or via HP channel partners acting as sales agent); and (iii) HP delaying the launch of its AMD-based x86 CPU business desktop in the [Europe, Middle East and Africa] region by six months;

(g)      Granting payments to Acer between September 2003 and January 2004 conditional on Acer delaying an AMD-based x86 CPU notebook;

(h)      Granting payments to Lenovo between June 2006 and December 2006 conditional on Lenovo delaying and finally cancelling its AMD-based x86 CPU notebooks.

Article 2

For the infringement referred to in Article 1, a fine of EUR 1 060 000 000 is imposed on Intel …

Article 3

Intel … shall immediately bring to an end the infringement referred to in Article 1 in so far as it has not already done so.

Intel … shall refrain from repeating any act or conduct described in Article 1, and from any act or conduct having the same or equivalent object or effect.

…’

 Procedure before the General Court and the Court of Justice

36      By application lodged at the Registry of the General Court on 22 July 2009, the applicant brought an action for annulment of the contested decision, relying on nine pleas in law.

37      By document lodged at the Registry on 14 October 2009, AMD sought leave to intervene in the present proceedings in support of the Commission. However, on 16 November 2009, AMD informed the Court that it was withdrawing its application to intervene in the case. In consequence, by order of the President of the Eighth Chamber of the General Court of 5 January 2010, AMD was removed from the case as an applicant for leave to intervene.

38      By document lodged at the Registry on 30 October 2009, the Union fédérale des consommateurs – Que choisir (UFC – Que choisir) (‘UFC’) sought leave to intervene in the present proceedings in support of the Commission. By order of 7 June 2010, the President of the Eighth Chamber of the General Court granted that leave to intervene. By letter lodged at the Registry on 22 September 2010, UFC informed the Court that it would not be lodging a statement in intervention, but that it would make oral submissions at the hearing.

39      By document lodged at the Registry on 2 November 2009, the Association for Competitive Technology, (‘ACT’) sought leave to intervene in the present proceedings in support of Intel. By order of 7 June 2010, the President of the Eighth Chamber of the General Court granted that leave to intervene. ACT submitted its statement in intervention within the prescribed period and the main parties submitted their observations on that statement.

40      Intel and the Commission requested that certain confidential matters contained in the application, defence, reply, rejoinder and in their observations on the statements in intervention respectively not be communicated to the interveners UFC and ACT. They provided a joint non-confidential version of those various pleadings. Only the non-confidential version of those pleadings was communicated. The interveners raised no objections in that regard.

41      Following a change in the composition of the Chambers of the General Court in September 2010, the Judge-Rapporteur was elected President of the Seventh Chamber, to which the present case was accordingly allocated.

42      By decision of 18 January 2012, the Court referred the case to the Seventh Chamber sitting in extended composition, pursuant to Article 14(1) and Article 51(1) of the Rules of Procedure of the General Court.

43      The parties presented oral argument at the hearing, which was held from 3 to 6 July 2012.

44      By judgment of 12 June 2014, Intel v Commission (T‑286/09, ‘the initial judgment’, EU:T:2014:547), the Court dismissed the action in its entirety.

45      In support of its first plea in law, in relation to horizontal issues concerning the legal assessments carried out by the Commission, Intel disputed the allocation of the burden of proof and the standard of proof required, the legal characterisation of the rebates and payments granted in consideration of exclusive supply as well as the legal characterisation of the payments, which the Commission referred to as ‘naked restrictions’, made to the OEMs so that they would delay, cancel or restrict the marketing of products equipped with AMD CPUs. Intel claimed, in particular, that the Commission’s analysis of the evidence did not satisfy the required standard. Accordingly, it argued, the Commission had failed to prove that Intel’s rebate agreements were conditional upon its customers purchasing all or almost all of their x86 CPU requirements from Intel. In addition, the Commission used an AEC test to determine whether Intel’s rebates could restrict competition, but it made numerous errors in its analysis and assessment of the evidence relating to the application of that test.

46      The Court held, in essence, in paragraph 79 of the initial judgment, that the rebates granted to Dell, HP, NEC and Lenovo were exclusivity rebates, since they were conditional upon customers purchasing from Intel either all their x86 CPU requirements or most of their requirements. In addition, the Court stated, in paragraphs 80 to 89 of the initial judgment, that the question whether such a rebate can be characterised as abusive does not depend on an analysis of the circumstances of the case aimed at establishing the capacity of that rebate to restrict competition.

47      For the sake of completeness, the Court considered, in paragraphs 172 to 197 of the initial judgment, that the Commission had established, to the requisite legal standard and on the basis of an analysis of the circumstances of the case, that the exclusivity rebates and payments granted by Intel to Dell, HP, NEC, Lenovo and MSH were capable of restricting competition.

48      As regards the second plea in law, alleging that the Commission did not establish its territorial jurisdiction to apply Articles 101 and 102 TFEU to the practices implemented in relation to Acer and Lenovo, the Court first of all considered, in paragraph 244 of the initial judgment, that, in order to justify the Commission’s jurisdiction under public international law, it was sufficient to establish either the qualified effects of the practice in the European Union or that it was implemented in the European Union. It then held, in paragraph 296 of the initial judgment, that the substantial, immediate and foreseeable effect that Intel’s conduct was capable of having within the European Economic Area (EEA) justified the Commission’s jurisdiction. Lastly, for the sake of completeness, the Court held, in paragraph 314 of the initial judgment, that that jurisdiction was also justified on account of the implementation of the conduct in question in the territory of the European Union and the EEA.

49      In support of its third plea in law, alleging that the Commission committed procedural irregularities, Intel submitted, inter alia, that its rights of defence had been infringed because of the absence of a record transcribing the meeting with D1, arguing that some of the evidence provided in that meeting could have been used as exculpatory evidence. Intel also maintained that the Commission had wrongly refused to hold a second hearing or to provide Intel with certain AMD documents which could have been relevant for Intel’s defence.

50      First, the Court considered, in paragraph 618 of the initial judgment, that the meeting in question did not constitute formal questioning for the purposes of Article 19 of Regulation No 1/2003 and that the Commission was not required to carry out such questioning. It therefore concluded, in that paragraph, that Article 3 of Commission Regulation (EC) No 773/2004 of 7 April 2004 relating to the conduct of proceedings by the Commission pursuant to Articles [101] and [102 TFEU] (OJ 2004 L 123, p. 18) was not applicable, with the result that the argument alleging infringement of the formal requirements laid down in that provision was ineffective.

51      Second, the Court held, in paragraphs 621 and 622 of the initial judgment, that, even though the Commission had infringed the principle of sound administration by failing to draw up a document containing a brief summary of the subjects addressed at that meeting and the names of the participants, it had nevertheless remedied that initial omission by making available to Intel the non-confidential version of an internal note relating to that meeting.

52      As regards the fourth plea in law, alleging errors of assessment concerning the practices relating to the various OEMs and MSH, the Court rejected all of the complaints raised by Intel in relation to Dell, HP, NEC, Lenovo, Acer and MSH in paragraphs 665, 894, 1032, 1221, 1371 and 1463 of the initial judgment.

53      As regards the fifth plea in law, by which Intel disputed the existence of an overall strategy aimed at foreclosing AMD’s access to the most important sales channels, the Court held, in paragraphs 1551 and 1552 of the initial judgment, that the Commission had, in essence, demonstrated to the requisite legal standard that Intel had attempted to conceal the anticompetitive nature of its practices and had implemented a long-term comprehensive strategy to foreclose AMD from those sales channels.

54      As regards the sixth plea in law, alleging that the Commission incorrectly applied the Guidelines on the method of setting fines imposed pursuant to Article 23(2)(a) of Regulation No 1/2003, the Court considered, inter alia, in paragraph 1598 of the initial judgment, that neither the principle of legal certainty nor the principle that offences and punishments are to be strictly defined by law precluded the Commission from deciding to adopt and apply new guidelines on the method of setting fines, even after the infringement has been committed. In addition, the Court took the view, in that paragraph, that effective enforcement of the competition rules justifies an undertaking being required to take account of the possibility of a modification to the general competition policy of the Commission as regards fines with respect both to the method of calculation and to the level of fines.

55      As regards the seventh plea in law, alleging the absence of an intentional or negligent infringement of Article 102 TFEU, the Court held, in essence, in paragraphs 1602 and 1603 of the initial judgment, that Intel could not have been unaware of the anticompetitive nature of its conduct and that the evidence relied on in the contested decision demonstrated to the requisite legal standard that Intel had implemented a long-term comprehensive strategy to foreclose AMD from the strategically most important sales channels and that it had attempted to conceal the anticompetitive nature of its conduct.

56      As regards the eighth plea in law, concerning the allegedly disproportionate nature of the fine, the Court held, in paragraphs 1614 to 1616 of the initial judgment, that the Commission’s practice in previous decisions could not serve as a legal framework for the fines imposed in competition matters and that, in any event, the decisions relied on in that respect by Intel were not relevant as regards observance of the principle of equal treatment. Furthermore, contrary to Intel’s assertions, the Court pointed out, in paragraphs 1627 and 1628 of the initial judgment, that the Commission had not taken into consideration the actual impact of the infringement on the market in order to determine its gravity.

57      As regards, lastly, the ninth plea in law, which sought annulment or reduction of the fine imposed on the applicant in the exercise of the Court’s unlimited jurisdiction, the Court held, inter alia, in paragraph 1647 of the initial judgment, that there was nothing in the complaints, arguments or matters of law and of fact put forward by Intel from which it might be concluded that the fine which had been imposed on it was disproportionate. The Court considered, in that paragraph, that the fine was appropriate to the circumstances of the case and emphasised out that it was well below the 10% ceiling set in Article 23(2) of Regulation No 1/2003.

58      By document lodged at the Registry of the Court of Justice on 26 August 2014, the applicant appealed against the initial judgment.

59      The applicant put forward six grounds in support of its appeal. By the first ground of appeal, it submitted that the General Court had erred in law by failing to examine the rebates at issue in the light of all the relevant circumstances. By the second ground of appeal, the applicant submitted that the General Court had erred in law in assessing the finding of an infringement in 2006 and 2007, inter alia as regards the assessment of the market coverage of the rebates at issue in those two years. By the third ground of appeal, the applicant argued that the General Court had erred in law as regards the legal characterisation of the exclusivity rebates which Intel concluded with HP and Lenovo. By the fourth ground of appeal, it submitted that the General Court had wrongly concluded that there was no material procedural irregularity affecting Intel’s rights of defence in the Commission’s treatment of the interview with D1. By the fifth ground of appeal, Intel argued that the General Court had misapplied the tests in relation to the Commission’s jurisdiction over the agreements concluded between Intel and Lenovo in 2006 and 2007. Lastly, by the sixth ground of appeal, the applicant asked the Court of Justice to annul or to reduce substantially the fine imposed on it, having regard to the principle of proportionality and the principle that the Guidelines on the method of setting fines imposed pursuant to Article 23(2)(a) of Regulation No 1/2003 should not be applied retroactively.

60      The Commission contended that the appeal should be dismissed. ACT argued that the appeal should be upheld in its entirety.

61      By its judgment of 6 September 2017, Intel v Commission (C‑413/14 P, ‘the judgment on the appeal’, EU:C:2017:632), as rectified, the Court of Justice set aside the initial judgment and referred the case back to the General Court.

 Procedure and forms of order sought following the referral back

62      The case was allocated to the Fourth Chamber (Extended Composition) of the General Court.

63      On 14, 15 and 16 November 2017 respectively, ACT, the applicant and the Commission submitted their written observations on the referral back, pursuant to Article 217(1) of the Rules of Procedure (‘the main observations’).

64      The applicant and the Commission requested that certain confidential matters contained in their respective main observations not be communicated to the interveners. They produced a non-confidential version of those various pleadings. The communication of those pleadings was restricted to that non-confidential version. The interveners raised no objections in that regard.

65      On 20 February 2018, ACT and, on 5 March 2018, the applicant and the Commission, respectively, lodged supplementary written observations pursuant to Article 217(3) of the Rules of Procedure (‘the supplementary observations’).

66      The applicant and the Commission requested that certain confidential matters contained in their supplementary observations not be communicated to the interveners. They produced a joint non-confidential version of those various pleadings. The communication of those pleadings was restricted to that non-confidential version. The interveners raised no objections in that regard.

67      In its observations, the applicant, supported by ACT, claims that the Court should:

–        annul the contested decision in whole or in part;

–        in the alternative, annul or reduce substantially the amount of the fine imposed;

–        order the Commission to pay the applicant’s costs.

68      In its observations, the Commission contends, in essence, that the Court should dismiss the action.

69      By letters of 7, 15 and 28 October 2019, Intel and the Commission partially waived confidentiality for the purposes of the hearing and the decision closing the proceedings, provided that no confidential document was served on the interveners. They stated, in essence, that all of the data in the file could be discussed in open court, with two exceptions, namely that the Court refrain from disclosing details regarding server X [confidential] and that the names of the natural persons mentioned in the pleadings should not be made public.

70      By letter of 27 January 2020, ACT applied for leave to participate in the part of the hearing which was to be held in camera because of the confidentiality of the information to be discussed, in accordance with the General Court’s decision of 10 December 2019.

71      By letter of 6 March 2020, UFC informed the Court that it was waiving its right to participate in the hearing scheduled from 10 to 12 March 2020 (‘the 2020 hearing’).

72      At the 2020 hearing, the President of the Fourth Chamber (Extended Composition), referred to the letter of 27 January 2020 lodged by ACT, in which ACT requested to participate in the hearing, which was initially to be held in camera. The President of the Fourth Chamber (Extended Composition), decided to add that letter to the file. However, since the 2020 hearing took place entirely in open court, the President stated that there was no longer any need to respond to ACT’s request. In addition, the President confirmed that the names of the natural persons would not be disclosed in open court or in the decision closing the proceedings.

73      Following the death of Judge Berke on 1 August 2021, the three judges whose signatures feature on the present judgment continued the deliberations, in accordance with Article 22 and Article 24(1) of the Rules of Procedure.

 Law

 Arguments of the parties relating to the subject matter of the proceedings following the referral back

74      As a preliminary point, it should be noted that, by its written observations submitted pursuant to Article 217(1) and (3) of the Rules of Procedure, the applicant withdrew the pleas in law in the action based on the Commission’s jurisdiction and procedural irregularities, which therefore no longer form part of the subject matter of the proceedings following the referral back.

75      The parties disagree as to the scope of the dispute following the referral back as regards the other pleas in the action.

76      The applicant, supported by ACT, submits, in essence, that, having regard to the fact that the initial judgment was set aside in its entirety, the General Court is required to issue a new judgment reconsidering all the grounds and arguments put forward in its application, other than those which it has waived, taking account of the legal framework laid down by the Court of Justice in the judgment on the appeal. The applicant adds, first, that the judgment on the appeal substantially clarified the framework within which the factual and economic evidence must be considered and, second, that the fact that the first ground of appeal has been upheld necessarily affects the basis for assessment of that evidence and the terms in which the contested decision is worded.

77      By contrast, the Commission contends, in essence, that, the findings of the initial judgment are final where they are unrelated to any error of law identified by the Court of Justice in its judgment on the appeal. According to the Commission, that is the case, in particular, where the findings of the General Court have not been challenged in the appeal or where the Court of Justice has dismissed the challenge to the findings in the initial judgment. The Commission adds, first, that it is apparent from paragraphs 147 and 149 of the judgment on the appeal that, in the context of the referral back the General Court is required to consider only the capability of the rebates to restrict competition and, second, that it is clear from paragraphs 109, 137 and 138 of the judgment on the appeal that the starting point for that assessment is the unappealed factual finding that the rebates in question are loyalty rebates. In the alternative, the Commission also submits that, were the General Court to decide to reconsider all the pleas in law and arguments put forward by the applicant in its action, there would be no reason for the General Court to reach different conclusions from those set out in its initial judgment on matters that were not raised on appeal.

78      In the present case, first, the question therefore arises as to whether, in the context of the referral back, all the pleas in law and arguments raised by the applicant in its action are once again brought before the General Court or whether, as, in essence, the Commission maintains, findings contained in the initial judgment may be regarded as having the force of res judicata.

79      The answer to that question is, as the applicant correctly submitted at the 2020 hearing, determined by the wording of the operative part of the judgment on the appeal.

80      It should be recalled that, once the Court of Justice has set aside a judgment and referred the case back to the General Court, the latter Court is seised, pursuant to Article 215 of the Rules of Procedure, of the case by the judgment of the Court of Justice and must rule again on all the pleas in law in support of annulment raised by the applicant, apart from those elements of the operative part not set aside by the Court of Justice and the considerations on which those elements are essentially founded, as those elements have acquired the authority of res judicata (judgment of 14 September 2011, Marcuccio v Commission, T‑236/02, EU:T:2011:465, paragraph 83).

81      In that regard, it must be stated that point 1 of the operative part of the judgment on the appeal sets aside the entirety of the initial judgment, since it states that it ‘sets aside’ that judgment.

82      It follows that, as the applicant and ACT maintain, the General Court must, in the present case, rule again on all the grounds and arguments put forward by the parties at first instance, with the exception of those referred to in paragraph 74 above concerning the Commission’s jurisdiction and procedural irregularities, which the applicant has expressly withdrawn.

83      Second, under Article 61 of the Statute of the Court of Justice of the European Union, where the appeal is well founded and the case is referred back to the General Court for judgment, the latter is bound by the decision of the Court of Justice on points of law. Consequently, as the Commission pointed out and as ACT essentially stated at the 2020 hearing, there is nothing, in principle, to preclude the Court to which the case is referred back from making the same assessment as the first-instance Court as regards the pleas in law and arguments which were not examined in the grounds of the judgment on the appeal. In that situation, there is no ‘decision of the Court of Justice on points of law’ within the meaning of the second paragraph of Article 61 of the Statute of the Court of Justice of the European Union, which would bind the referring Court (judgment of 14 September 2011, Marcuccio v Commission, T‑236/02, EU:T:2011:465, paragraph 86).

84      In the present case, it must be stated that the only error which justified the setting aside of the initial judgment lies in the fact that ‘in its analysis of whether the rebates at issue were capable of restricting competition, the General Court wrongly failed to take into consideration Intel’s line of argument seeking to expose alleged errors committed by the Commission in the AEC test’ (judgment on the appeal, paragraph 147).

85      It follows that, in accordance with the case-law cited in paragraphs 80 and 83 above, the General Court is required to consider, in the context of the referral back, whether the rebates at issue had the capacity to restrict competition in the light of (i) the clarifications contained in paragraphs 133 and 141 of the judgment on the appeal regarding the principles laid down in the judgment of 13 February 1979, Hoffmann-La Roche v Commission (85/76, EU:C:1979:36), and (ii) the parties’ main and supplementary observations on the conclusions to be drawn from those clarifications. Therefore, while the Court is required to consider Intel’s arguments which seek to demonstrate that the Commission made errors in its AEC analysis, it may, as to the remainder, accept, in the context of its examination, all the findings which are not called into question in the context of the appeal or, in any event, those relating to ‘points of law not decided by the Court’ in the judgment on the appeal.

86      This applies, in particular, to the findings on the legal characterisation of the so-called ‘naked restrictions’, under Title II ‘The heads of claim for annulment of the contested decision’, Section A, entitled ‘Horizontal questions concerning the legal assessments carried out by the Commission’, part 3, entitled ‘The legal characterisation of the so-called “naked restrictions”, in the initial judgment (paragraphs 198 to 220 of that judgment), and the findings on the naked restrictions and the existence of exclusivity rebates, also under Title II, but in Section D, entitled ‘Errors of assessment concerning the practices relating to the various OEMs and MSH’ of the initial judgment (paragraphs 437 to 1522 of that judgment).

87      As regards the findings concerning naked restrictions, Intel and ACT submit, in their main observations, that it is apparent from the judgment on the appeal that the Commission ought to have examined, in the contested decision, whether the naked restrictions were capable of producing the foreclosure effects alleged against the applicant, by applying the factors listed in paragraph 139 of the judgment on the appeal and the AEC test. ACT adds that naked restrictions are ultimately a form of exclusivity rebate or exclusivity payment and that the principle of legal certainty dictates that those two pricing practices are not to be distinguished from one another.

88      It should be noted at the outset that it is apparent from recital 1641 et seq. of the contested decision that a distinction was drawn between Intel’s actions towards Acer, HP and Lenovo, which are characterised as naked restrictions, and Intel’s other actions, which, for their part, were covered by the AEC test applied in the contested decision. In that regard, that distinction arises from the fact that the naked restrictions are based on two pillars, the second of which distinguishes them from Intel’s other actions referred to in the contested decision. Apart from the fact that Intel offered rebates or payments to the OEMs concerned (HP, Acer and Lenovo respectively), they were requested to abstain from certain specific actions as consideration for those payments, namely to cancel or restrict in one way or another the marketing of certain products with AMD-based CPUs.

89      More specifically, as those practices were described by the Court in paragraph 198 of the initial judgment, Intel granted payments subject to the following conditions:

–        first, HP was to direct HP’s AMD-based x86 CPU corporate desktops to small and medium-sized business (SMB) and the Government, Educational and Medical (‘GEM’) customers rather than to enterprise business customers;

–        second, HP was to preclude its channel partners from stocking HP’s AMD-based x86 CPU corporate desktops so that such desktops would only be available to customers by ordering them from HP either directly or via HP channel partners acting as sales agent;

–        third, Acer, HP and Lenovo were to delay or cancel the launch of computers equipped with AMD CPUs.

90      In the light of the foregoing, first of all, the General Court observes that nothing in the judgment on the appeal permits the inference that the Court of Justice held that the method defined in paragraph 138 et seq. of the judgment on the appeal should also apply to naked restrictions. Similarly, it does not in any way follow from that judgment that the Court of Justice required an AEC test be carried out in relation to naked restrictions, as the applicant maintained at first instance. Although those practices were clearly distinguished both in the Commission’s contested decision and in the initial judgment of the General Court, it is clear that the judgment on the appeal does not analyse those practices as such, but merely refers to them in paragraphs 11 and 15 thereof, in the context of the background to the dispute and the summary of the procedure before the General Court, without further assessment.

91      As the Commission correctly observes, the Court of Justice’s summary of the first ground of appeal and its statement of reasons for its assessment in paragraph 137 et seq. of its judgment support the argument that the Court of Justice in no way assessed the naked restrictions at issue. It is clear from those paragraphs that the Court of Justice assessed only the fidelity rebates within the meaning of the judgment of 13 February 1979, Hoffmann-La Roche v Commission (85/76, EU:C:1979:36).

92      That is all the more evident from reading paragraph 141 of the judgment on the appeal, which states that ‘if, in a decision finding a rebate scheme abusive, the Commission carries out [an AEC] analysis, the General Court must examine all of the applicant’s arguments seeking to call into question the validity of the Commission’s findings concerning the foreclosure capability of the rebate concerned’. Since the Commission failed entirely to carry out any AEC analysis of the naked restrictions and since the General Court, in essence, endorsed that approach in paragraphs 198 to 220 of the initial judgment, there is no doubt that the Court of Justice was referring to the ‘legal test’ to be applied to the rebates and payments granted to the OEMs and MSH respectively and not to naked restrictions.

93      Accordingly, contrary to what Intel and ACT assert, it is not apparent from the judgment on the appeal that naked restrictions ought to be subject, as regards establishing their abusive nature, to the same principles as the rebates at issue.

94      Next, contrary to ACT’s assertion, the naked restrictions as identified by the Commission are characterised by the fact that they are practices based on two pillars, the second of which distinguishes them from Intel’s other actions referred to in the contested decision, as stated in paragraph 88 above. Consequently, the principle of legal certainty does not preclude conditional rebates and naked restrictions from being subject to different legal tests, on the ground that a distinction between those two types of conduct is not likely to be applied consistently by the competent authorities and courts.

95      Lastly, in their main and supplementary observations, Intel and ACT do not put forward any argument capable of demonstrating that certain factual matters considered in the initial judgment relating to naked restrictions ought to be reconsidered following the referral back.

96      In those circumstances, the Court considers that it is necessary to accept the findings in paragraphs 198 to 220, 799 to 873, 1043 to 1144, 1222 to 1361 and 1371 of the initial judgment, solely in so far as they concern naked restrictions and their unlawfulness under Article 102 TFEU.

97      As regards the assessments concerning the rebates at issue being characterised as ‘exclusivity rebates’ under Title II, Section D, of the initial judgment, the General Court also considers that those should be accepted. First, they were not examined in the grounds of the judgment on the appeal and cannot therefore be regarded as a point of law decided by the Court of Justice, as provided for in the second paragraph of Article 61 of the Statute of the Court of Justice of the European Union. Second, it must be observed, as the Commission did at the 2020 hearing, that the applicant has not put forward any argument capable of demonstrating that certain factual matters considered in the initial judgment for the purposes of characterising the rebates at issue as ‘exclusivity rebates’ ought to be re-examined, in particular following the clarifications provided in the judgment on the appeal with regard to the principles laid down in the judgment of 13 February 1979, Hoffmann-La Roche v Commission (85/76, EU:C:1979:36).

98      In those circumstances, the Court considers it appropriate to accept the considerations set out in Title II, Section D, of the initial judgment that the Commission, in the contested decision, established that (i) the applicant had informed Dell that, from December 2002 to December 2005, the level of the rebates granted under a meet-competition programme was subject to an exclusivity condition (initial judgment, paragraphs 444 to 584), (ii) the two agreements concluded between the applicant and HP between November 2002 and May 2005 (‘the HPA agreements’) consisted of exclusivity rebates (initial judgment, paragraphs 673 to 798), (iii) the applicant had granted exclusivity rebates to NEC between October 2002 and November 2005 (initial judgment, paragraphs 900 to 1018), (iv) the applicant and Lenovo had concluded a declaration of intent, the 2007 Memorandum of Understanding (‘the 2007 MoU’) which was subject to an unwritten exclusivity condition (initial judgment, paragraphs 1045 to 1208), and (v) the applicant had made payments to MSH between October 2002 and December 2007 which were conditional on MSH selling exclusively computers equipped with Intel CPUs (initial judgment, paragraphs 1372 to 1502).

99      It should be added that the considerations referred to in paragraph 98 above are subject to two qualifications.

100    First of all, the considerations referred to in paragraph 98 above, according to which the Commission was not required to quantify precisely the part of the rebates constituting the consideration for exclusivity (initial judgment, paragraphs 453, 538, 916 and 1500), apply only in so far as they are called upon in support of the rebates at issue being characterised as ‘exclusivity rebates’.

101    Next, since the General Court is bound by the decision of the Court of Justice on the point of law in the judgment on the appeal which is set out in paragraph 84 above, it accepts the considerations referred to in paragraph 98 above, except for those from which it is apparent, first, that it was not necessary to apply the AEC test in the context of analysing the capacity of the rebates at issue to restrict competition and second, that characterising the rebates at issue as exclusivity rebates was sufficient for them also to be characterised as abusive under Article 102 TFEU.

102    In the light of the foregoing considerations, it must be held, in response to the parties’ arguments, that the subject matter of the dispute concerns, in essence, the General Court’s analysis of the capacity of the rebates at issue to restrict competition in the light, first, of the clarifications provided regarding the principles laid down in the judgment of 13 February 1979, Hoffmann-La Roche v Commission (85/76, EU:C:1979:36), contained in paragraph 133 et seq. of the judgment on the appeal, and, second, the parties’ observations on the conclusions to be drawn from those clarifications.

 The Commission’s arguments concerning the admissibility of certain arguments contained in the main observations of the applicant and ACT

103    In its supplementary observations, the Commission submits that the applicant’s main observations are, to a substantial extent, inadmissible or, at least, irrelevant. According to the Commission, the proper function of written observations pursuant to Article 217 of the Rules of Procedure is to comment on the implications of the judgment on the appeal in the context of the referral back. In view of the fact that it was held in the judgment on the appeal that the failure to address Intel’s arguments concerning the AEC test was an error, but that no findings were made as to whether the AEC test in the contested decision was well founded, there is, in the Commission’s view, nothing in the judgment on the appeal that justifies the applicant devoting a substantial part of its observations to restating the arguments put forward concerning the AEC test.

104    The Commission also contends that the judgment on the appeal does not constitute a new factor justifying the applicant being able to amend or enlarge the complaints made in the proceedings which gave rise to the initial judgment and maintains, on that basis, that certain arguments set out in the main observations of the applicant or ACT are inadmissible.

105    In that regard, it is appropriate to recall the settled case-law that, under Article 76(1)(d) of the Rules of Procedure, an applicant is required to state in the application the subject matter of the proceedings and the form of order sought (see judgment of 20 May 2009, VIP Car Solutions v Parliament, T‑89/07, EU:T:2009:163, paragraph 110 and the case-law cited). Moreover, according to Article 84(1) of the Rules of Procedure, which is applicable by virtue of Article 218 of those rules, where, as in the present case, the General Court is seised of a case by a judgment of the Court of Justice referring it back, no new plea in law may be introduced in the course of proceedings unless it is based on matters of law or of fact which come to light in the course of the procedure. Although Article 84(2) of those rules authorises, in certain circumstances, new pleas in law to be introduced in the course of proceedings, that provision cannot in any circumstances be interpreted as authorising the applicant to bring new claims before the EU Courts and thereby to modify the subject matter of the proceedings or the nature of the action (judgments of 20 May 2009, VIP Car Solutions v Parliament, T‑89/07, EU:T:2009:163, paragraph 110, and of 13 June 2012, Insula v Commission, T‑246/09, not published, EU:T:2012:287, paragraphs 100 and 103).

106    It follows that, following a referral judgment of the Court of Justice, the parties are not entitled, in principle, to rely on pleas which were not raised in the procedure which gave rise to the judgment of the General Court set aside by the Court of Justice (judgment of 1 July 2008, Chronopost and La Poste v UFEX and Others, C‑341/06 P and C‑342/06 P, EU:C:2008:375, paragraph 71). Only a plea which may be regarded as amplifying a plea put forward previously, whether directly or by implication, in the original application, and which is closely connected therewith must be declared admissible (judgment of 11 March 2020, Commission v Gmina Miasto Gdynia and Port Lotniczy Gdynia Kosakowo, C‑56/18 P, EU:C:2020:192, paragraph 66).

107    It should also be recalled that arguments submitted by an intervener are not admissible unless they fall within the framework provided by the forms of order and the pleas in law of the main parties (judgment of 4 February 2020, Uniwersytet Wrocławski and Poland v REA, C‑515/17 P and C‑561/17 P, EU:C:2020:73, paragraph 51).

108    In the present case, the action in Case T‑286/09 was defined on the basis of the application, referred to in paragraph 36 above.

109    In that respect, it should be noted that, in the application, the applicant submitted that, in the contested decision, the Commission had made a series of ‘manifest errors’ when applying the AEC test, and then set out in detail its claims concerning those errors as regards the rebates and payments granted to Dell, Lenovo, HP, NEC and MSH respectively. Therefore, Article 84(1) of the Rules of Procedure does not preclude the applicant from devoting a substantial part of its main observations to restating the arguments in the application concerning the AEC test, or even to amplify those arguments. Such a practice cannot be treated as equivalent to introducing new pleas in law in the course of the proceedings.

110    In those circumstances, it is necessary to reject the Commission’s argument regarding Article 217 of the Rules of Procedure, according to which the applicant’s main observations are, to a substantial extent, inadmissible or, at least, irrelevant.

111    However, the Commission is correct in submitting that, despite the fact that under Article 61 of the Statute of the Court of Justice of the European Union the General Court is bound by the interpretation of the law given in the judgment on the appeal, the fact remains that the General Court is also bound by the pleas raised by the applicant in the application and that the appeal does not, as such, constitute a new factor justifying an amendment to or enlargement of Intel’s complaints against the contested decision. The Commission is also fully entitled to maintain that, following the referral back, ACT cannot put forward arguments in support of the applicant which do not correspond to pleas raised by the applicant in the application.

112    The Commission’s allegations that certain arguments put forward in the applicant’s main observations or in ACT’s main observations have amended or enlarged the complaints made in the proceedings which gave rise to the initial judgment will, if necessary for the purposes of resolving the present case, be considered at a later point in the present judgment (see in particular paragraphs 401 and 506 below).

 Substance

 The heads of claim for annulment of the contested decision

113    The applicant, supported by ACT, submits that the contested decision must be annulled on the ground, first, that it is founded on an incorrect legal assessment, second, that the contested decision failed properly to assess and did not take due account of the criteria referred to in paragraph 139 of the judgment on the appeal and, third, that the contested decision contains an AEC analysis which is vitiated by several errors.

114    The Commission asserts, in essence, that the contested decision must be upheld in its entirety on the grounds that, first, it is consistent with the approach adopted by the judgment on the appeal, second, it took account of all the criteria referred to in paragraph 139 of the judgment on the appeal, and, third, the AEC analysis is not vitiated by any errors.

115    In the present case, as has been stated in paragraph 102 above, the General Court is required to consider, in the light of the judgment on the appeal, the substance of the pleas and arguments put forward by the applicant to show that the rebates at issue were not capable of restricting competition. To that end, it is necessary, first of all, to recall the method defined by the Court of Justice for analysing whether rebates such as those at issue in the present case are capable of restricting competition and, second, to draw the key lessons from them.

I.      The method defined by the Court of Justice for assessing whether a system of rebates has the capacity to restrict competition

116    In the first place, the Court of Justice recalled, in paragraphs 133 to 137 of the judgment on the appeal, the nature and purpose of Article 102 TFEU. Referring in particular to the judgment of 27 March 2012, Post Danmark (C‑209/10, EU:C:2012:172), the Court of Justice, in essence, emphasised that competition on the merits may lead to the exclusion from the market of less efficient competitors (judgment on the appeal, paragraph 134), while recalling the special responsibility on undertakings in a dominant position not to impair genuine, undistorted competition (judgment on the appeal, paragraph 135). Moreover, the Court of Justice pointed out that not all competition by means of price could be regarded as legitimate (judgment on the appeal, paragraph 136).

117    In the second place, in paragraph 137 of the judgment on the appeal, the Court of Justice recalled its settled case-law resulting from the judgment of 13 February 1979, Hoffmann-La Roche v Commission (85/76, EU:C:1979:36) (‘the Hoffmann-La Roche case-law’), according to which, in essence, loyalty rebates constituted an abuse of a dominant position within the meaning of Article 102 TFEU.

118    In the third place, however, the Court of Justice held, in paragraph 138 of the judgment on the appeal, that the Hoffmann-La Roche case-law had to be clarified in the situation where an undertaking in a dominant position ‘submit[ted], during the administrative procedure, on the basis of supporting evidence, that its conduct [had not been] capable of restricting competition and, in particular, of producing the alleged foreclosure effects’.

119    In relation to such a situation, the Court of Justice set out, in paragraph 139 of the judgment on the appeal, the criteria to be taken into account in determining whether there had been an infringement of Article 102 TFEU. According to the Court of Justice, the Commission is required to analyse, first, the extent of the undertaking’s dominant position on the relevant market, second, the share of the market covered by the contested practice, third, the conditions and arrangements for granting the rebates in question, and fourth, their duration and their amount; but it is also required to assess, fifth, the possible existence of a strategy aiming to exclude competitors that are at least as efficient as the dominant undertaking from the market.

120    In the fourth place, in paragraph 141 of the judgment on the appeal, the Court of Justice held that ‘if, [as in the present case], in a decision finding a rebate scheme abusive, the Commission carrie[d] out [an analysis of foreclosure capability], the General Court [had to] examine all of the applicant’s arguments seeking to call into question the validity of the Commission’s findings concerning the foreclosure capability of the rebate concerned’.

121    In the fifth place, in paragraph 142 of the judgment on the appeal, the Court of Justice stated that, although the Commission had stated in the contested decision that ‘the rebates at issue were by their very nature capable of restricting competition such that an analysis of all the circumstances of the case and, in particular, an AEC test were not necessary in order to find an abuse of a dominant position (see, inter alia, paragraphs 925 and 1760 of that decision), it nevertheless carried out an in-depth examination of those circumstances, setting out, in paragraphs 1002 to 1576 of that decision, a very detailed analysis of the AEC test, which led it to conclude, in paragraphs 1574 and 1575 of that decision, that an as-efficient competitor would have had to offer prices which would not have been viable and that, accordingly, the rebate scheme at issue was capable of having foreclosure effects on such a competitor’.

122    It is for that reason that the Court of Justice concluded, in paragraphs 143 and 144 of the judgment on the appeal, that, since the AEC test had played an important role in the Commission’s assessment of whether the rebate scheme at issue was capable of having foreclosure effects on as-efficient competitors, the General Court was required to examine all of Intel’s arguments concerning the AEC test applied by the Commission in the contested decision, something which it had failed to do.

II.    The principles arising from the judgment on the appeal

123    It should be noted, as the parties do, that the judgment on the appeal clarifies the Hoffmann-La Roche case-law, from which three instructive inferences can be drawn.

124    First, it is apparent from paragraphs 137 and 138 of the judgment on the appeal that, although a system of rebates set up by an undertaking in a dominant position on the market may be characterised as a restriction of competition, since, given its nature, it may be assumed to have restrictive effects on competition, the fact remains that what is involved is, in that regard, a mere presumption and not a per se infringement of Article 102 TFEU, which would relieve the Commission in all cases of the obligation to conduct an effects analysis.

125    Second, the Court of Justice has held that, where an undertaking in a dominant position ‘submits, during the administrative procedure, on the basis of supporting evidence, that its conduct was not capable of restricting competition and, in particular, of producing the … foreclosure effects [alleged against it]’, the Commission must analyse the foreclosure capability of the scheme of rebates by applying the five criteria set out in paragraph 139 of the judgment on the appeal (see paragraph 119 above). Having regard to the wording of paragraph 139 of the judgment on the appeal, the Commission is, as a minimum, required to examine those five criteria for the purposes of assessing the foreclosure capability of a system of rebates, such as that at issue in the present case.

126    Lastly and third, it must be pointed out that, although the Court of Justice did not hold that an AEC test necessarily had to be carried out in order to examine the foreclosure capability of all rebate systems, it follows from the judgment on the appeal, in essence, that, where the Commission has carried out an AEC test, that test is one of the factors which must be taken into account by the Commission to assess whether the rebate scheme is capable of restricting competition.

127    It is in the light of those instructive inferences and the method defined by the Court of Justice that the General Court will examine the pleas and arguments raised by the applicant to challenge the capability of the rebates at issue to restrict competition, starting first with an assessment of the substance of the argument put forward by the applicant and ACT that, in essence, the contested decision is founded on an incorrect legal analysis which is capable, in itself, of resulting in it being set aside.

III. The substance of the arguments raised by the applicant and ACT

A.      The argument that the contested decision is founded on an incorrect legal analysis

128    First, as they argued at first instance, the applicant and ACT maintain that the Commission relied on a legal analysis vitiated by a fundamental error which had consequences for the entire contested decision and must, in itself, cause that decision to be set aside.

129    In the view of the applicant and ACT, the contested decision’s finding of an infringement can be upheld only if it can be shown that it is based on a legal analysis consistent with that described in paragraphs 138 and 139 of the judgment on the appeal. However, in their view, it is clear that that is not so in the present case. Rather than interpreting the Hoffmann-La Roche case-law as meaning that a mere presumption of illegality arises, the Commission simply found that the rebates at issue were by their very nature abusive, with the result that it was not necessary to consider and, a fortiori, to take account of their foreclosure capability in coming to the conclusion that they were abusive.

130    Second, ACT adds, in essence, that, even if the contested decision were to contain findings concerning the capacity of the rebates at issue to restrict competition, such additional findings would in no way form part of the legal analysis carried out in order to determine that those rebates were abusive and infringed Article 102 TFEU. In addition, ACT observes that the Commission took the view that factors such as market coverage, the duration and the amount of the rebates were irrelevant for the purpose of establishing the existence of an abuse, confirming that they had not been taken into account for that purpose. In the light of the method defined in the judgment on the appeal, ACT argues that that is sufficient for a finding to be made that the entire analysis of the contested decision is vitiated by an error which justifies its annulment.

131    The Commission disputes the substance of the argument that the contested decision is founded on an incorrect legal analysis which is capable, in itself, of causing it to be set aside.

132    At the 2020 hearing, in response to a question from the General Court, the Commission stated, in essence, that the contested decision was based primarily on a traditional interpretation of the Hoffman-La Roche case-law. That is why it argued in its pleadings at first instance that the contested decision did not need to rely on the AEC test, on the ground that that was irrelevant. It observed, however, that the Court of Justice had stated, in paragraph 143 of the judgment on the appeal, that the AEC test had played an important role in its assessment of whether the rebate scheme at issue was capable of having foreclosure effects on as-efficient competitors, which was consistent with recital 925 of the contested decision. Lastly, the Commission observed, in essence, that, although it had not relied primarily on the AEC test in the contested decision, that test had been carried out as a complement and had made it possible to demonstrate that the rebates at issue were capable of causing anticompetitive foreclosure.

133    In that regard, it should be stated at the outset that it is common ground, in the present case, that the applicant submitted, during the administrative procedure, on the basis of supporting evidence, that its conduct was not capable of restricting competition and, in particular, was not capable of producing the alleged foreclosure effects, which therefore required, as is apparent from paragraphs 138 and 139 of the judgment on the appeal, the Commission to analyse the foreclosure capability of the system of rebates. Furthermore, it is apparent in particular from recitals 920 to 926, 950, 972, 981, 989, 1000 and 1001 of the contested decision and from the explanations provided by the Commission in its pleadings at first instance and at the 2020 hearing that the Commission took the view that, in the light of the principles stemming from the judgment of 13 February 1979, Hoffmann-La Roche v Commission (85/76, EU:C:1979:36), it was not necessary to demonstrate that the rebates at issue were capable of having a foreclosure effect in order to establish an infringement of Article 102 TFEU, since those rebates were by their nature anticompetitive.

134    In the first place, as regards the abovementioned recitals of the contested decision, all of which are found in the part of that decision dealing with the conditionality of the rebates and preceding the AEC analysis, the Commission stated, in particular, in recital 923 of the contested decision that, ‘contrary to what Intel argue[d], however, this [did] not require evidence of actual foreclosure’ and that, ‘in addition, a violation of Article [102 TFEU] [might] also result from the anticompetitive object of practices pursued by a dominant undertaking’.

135    In addition, in recital 925 of the contested decision, the Commission stated as follows:

‘Whilst the findings referred to in the preceding recital, in the absence of any objective justification, are in themselves sufficient to find an infringement under Article [102 TFEU] according to the case[-]law, the Commission will in addition demonstrate in sections 4.2.3 to 4.2.6, that, on top of fulfilling the conditions of the case-law referred to in recitals (920), (921) and (923), the conditional rebates that Intel granted to Dell, HP, NEC and Lenovo and the conditional payments granted to MSH were capable of causing or likely to cause anticompetitive foreclosure (which is likely to result in consumer harm). Although not indispensable for finding an infringement under Article [102 TFEU] according to the case[-]law, one possible way of showing whether Intel’s rebates and payments were capable of causing or likely to cause anticompetitive foreclosure is to conduct an as-efficient-competitor analysis (section 4.2.3). On the basis of the results of this analysis and the qualitative and quantitative evidence (sections 4.2.4 and 4.2.5), and given the lack of objective justification and efficiencies (section 4.2.6), the Commission concludes that Intel’s conditional rebates to Dell, HP, NEC and Lenovo, as well as Intel’s conditional payments to MSH, were an abuse under Article [102 TFEU] which deserves its particular attention.’

136    In recital 926 of the contested decision, which introduces the Commission’s analysis of the nature and functioning of the rebates, it is stated, in particular, that ‘although this is not indispensable according to the case-law quoted in recitals 920, 921 and 923 [of that decision], the Commission will show how the conditional rebate schemes prevented or made it more difficult for each of those OEMs to source x86 CPUs from AMD … [and] will show how the Intel conditional payment schemes to MSH operated as an incentive for MSH to sell exclusively Intel-based PCs, and prevented or made it more difficult for MSH to sell AMD-based PCs’.

137    As regards recitals 950 (concerning Dell), 972 (concerning HP), 981 (concerning NEC), 989 (concerning Lenovo) and 1000 (concerning MSH) of the contested decision, which concludes the analysis of the conditionality of the rebates granted to each OEM or to MSH, the Commission systematically took the view, first of all, that the level of the rebates or payments granted by Intel to those OEMs or to MSH was de facto conditional on their obtaining all of their x86 CPU requirements from Intel; next, that those rebates or payments fulfilled the conditions of the case-law referred to in recitals 920, 921 and 923 of the contested decision for characterisation as abusive and, lastly, that those rebates or payments had the effect of (i) restricting the freedom of the OEMs or of MSH to choose their sources of x86 CPU supply and (ii) preventing other competitors from supplying those OEMs or MSH with x86 CPUs.

138    Lastly, in recital 1001 of the contested decision, which concludes the analysis in Section 4.2.2, entitled ‘Nature and operation of the rebates’, the Commission found as follows:

‘On the basis of the evidence presented in sections 4.2.2.2 to 4.2.2.6 [which concern the nature and operation of the rebates to the OEMs and MSH] and the case[-]law referred to in section 4.2.1 [namely the Hoffmann-La Roche case-law], it is concluded that the level of the rebates granted by Intel to Dell, HP, NEC between the fourth quarter of 2002 and December 2005 was de facto conditional upon those customers sourcing their x86 CPUs exclusively (Dell) or, within defined segments, almost exclusively (HP and NEC) from Intel … The rebates and payments in question constitute fidelity rebates which fulfil the conditions of the relevant case-law for qualification as abusive (see recitals (920), (921) and (923). In addition, they had the effect of restricting the freedom to choose of the respective OEMs and MSH.’

139    In the second place, as regards the explanations provided by the Commission in its written pleadings at first instance and at the 2020 hearing, the Commission stated, first, in paragraph 144 of the defence, as follows:

‘Contrary to the Application …, it is apparent from the structure and the text of the Decision that the Commission does not need to prove the potential effects of Intel’s practices. Recitals 920 to 925 explain very clearly the role of the AEC analysis in the [contested decision], leaving no doubt that the analysis in the preceding recitals establishes, absent an objective justification, the unlawfulness of Intel’s exclusivity rebates and payments because they constitute fidelity rebates within the meaning of the Hoffmann-La Roche line of case-law and they pursued an anticompetitive object/were part of an anticompetitive strategy. For each of these reasons, the [contested decision] concludes ([in] recital 925) that establishing the potential foreclosure effects of Intel’s exclusivity rebates and payments was unnecessary for finding that these practices infringe Article 102 TFEU.’

140    Second, in paragraph 145 of the defence, the Commission stated that ‘the contested Decision ([in] recital 925) also clearly explain[ed] that the Commission established the potential effects of Intel’s rebates only to show that Intel’s practices deserved the Commission’s particular attention’.

141    Third, in paragraph 283 of the defence, the Commission claimed that, ‘contrary to the Applicant’s contentions, the Commission was under no obligation to show that Intel’s exclusivity rebates were capable of foreclosing an as-efficient a competitor’ and that, ‘as stated in recitals 925 and 926 of the [contested decision], the Commission’s findings on the potential effects of Intel’s exclusivity rebates on the market were not part of the legal assessment made to establish their abusive nature but rather a factor that led the Commission to conclude that the infringement merited its particular attention’.

142    Fourth, in paragraph 109 of the rejoinder, the Commission observed that, ‘as pointed out in the [contested decision], the efforts devoted to the AEC analysis [were] not to be taken as an indication that the Commission intended to depart from long-standing case-law on fidelity rebates’.

143    Fifth and last, as regards the coverage rate, duration and amount of discounts, the Commission, first of all, stated in paragraph 68 of the rejoinder that the ‘issue Intel raise[d] about duration [was] legally irrelevant’, since ‘the Decision in Hoffmann-La Roche [had not] identif[ied] duration as a relevant element in the Commission’s assessment of the conduct as abusive’. Next, in paragraph 166 of the defence, it stated that ‘the applicant’s argument [that it failed to take account of the level of the rebates] is misdirected [on the ground that], as noted [in] recital 1620 [of the contested decision], the Decision objects not to the level of the rebates but to the exclusivity for which they [had been] given and to the anticompetitive object that they pursued’. Lastly, in paragraphs 169 and 170 of the defence, the Commission stated that, ‘if Intel wish[ed] to argue that its exclusivity rebates [had] restricted competition “only” with regard to certain types of x86 CPUs, the Commission would [have] expect[ed] to find such contentions in the application’s section on the level of the fines’, since ‘nothing in the case-law on fidelity rebates indicate[d] that their unlawfulness depend[ed] on whether they cover[ed] the entirety of a market or “solely” one segment’.

144    It therefore follows from paragraphs 134 to 143 above that the Commission inferred from the Hoffmann-La Roche case-law, first, that the rebates at issue were by their nature anticompetitive, with the result that there was no need to demonstrate foreclosure capability in order to establish an infringement of Article 102 TFEU. Second, although the contested decision contains an additional analysis of the foreclosure capability of those rebates, the Commission took the view that, in accordance with that case-law, it was not required to take that analysis into account in order to conclude that those rebates were abusive. Third and lastly, still on the basis of that same case-law, the Commission held, inter alia, that a number of factors were irrelevant for determining whether there was abuse.

145    It must be stated that that position is not consistent with the Hoffman-La Roche case-law, as clarified by the Court of Justice in paragraphs 137 to 139 of the judgment on the appeal. It must therefore be found that the applicant and ACT are correct in maintaining that the Commission vitiated the contested decision by an error of law in taking as a starting point the premiss that, in essence, the Hoffman-La Roche case-law allowed it simply to find that the rebates at issue infringed Article 102 TFEU on the ground that they were by their very nature abusive, without necessarily having to take account of the capability of those rebates to restrict competition in order to reach the conclusion that they constituted an abuse.

146    It is true that the Commission did state, in recital 925 of the contested decision, that it had also demonstrated that Intel’s rebates granted to the OEMs and its conditional payments granted to MSH were capable of causing or likely to cause anticompetitive foreclosure, by means of conducting an AEC analysis, set out in Section 4.2.3 of the contested decision, and taking account of the qualitative and quantitative evidence, set out in Sections 4.2.4 and 4.2.5 of that decision. However, it is apparent from the contested decision that the findings in Sections 4.2.3 to 4.2.5 of the contested decision were not regarded as necessary for the legal analysis which was carried out in order to establish that Intel’s practices were abusive.

147    It follows that the Commission, in the contested decision, took the view that the AEC test was not necessary for the purposes of assessing whether Intel’s practices were abusive and for concluding that those practices were abusive.

148    In that regard, it is necessary to reject the argument put forward by the Commission at the 2020 hearing that, in essence, that conclusion is inconsistent with the fact that the Court of Justice stated, in paragraph 143 of the judgment on the appeal, that the AEC test had played an important role in the Commission’s assessment of whether the rebate scheme at issue was capable of having foreclosure effects on as-efficient competitors. Indeed, when read in the light of paragraph 142 of the judgment on the appeal, that paragraph 143 must be interpreted as meaning that the Court of Justice concluded that the AEC test played an important role given the detailed background and the number of recitals devoted to the test in the contested decision. By contrast, contrary to the Commission’s implication, the wording of paragraph 143 of the judgment on the appeal does not support the argument that, in essence, the Court of Justice ruled that the AEC test had formed part of the evidence which the Commission had considered necessary in order to conclude that the rebates were abusive.

149    In the light of all the foregoing considerations, it is necessary to uphold the argument that the contested decision is vitiated by an error of law. However, as is apparent from paragraphs 143 and 144 of the judgment on the appeal, the AEC test played an important role in the Commission’s assessment of whether the rebate scheme at issue was capable of having foreclosure effects on as-efficient competitors and, in those circumstances, the General Court is required to examine all of Intel’s arguments concerning that test.

B.      The argument that the contested decision must be annulled on the ground that it contains an AEC analysis which is vitiated by numerous errors

1.      The scope of the Court’s review

150    The system of judicial review of Commission decisions relating to proceedings under Articles 101 and 102 TFEU consists in a review of the legality of the acts of the institutions for which provision is made in Article 263 TFEU (see judgment of 26 September 2018, Infineon Technologies v Commission, C‑99/17 P, EU:C:2018:773, paragraph 47 and the case-law cited). The scope of that judicial review extends to all the elements of Commission decisions relating to proceedings applying Articles 101 and 102 TFEU which are subject to in-depth review by the EU Courts, in law and in fact, in the light of the pleas raised by an applicant and taking into account all the relevant evidence submitted by the latter (see judgment of 26 September 2018, Infineon Technologies v Commission, C‑99/17 P, EU:C:2018:773, paragraph 48 and the case-law cited). It should, however, be recalled that the EU Courts cannot, in reviewing the legality of acts under Article 263 TFEU, substitute their own reasoning for that of the author of the act in question (see judgment of 24 January 2013, Frucona Košice v Commission, C‑73/11 P, EU:C:2013:32, paragraph 89 and the case-law cited).

151    Before examining the substance of the applicant’s argument, it is appropriate, first, to set out general considerations concerning the AEC test and, second, to recall the rules on the burden of proof and the standard of proof required.

2.      General considerations concerning the AEC analysis

152    The starting point for the AEC test, as defined in recital 1003 et seq. of the contested decision and applied by the Commission in the present case, is that, given, in particular, the nature of its product, its brand image and its profile, Intel was an unavoidable trading partner and that the OEMs would always have purchased at least part of their CPU requirements from Intel, regardless of the quality of products offered by alternative suppliers. Therefore, customers were willing and able to switch only a certain share of their demand to that alternative supplier (‘the contestable share’). Intel’s power to use the non-contestable share as leverage to reduce the price on the contestable share of the market was the result of that status as an unavoidable trading partner.

153    As the Court observed in paragraph 141 of the initial judgment, the AEC test carried out in the contested decision starts from the premiss that an as-efficient competitor, which seeks to obtain the contestable share of the orders hitherto satisfied by a dominant undertaking, must compensate the customer for the exclusivity rebate which it would lose if it purchased a smaller portion than that stipulated by the exclusivity or quasi-exclusivity condition. The AEC test is designed to determine whether the competitor which is as efficient as the undertaking in a dominant position, which faces the same costs as the latter, can still cover its costs in that case.

154    The AEC test, as applied in the present case, establishes the price at which a competitor as efficient as Intel would have had to offer its x86 CPUs in order to compensate an OEM for the loss of any exclusivity payment granted by Intel. In the AEC test, that price is called the ‘effective price’ or ‘EP’.

155    In principle, the portion of the total rebates for which an as-efficient competitor must offer compensation comprises only the amount of the rebates which is subject to the exclusive supply condition, excluding quantity rebates (‘the conditional portion of the rebates’). As is apparent in particular from recital 1460 of the contested decision, in order to take into consideration only the conditional part of a payment, the AEC test refers, in the present case, to the average sales price (‘the ASP’), namely the list price, after deduction of conditional discounts.

156    The smaller the contestable share and, therefore, the smaller the quantity of products for which the alternative supplier may compete, the greater the likelihood that the exclusivity payment will be capable of foreclosing an as-efficient competitor. If the loss of payments granted by Intel to its customer must be spread over a small quantity of products offered by the alternative supplier in the contestable share, that leads to an appreciable reduction in the effective price. It is therefore more probable that the effective price will be below Intel’s measure of viable cost.

157    The effective price must be compared with Intel’s measure of viable cost. Intel’s measure of viable cost adopted in the contested decision is that of the average avoidable cost (‘AAC’).

158    As is apparent in particular from recital 1006 of the contested decision, it may be concluded that a system of exclusivity payments is capable of foreclosing market access for equally efficient competitors if the effective price is below Intel’s AAC. In that situation, the AEC test has a negative result. If, by contrast, the effective price is above the AAC, an as-efficient competitor is deemed to be able to cover its costs and to therefore be in a position to enter the market. In that situation, the AEC test has a positive result.

159    It is in the light of those general considerations that the Court must examine the substance of the applicant’s arguments that the AEC analysis is vitiated by numerous errors.

3.      The burden of proof and the standard of proof required

160    The applicant refers to the case-law of the EU Courts and states, in particular, that competition cases are quasi-criminal in nature, which means that they require a high standard of proof and that the presumption of innocence applies.

161    As was pointed out in paragraph 62 et seq. of the initial judgment, according to Article 2 of Regulation No 1/2003, in any proceedings for the application of Article 102 TFEU, the burden of proving an infringement of that article rests on the party or the authority alleging the infringement, that is to say, in the present case, the Commission. Furthermore, according to well-established case-law, any doubt in the mind of the Court must operate to the advantage of the undertaking to which the decision finding an infringement was addressed. The Court cannot therefore conclude that the Commission has established the infringement at issue to the requisite legal standard if it still entertains any doubts on that point, in particular in proceedings for annulment of a decision imposing a fine (judgments of 8 July 2004, JFE Engineering v Commission, T‑67/00, T‑68/00, T‑71/00 and T‑78/00, EU:T:2004:221, paragraph 177, and of 12 July 2011, Hitachi and Others v Commission, T‑112/07, EU:T:2011:342, paragraph 58).

162    In the latter situation, account must be taken of the principle of the presumption of innocence, which constitutes a general principle of EU law, currently laid down in Article 48(1) of the Charter of Fundamental Rights of the European Union (see judgment of 22 November 2012, E.ON Energie v Commission, C‑89/11 P, EU:C:2012:738, paragraph 72 and the case-law cited). In addition, according to the case‑law of the Court of Justice, the principle of the presumption of innocence applies to the procedures relating to infringements of the competition rules applicable to undertakings that may result in the imposition of fines or periodic penalty payments (see judgment of 22 November 2012, E.ON Energie v Commission, C‑89/11 P, EU:C:2012:738, paragraph 73 and the case-law cited).

163    Although the Commission must produce sufficiently precise and consistent evidence to support the firm conviction that the alleged infringement took place, it is important to emphasise that it is not necessary for every item of evidence produced by the Commission to satisfy those criteria in relation to every aspect of the infringement. It is sufficient if the body of evidence relied on by that institution, viewed as a whole, meets that requirement, as is stated in the case-law concerning the implementation of Article 101 TFEU (see judgment of 26 January 2017, Commission v Keramag Keramische Werke and Others, C‑613/13 P, EU:C:2017:49, paragraph 52 and the case-law cited). That principle also applies in cases concerning the implementation of Article 102 TFEU (judgment of 1 July 2010, AstraZeneca v Commission, T‑321/05, EU:T:2010:266, paragraph 477).

164    With respect to the probative force of the evidence relied on by the Commission, a distinction should be drawn between two situations.

165    First, where the Commission finds that there has been an infringement of the competition rules based on the supposition that the facts established cannot be explained other than by the existence of anticompetitive behaviour, the EU Courts will find it necessary to annul the decision in question where those undertakings put forward arguments which cast the facts established by the Commission in a different light and which thus allow another plausible explanation of the facts to be substituted for the one adopted by the Commission in concluding that an infringement occurred. In such a case, it cannot be considered that the Commission has adduced proof of an infringement of competition law (see, to that effect, judgments of 28 March 1984, Compagnie royale asturienne des mines and Rheinzink v Commission, 29/83 and 30/83, EU:C:1984:130, paragraph 16, and of 31 March 1993, Ahlström Osakeyhtiö and Others v Commission, C‑89/85, C‑104/85, C‑114/85, C‑116/85, C‑117/85 and C‑125/85 to C‑129/85, EU:C:1993:120, paragraphs 126 and 127).

166    Second, when the Commission relies on evidence which is, in principle, sufficient to demonstrate the existence of the infringement, it is not sufficient for the undertaking concerned to raise the possibility that a circumstance arose which might affect the probative value of that evidence in order for the Commission to bear the burden of proving that that circumstance was not capable of affecting the probative value of the evidence. On the contrary, except in cases where such proof could not be provided by the undertaking concerned on account of the conduct of the Commission itself, it is for the undertaking concerned to prove to the requisite legal standard, on the one hand, the existence of the circumstance on which it relies and, on the other, that that circumstance calls into question the probative value of the evidence relied on by the Commission (see, to that effect, judgment of 15 December 2010, E.ON Energie v Commission, T‑141/08, EU:T:2010:516, paragraph 56 and the case-law cited).

4.      The substance of the arguments that the contested decision is vitiated by numerous errors in the AEC test

167    The applicant claims, in essence, that the AEC analysis carried out for all the OEMs and for MSH contains numerous errors, in particular as regards the contestable share, the conditional portion of the rebates and the AAC. It presents general arguments in respect of which it subsequently provides more detail for each OEM and for MSH, regarding each of the three aspects mentioned above.

(a)    General arguments concerning the alleged errors in the AEC test applied to Dell

168    As regards the AEC test carried out in relation to the rebates granted to Dell, Intel submits, in essence, that the Commission makes errors when it assesses each of the three key elements of the AEC test, namely the contestable share, the conditional portion of the rebates and the costs. According to Intel, in most cases, correcting only one of those errors is sufficient to demonstrate that it satisfies the AEC test, even if the other errors are left uncorrected. Intel submits that the Commission selected data from inconsistent sources in order to slant the results against Intel and that the Commission made selective and inconsistent use of the documents. That is made clear, in particular, if the results of the Commission’s AEC test analysis are compared with the facts as they actually occurred when Dell began sourcing from AMD in 2006.

169    In addition, Intel asserts that the Commission acknowledges, in the contested decision, that the rebates passed the AEC test during the first four quarters concerned (between December 2002 and October 2003). Intel argues that, despite that, in recital 1281 of the contested decision, the Commission inexplicably concluded that the rebates granted by Intel ‘from December 2002 to December 2005’ were ‘capable of having or likely to have anticompetitive foreclosure effects’. According to the applicant, the contested decision does not even attempt to explain or to justify that inconsistency in its reasoning.

170    The Commission submits, in essence, that the contested decision shows that the exclusivity rebates granted to Dell were capable of foreclosing an as-efficient competitor. In particular, it did not state that the rebates which Intel granted to Dell satisfied the AEC analysis between December 2002 and October 2003. According to the Commission, Intel’s calculations were based solely on optimistic assumptions that were favourable to it. The Commission contends that Intel did not provide contemporaneous documents substantiating its claims concerning the contestable share. As regards the events which took place after Dell announced that it would start, from May 2006, to obtain part of its supply from AMD, the Commission argues that those events confirm the findings that the rebates granted by Intel to Dell were capable of having a foreclosure effect on a competitor as efficient as Intel. The Commission also asserts that the purpose of the AEC analysis is not to give predictions of actual market development, but to identify the economic incentives provided by rebate schemes in a theoretical situation.

(1)    The assessment of the contestable share

171    The contested decision used a contestable share of 7.1% for the AEC analysis of the rebates which Intel granted to Dell. According to the Commission, that figure comes from a spreadsheet dating from January 2004 (‘the 2004 spreadsheet’) which Dell had provided to it during the administrative procedure. The Commission stated in recitals 1202 to 1208 of the contested decision that the 2004 spreadsheet included, inter alia, a specific analysis of the time profile of a shift in supplies to AMD, whereas previous presentations, including one dated 26 February 2003 entitled ‘AMD Update – Dimension LOB’ and another dated 17 March 2003 entitled ‘AMD Update’, did not contain such a specific analysis and, therefore, were disregarded by the Commission.

172    In recitals 1209 to 1212 of the contested decision, the Commission stated that the 2004 spreadsheet was an internal Dell document speculating how the relationship between Dell and AMD could evolve, with AMD’s penetration increasing in the various business segments for which an assessment was made, that spreadsheet having to be read together with the covering letter of 18 April 2007 sent by Dell to the Commission, to which footnote 1542 to recital 1209 of the contested decision refers.

173    The contested decision highlighted, in recitals 1210 to 1213, that, at the time when the 2004 spreadsheet was drawn up, a switch of x86 CPU supplier was envisaged by Dell for certain segments of manufactured products. According to the contested decision, in view of the estimate of the total volume in each of those segments, it was possible to calculate that AMD’s total share in the four years in question, namely the tax years 2005 to 2008, would be 7.1% in the first year and 17.3%, 22.5% and 24.2% over the following three years. The Commission concludes that it was therefore appropriate to use a contestable share of 7.1% for the purposes of the AEC test analysis.

174    In recitals 1214 to 1254 of the contested decision, the Commission rejected a number of arguments raised by Intel concerning the contestable share, which covered, first, the date which should be used as the starting point for the calculations in the 2004 spreadsheet, second, Dell’s internal presentation entitled ‘MAID status review’ of 17 February 2004 (‘Dell’s 17 February 2004 presentation’), third, Intel’s internal estimates, fourth, Dell’s actual switching of part of its supply needs to AMD in 2006 and, fifth and lastly, the statements by Dell executives in the context of the private litigation between AMD and Intel in the State of Delaware.

175    In recitals 1255 to 1259 of the contested decision, the Commission compared the required share with the contestable share. In essence, the Commission took the 7.1% indicated in recital 1213 of the contested decision as the relevant percentage to be applied for the contestable share and compared it with the required share shown in Table 22 in recital 1194 of the contested decision (‘Table 22’). It accordingly took the view that for 9 out of 13 quarters, the required share was higher than the contestable share. It observed that that conclusion was not affected by the use of Intel’s assessment of the ratio between the AAC and the ASP, even if the AAC was underestimated.

176    The Commission then recalled, in recital 1257 of the contested decision, that the contestable share figure of 7.1% had been established using Dell’s internal estimates made in January 2004, that is to say, in preparation for a switch of supplier which could have taken place at the earliest during the first quarter of Dell’s 2005 tax year, whilst the corresponding required share was 7.9%. The Commission also stated, in recital 1258 of the contested decision, the reasons why, before the first quarter of the 2005 tax year, it was possible that the contestable share might have been lower than 7.1%. The Commission accordingly took the view that the difference between the required share and the contestable share during the first quarters of the relevant period could be lower than the figures in Table 22 would suggest at first sight.

177    In recitals 1260 to 1265 of the contested decision, the Commission refers to a number of reinforcing factors which, if included in the analysis, would, according to the contested decision, reinforce the estimated foreclosure capability of the rebates. According to the contested decision, those factors are, in essence, (i) that Dell clearly considered that any loss of Intel’s rebates would also be accompanied by increased rebates from Intel to Dell’s OEM competitors and (ii) that the estimate of the contestable share does not take account of the fact that Dell also purchased from Intel products other than x86 CPUs, in particular chipsets.

178    Lastly, in recitals 1266 to 1280 of the contested decision, the Commission used an alternative method of calculating the contestable share.

179    The applicant’s claims concern, first, the use of the 2004 spreadsheet and the Commission’s assessment of its content and, second, certain other items of evidence upon which, in the applicant’s view, the assessment of the contestable share ought to have been based.

180    In the first place, according to the applicant, the Commission could not found its assessment of the contestable share on a document which was not ‘knowable’ to Intel. Such an approach constitutes an infringement of the principle of legal certainty. In addition, the applicant argues that the assessment of the contestable share on the basis of the 2004 spreadsheet is incorrect, because that figure of 7.1% is based solely on eight months of sales of AMD’s x86 CPUs and because the selective and inconsistent analysis of the 2004 spreadsheet deprives the Commission’s conclusions of any credibility. Lastly, the applicant asserts that the contested decision accepts that Intel’s rebates to Dell passed the AEC test for the first four quarters mentioned, that is to say, between December 2002 and October 2003.

181    The applicant refers, in respect of the period taken into account in the 2004 spreadsheet, also to paragraphs 82 to 86 and 121 to 131 of Professor Shapiro’s report of 4 January 2008, submitting that, if Dell had feared retaliation by Intel following the commencement of sourcing from AMD, Dell would have avoided disclosing its decision to switch to a competitor and would have kept secret that decision until the last moment, after having concluded an agreement with AMD on the rebate conditions and percentage for the following quarter.

182    According to the applicant, paragraphs 82 to 86 of Professor Shapiro’s report of 4 January 2008 emphasise the importance of the date on which Dell took the decision to purchase the x86 CPUs from AMD and that date is compared with the date on which the first supplies of AMD x86 CPUs to Dell could in fact take place. Professor Shapiro relies on Dell’s 17 February 2004 presentation in stating that those two dates could have been three or four months apart (in particular, February 2004 for the first date and June 2004 for the second). According to Professor Shapiro, taking account also of the actual date when supply of AMD’s x86 CPUs commenced (and therefore the fact that the 2004 spreadsheet corresponded, according to him, to only eight months of the first year mentioned), Dell’s contestable share was rather 10.65%.

183    In the second place, the applicant submits, in essence, that the assessment of the contestable share in the contested decision is incorrect because the Commission incorrectly rejected (i) evidence from Dell’s senior executives demonstrating that the contestable share was far higher than that used by the Commission, in the range of 12.5% to 17.5%, (ii) evidence demonstrating that Intel believed Dell’s contestable share to be in the range of 15 to 25%, and (iii) evidence relating to Dell switching supplies to AMD in 2006.

184    The Commission’s answer is, in essence, in the first place, that the 2004 spreadsheet is more reliable for assessing the contestable share than the documents submitted by Intel, since it is a contemporaneous document of Dell containing an extremely thorough and detailed quantitative analysis of the potential switch of supplies of x86 CPUs from Intel to AMD.

185    In the defence, the Commission argues that the analysis of the documents dating from the period between May and July 2006, which therefore post-dates the announcement of Dell’s partial switch of supplies to AMD, although having only limited weight in relation to the AEC test conducted in the contested decision, confirms that Intel was in a position to reduce the rebates granted to Dell immediately after the announcement of the partial switch of its supplies to AMD, that is to say, four months before Dell started selling products with AMD-based x86 CPUs. The Commission further states that, although it is correct to assert that the 2004 spreadsheet related, for the first year, only to Dell’s sales plans for AMD-based x86 CPU products starting after the first four months of 2004 had elapsed, Dell expected, however, a loss of 50% of the rebates for the whole of 2004, including the four months preceding the first sale.

186    The Commission also states, in paragraph 46 of its main observations, as follows:

‘The [contested] decision … finds the relevant period for the AEC analysis begins at the latest when Intel would have been in a position to start withdrawing rebates from its customer. The reason for this is straightforward: when Intel’s customers analysed the pros and cons of switching to AMD, they had to consider the entire period during which that decision would have financial consequences.’

187    In the second place, according to the Commission, Intel did not provide any document dating from the material time to support its assertion that it took the view that the contestable share was between 15% and 25%. The only document that Intel provided is an ad hoc document drawn up by one of its executives for the purposes of the administrative procedure, containing information which is contradicted, at least in part, by a document dating from the material time, written by the same Intel executive. The Commission submits that, for that reason, the contested decision does not define a position on whether the assessment of the contestable share was to be based on the dominant undertaking’s expectations.

188    It is necessary to examine, first, the applicant’s allegations concerning the principle of legal certainty and then its allegations relating to the 2004 spreadsheet, on the basis of which, in essence, the contestable share disputed by the applicant was calculated.

(i)    The arguments concerning the principle of legal certainty

189    The applicant relies on the principle of legal certainty to criticise the Commission for having established Dell’s contestable share as being 7.1% by relying on the 2004 spreadsheet, which was sent to the Commission as an annex to Dell’s covering letter of 18 April 2007, although that is an internal Dell document which contained confidential information of which it had not been aware during the relevant period, namely from December 2002 to December 2005.

190    In that regard, it should be noted that, in the judgment of 14 October 2010, Deutsche Telekom v Commission (C‑280/08 P, EU:C:2010:603, paragraphs 198 to 202), the Court of Justice held that, in order to assess whether the pricing practices of a dominant undertaking were likely to eliminate a competitor contrary to Article 102 TFEU, it is necessary to adopt a test based on the costs and strategy of the dominant undertaking itself. Since the abusive nature of the pricing practices at issue in that case stems from their exclusionary effect on the dominant undertaking’s competitors, the General Court did not therefore err in law when it held that the Commission had been able to analyse the abusive nature of the dominant undertaking’s pricing practices solely on the basis of the dominant undertaking’s charges and costs. Since such a test can establish whether the dominant undertaking itself would have been able to offer its retail services to end-users otherwise than at a loss if it had first been obliged to pay its own wholesale prices for local loop access services, it was suitable for determining whether the applicant’s pricing practices had an exclusionary effect on competitors by squeezing their margins. The Court of Justice held that such an approach was particularly justified because, as the General Court had indicated, in essence, in paragraph 192 of the judgment of 10 April 2008, Deutsche Telekom v Commission (T‑271/03, EU:T:2008:101), it was also consistent with the general principle of legal certainty in so far as the account taken of the costs of the dominant undertaking allowed that undertaking, in the light of its special responsibility under Article 102 TFEU, to assess the lawfulness of its own conduct, since, while a dominant undertaking knew what its own costs and charges were, it did not, as a general rule, know what its competitors’ costs and charges were.

191    That case-law was clarified in the judgment of 17 February 2011, TeliaSonera Sverige (C‑52/09, EU:C:2011:83, paragraphs 41 to 46). In paragraphs 45 and 46 of that judgment, the Court of Justice held that it could not be ruled out that the costs and prices of competitors may be relevant to the examination of the pricing practice at issue. According to the Court of Justice, that might in particular be the case where the cost structure of the dominant undertaking is not precisely identifiable for objective reasons, or where the service supplied to competitors consists in the mere use of an infrastructure, the production cost of which has already been written off, so that access to such an infrastructure no longer represents a cost for the dominant undertaking which is economically comparable to the cost which its competitors have to incur to have access to it, or again where the particular market conditions of competition dictate it, by reason, for example, of the fact that the level of the dominant undertaking’s costs is specifically attributable to the intensity of the competition to which it is subject. It must therefore be concluded that, when assessing whether a pricing practice which causes a margin squeeze is abusive, account should as a general rule be taken primarily of the prices and costs of the dominant undertaking concerned on the retail services market. Only where it is not possible, in the light of the particular circumstances, to refer to those prices and costs should those of its competitors on the same market be examined (see, to that effect, judgment of 13 December 2018, Deutsche Telekom v Commission, T‑827/14, EU:T:2018:930, paragraph 165).

192    Even if that case-law, developed in cases relating to predatory pricing or margin squeeze practices, could be transposed to the present case, for the purposes of determining the contestable share for the AEC test applied to pricing rebates, Intel’s arguments cannot succeed.

193    It is apparent from the case-law cited in paragraph 191 above that there is an exception from the principle that reliance should be placed primarily on data known to the dominant undertaking for the purposes of examining whether conduct is abusive where it is not possible, in the light of the circumstances, to rely on such data and that it is therefore necessary to rely on data known to other economic operators.

194    In the present case, Intel states that, during the relevant period, Dell’s contestable share was, in its view, between 15% and 25% ‘and that Dell’s contemporaneous documents [agreed] with this estimate, confirmed by’ the declaration by I1, the person at Intel responsible for the relationship with Dell at the material time, dated 21 December 2007 (‘the I1 declaration of 21 December 2007’).

195    In that regard, it should be noted that the I1 declaration of 21 December 2007 was made by a representative of the applicant and seeks to mitigate the applicant’s responsibility for the established infringement. The evidential value of that statement therefore is poor, and, at the very least, lower than that of the documentary evidence provided in the administrative procedure or before the General Court (see, to that effect, judgment of 8 July 2008, Lafarge v Commission, T‑54/03, not published, EU:T:2008:255, paragraph 379).

196    As to the ‘contemporaneous Dell documents’ to which the applicant refers, this consists of (i) an internal Dell document, namely an email from D1 dated 10 November 2005 (‘the D1 email of 10 November 2005’), which is presented as a convincing document produced on 18 February 2009 and of which Intel does not claim to have been aware during the relevant period, and (ii) statements by D3 made on 11 February 2009 in the context of the private litigation between Intel and AMD in the State of Delaware, which therefore post-date the relevant period.

197    It follows from the foregoing that, in support of its allegations that it was aware of certain estimates of Dell’s contestable share, to which it could have referred in order to assess the lawfulness of its practices during the relevant period, the only relevant item on which the applicant relies is a statement by one of its executives which seeks to mitigate the applicant’s responsibility for the established infringement.

198    As the Commission correctly observes, the applicant has not produced before the Court any documentary evidence relating to an estimate of Dell’s contestable share of which it was aware during the relevant period. In order to substantiate the content of the statement referred to in paragraph 197 above, the applicant relies on internal Dell documents or on statements made by a Dell executive even though it has not been established that the applicant was aware of them during the relevant period.

199    It follows that, if it were necessary, in the present case, as Intel maintains, to apply the principle of legal certainty, in order to determine Dell’s contestable share, the Commission would have been obliged to rely solely on a statement made by a representative of the applicant and which sought to mitigate the applicant’s responsibility for the established infringement, without being able to rely on internal Dell documents, some of which are, moreover, prima facie relevant in the applicant’s view, since the applicant relies on them in order to substantiate the merits of that statement.

200    Consequently, unless it is accepted that, for that undertaking to escape all liability, it is sufficient for a representative of the dominant undertaking to make certain exculpatory statements for the purposes of the administrative procedure, it must be held that, in the circumstances of the present case, the Commission was not required to rely solely on evidence concerning the data known to Intel during the relevant period and that it could take account of other evidence concerning data known to other economic operators, in the present instance, internal Dell documents.

201    Accordingly, it is necessary to reject the applicant’s arguments relating to the principle of legal certainty, criticising the Commission for having relied on the 2004 spreadsheet of which the applicant was unaware during the relevant period, rather than on the applicant’s own estimates of the contestable share during the relevant period.

(ii) The assessment of the contestable share as 7.1%

202    The applicant submits that the Commission erred in relying solely on the 2004 spreadsheet to assess Dell’s contestable share as 7.1% and in unjustifiably rejecting other documents or evidence with a higher evidential value from which a higher contestable share can be inferred.

203    In the first place, the applicant relies on a number of items of evidence to dispute the contestable share of 7.1% used by the Commission.

204    First, Intel refers to the D1 email of 10 November 2005, in which D1 indicated to D3, [confidential], and to D4, at the time [confidential], that ‘the assumptions in [the MAID project] were over the first six months to twelve months that we could move … approx[imately] 25% of our total volume’ to AMD. The MAID project was one of the concrete programmes in which Dell planned to switch part of its requirements to AMD. Intel asserts that, on the basis of the calculations made in the report of Professor Salop and Dr Hayes of 22 July 2009 (‘the Salop-Hayes report’), the volume projection of 25% of Dell’s requirements translated to a contestable share of 17.5% for the first year (or 12.5% using the Commission’s approach, which the applicant considers to be unreasonable).

205    Second, Intel relies on an internal Dell email, from D5 to D1 of 9 March 2004 (‘the D5 email of 9 March 2004’) which involved a different assumption, namely a switch of Dell supplies to AMD for 25% of the total volume of its x86 CPU requirements ‘inside 90 days’.

206    Third, the applicant relies on I1’s declaration of 21 December 2007 to assert that, during the period which was relevant to the allocation of rebates to Dell, its internal estimate of Dell’s contestable share of x86 CPUs was between 15% and 25%. In that statement, I1 wrote that, during that period, he ‘believed that if Dell were to add AMD as a second source, it would likely source 15-25% of its microprocessors from AMD in the first year and a quarter to a third of its microprocessors by the third year of a ramp’.

207    The Commission maintains, first, as regards the D1 email of 10 November 2005, that it is less reliable for assessing the contestable share than the 2004 spreadsheet, since it is a schematic summary of D1’s recollections of the MAID programme written two years after the events. By contrast, the 2004 spreadsheet assesses the potential share of purchases which Dell might switch to AMD, by product line and by segment, in the context of the MAID project that Dell was undertaking at that time. In addition, the Commission maintains, in paragraphs 287 to 290 of the defence, also referring to Annex B.31 to the defence, that it had demonstrated that Intel’s claim that the D1 email of 10 November 2005 disproves the 7.1% contestable share estimate used in the contested decision was based on hypothetical calculations which rely on speculative ‘ramp-up’ scenarios from AMD biased in Intel’s favour. Despite that email mentioning a ramp-up period for x86 CPUs from sources other than Intel, in the present case AMD, of between 6 and 12 months, Intel made no provisional calculations for the latter scenario, that is to say, a slow ramp-up of 12 months. In addition, according to the Commission, in Intel’s calculations, the starting level for the ramp-up was 5% instead of 0%, there being no logic to support such a sudden discontinuous increment. In paragraph 198 of the rejoinder, the Commission submits, referring to Annex D.9, that the claims put forward in the reply – that Intel’s calculations were not biased – are unfounded and rely on serious distortions of actual data.

208    Thus, according to the Commission, the scenarios in the D1 email of 10 November 2005 which were less favourable to Intel were systematically ignored. By also incorporating assumptions which are not biased towards particular scenarios, it appears, in the view of the Commission, that the contestable share resulting from the data contained in D1’s email was between 5.6% and 10.4% That value is consistent with the figure of 7.1% in the contested decision, which is based on more accurate data.

209    Second, the Commission argues that, even though Intel states that it believed that Dell would source 15 to 25% of its x86 CPU requirements from AMD in the first year, as explained in detail in recitals 1231 to 1238 of the contested decision, Intel did not provide any contemporaneous documents in support of that contention. According to the Commission, Intel is relying in that regard on a mere ad hoc document written by one of its executives, I1, for the purposes of the administrative procedure, containing information which is contradicted, at least in one instance, by a contemporaneous document written by that executive. According to the Commission, it cannot therefore be accepted as credible evidence of Intel’s internal estimates of the contestable share.

210    In recitals 1251 and 1252 of the contested decision, the Commission found, as regards the D1 email of 10 November 2005, in essence, that the figure was an aspiration rather than an actual reasonable estimate. In addition, it was not possible to determine exactly the starting point for the ramp-up of the goods in question. The Commission recalls that the relevant starting point for the one-year period assessed in the as-efficient-competitor analysis is the date when Intel could start to react to Dell’s shift of supplier. That date, according to the contested decision, predated the actual date of Dell’s first sales of computers with AMD-based x86 CPUs.

211    In recitals 1233 to 1236 of the contested decision, the Commission asserts that the credibility of the I1 declaration of 21 December 2007, which was prepared solely for the administrative procedure, is weakened by the fact, first, that Intel was unable to corroborate it by evidence dating from the material time and, second, that it contains, as regards another point concerning Intel’s reaction if Dell were to terminate its exclusive supply to it, information which contradicts a presentation made by I1 on 10 January 2003 entitled ‘Dell F1H’04 MCP’.

212    In recital 1237 of the contested decision, the Commission further states that Intel had itself drawn the Commission’s attention to the fact that ‘[D1] [had] testified that Dell had no viable AMD option in early 2003’. The Commission goes on to state that ‘Intel would therefore have the Commission simultaneously conclude that Dell had no viable AMD option in early 2003, and that Dell would be likely to source 15-25% from AMD in the first year on the basis of two non-contemporaneous statements by [I1] and [D1]’.

213    It should be noted at the outset that, contrary to Intel’s assertions, the documents on which Intel relies do not, in themselves, have a greater evidential value than the 2004 spreadsheet.

214    First of all, just as is the case for the D1 email of 10 November 2005 and the D5 email of 9 March 2004, the 2004 spreadsheet is an internal Dell document which was drawn up during the relevant period and relates to the x86 CPU requirements which that OEM intended to switch to AMD.

215    Next, Intel submits that (i) the documents on which it relies were drafted by senior Dell executives, (ii) D1 confirmed, under oath, in the course of the private litigation between Intel and AMD in the State of Delaware, the content of his email of 10 November 2005, and (iii) in the course of that litigation, D3 stated that he had no reason to question the accuracy of D1’s remarks.

216    However, it is apparent from the case-law that answers given on behalf of an undertaking as such carry more weight than the answer given by an employee or by one of its directors, whatever that person’s individual experience or opinion (see judgment of 8 July 2004, JFE Engineering v Commission, T‑67/00, T‑68/00, T‑71/00 and T‑78/00, EU:T:2004:221, paragraph 205 and the case-law cited).

217    Thus, the Commission was fully entitled to take the view that the 2004 spreadsheet has a greater evidential value than the documents or statements of Dell’s senior executives relied on by Intel.

218    The Commission was also correct in relying on the precise and detailed nature of the information contained in the 2004 spreadsheet, since those characteristics are, in principle, likely to strengthen the evidential value of a document (see, to that effect, judgment of 20 April 1999, Limburgse Vinyl Maatschappij and Others v Commission, T‑305/94 to T‑307/94, T‑313/94 to T‑316/94, T‑318/94, T‑325/94, T‑328/94, T‑329/94 and T‑335/94, EU:T:1999:80, paragraph 593).

219    The fact remains, however, that the evidence relied upon by Intel is not completing lacking in evidential value.

220    It is necessary to assess the evidence consisting of (i) assessments of the contestable share by Dell, namely assessments made by D1 and D5, (ii) statements made by Dell executives in the context of the private litigation between Intel and AMD in the State of Delaware, and (iii) a document which is the I1 declaration of 21 December 2007.

221    First, it is apparent from the D1 email of 10 November 2005 that ‘the assumptions in [the MAID project] were over the first six months to twelve months that we could move … approx[imately] 25% of … total volume’ of Dell’s supplies to AMD. As regards the Commission’s criticisms concerning the objective reliability of the D1 email of 10 November 2005, it is clear, first of all, that its sender, D1, was [confidential] at the material time. Next, that email was written during the relevant period. Finally, its content is sufficiently clear and relates precisely to the contestable share at the material time. In the light of those factors, it is necessary to take that email into account and to attach genuine relevance to it, since its reliability is not weakened by the fact that it is a schematic summary of D1’s recollections.

222    Second, contrary to what the Commission states in recital 1251 of the contested decision, the assertion that ‘the figure of 25% achieved following a 6 to 12 month period [mentioned in the D1 email of 10 November 2005]’ was akin to ‘an “aspiration” more than an actual reasonable estimate’ in fact concerned, as the applicant correctly submits, a different item of evidence, namely the D5 email of 9 March 2004, and related to a different assumption, namely Dell switching to supply from AMD for 25% of the total volume of its CPU x86 requirements ‘inside 90 days’. It is only in the latter email that the terms ‘aspiration’ or ‘planning guideline’ are used.

223    Moreover, as regards the D5 email of 9 March 2004, it must also be observed that, even taking into consideration the fact that a particularly rapid switch to supply by AMD for 25% of the total volume of Dell’s x86 CPU requirements ‘inside 90 days’ was mentioned in that email only as an aspiration, that already demonstrates that such a possibility could be raised in an internal discussion in Dell, at least in terms of motivation or targeted planning, which must be regarded as additional support for the possibility that the contestable share might have been quite high. That is all the more so since that email post-dates the 2004 spreadsheet by only a few months and since, in the same way as the D1 email of 10 November 2005, that email refers to switching approximately 25% of Dell’s requirements to AMD.

224    Third, during the private litigation between Intel and AMD in the State of Delaware, D1 confirmed that he assumed in the context of the MAID project that, during the first 6 to 12 months, the switch of the requirements to AMD would involve approximately 25% of the volume of x86 CPUs and D3 stated that he had no reason to question the accuracy of D1’s remarks.

225    Thus, the statements made by Dell’s executives in the context of the private litigation between Intel and AMD in the State of Delaware support the assumption that, in the context of the MAID project, during the first 6 to 12 months, the switch of Dell’s requirements to AMD could involve approximately 25% of the volume of x86 CPUs.

226    Fourth, the I1 declaration of 21 December 2007 remains to be assessed. As regards the Commission’s criticisms of that declaration in the contested decision, those fall into three categories, the first of which is that that document was drawn up solely for the administrative procedure; the second, that it was not supported by other evidence dating from the material time; and, the third, that it contains certain contradictions compared with a presentation made by I1 on 10 January 2003 to Dell (see paragraph 211 above).

227    In that regard, first of all, it is true that, as is apparent from paragraph 195 above, the I1 declaration of 21 December 2007 was made by a representative of the applicant and seeks to mitigate the applicant’s responsibility for the established infringement, with the result that it, in itself, is of low evidential value.

228    The fact remains that the I1 declaration of 21 December 2007 was made under oath and that I1 was, as is apparent from paragraph 1 of that declaration, [confidential], since 1999. Due to his position and seniority at Intel, I1 must have been fully aware of the main features of the relationship with Dell, which includes the question of the likely contestable share for the relevant period.

229    Next, as is apparent from paragraphs 221 to 223 above, internal Dell documents from the relevant period support the I1 declaration of 21 December 2007 in so far as the switch of Dell’s requirements to AMD could have involved up to 25% of the volume of x86 CPUs. At the very least, that declaration, in so far as it refers to a volume of x86 CPUs between 15 and 25%, shows, as do those documents, that the switch of Dell’s requirements to AMD could exceed the 7% volume found in the 2004 spreadsheet.

230    As regards the alleged contradictions concerning the economic logic of switching to AMD raised by the Commission, or as regards contradictions in the declarations of I1, it should be noted that I1 provides details, in his presentation of 10 January 2003, concerning the nature of the relationship between Intel and Dell, highlighting, in particular, that it was necessary to ensure that Dell understood the particular nature of that relationship, should Dell intend to switch to AMD. As the Commission correctly points out in recitals 1235 and 1236 of the contested decision, that part of that presentation may appear to contradict Section 4 of the I1 declaration of 21 December 2007 concerning the absence of any conditionality of the rebates offered by Intel. However, in contrast to the inferences drawn by the Commission, it must be stated that, in so far as that contradiction relates to a part of the I1 declaration of 21 December 2007 which differs from the part which is relevant for assessing the contestable share, it cannot be used to infer that that declaration has no evidential value overall and therefore also in so far as it concerns the contestable share.

231    It should be added that the I1 declaration of 21 December 2007, indicating that any potential supply by AMD of x86 CPUs to Dell would be of a significant size, given the costs, increased complexity and additional engineering, assistance and sales resources linked to the addition of AMD platforms, is neither illogical nor contradictory. It is apparent from the I1 declaration of 21 December 2007 that I1 went to some lengths to provide an objective overview, since he also states that, in his view, the likelihood of a partial switch of Dell’s supply to AMD was only ‘low’ during the relevant period. However, I1 explains clearly in his declaration that, for the reasons mentioned above, if Dell were to add AMD as the second source of supply of x86 CPUs, that would necessarily have involved 15 to 25% of its requirements.

232    It cannot be ruled out that Dell might actually have intended, during the relevant period, to obtain part of its supplies of x86 CPUs from AMD. It is apparent from a number of documents in the file, including the 2004 spreadsheet, that Dell considered, and regularly analysed at an internal level, throughout the relevant period, the possibility of a partial transition to AMD. It should also be noted that D1’s witness statement, referred to in paragraph 212 above, according to which AMD was not a viable option for Dell, concerned only the year 2003. The Commission itself stated, in particular in recital 1258 of the contested decision, that it could not be ruled out that Dell’s contestable share would vary over time, and could, inter alia, increase in due time, owing to the fact that consumers would increasingly become accustomed to x86 CPUs produced by AMD. Consequently, it cannot be considered that the situation regarding Dell’s contestable share in 2003 necessarily had to be the same as the situation in 2004 and 2005. Accordingly, the I1 declaration of 21 December 2007, which is supported by the evidence referred to in paragraphs 221 and 222 above, must also be regarded as reliable in so far as it concerns Dell’s contestable share.

233    Consequently, from the D1 email of 10 November 2005, the D5 email of 9 March 2004, the statements made by Dell’s executives in the context of the private litigation between Intel and AMD in the State of Delaware and the I1 declaration of 21 December 2007, which, taken together, support one another, it is apparent that, during 2005, the switch of Dell’s requirements to AMD could have involved up to 25% of the volume of x86 CPUs, and not 7% as is set out in the 2004 spreadsheet.

234    It follows that the evidence relied on by Intel casts doubt on the argument that Dell’s contestable share had to be assessed solely on the basis of the 2004 spreadsheet which mentioned a 7% volume switch of Dell’s requirements to AMD for 2005, from which the Commission inferred a contestable share of 7.1%.

235    The conclusion reached by the Court cannot be invalidated by the economic analyses submitted by the Commission to the Court in Annex B.31, which illustrates its arguments raised in paragraph 290 of the defence, and in paragraphs 196 and 199 of the rejoinder, which refer to Annex D.9, which aim to show that, even if it were necessary to calculate the contestable share on the basis of the documents referred to in paragraph 233 above, a contestable share of between 12.5% and 17.5% could not be inferred from it, as Intel maintains.

236    The Court cannot take account of those additional analyses, which were submitted for the first time during the procedure before it, in order to substantiate the AEC test contained in the contested decision without substituting its own reasoning for that of the Commission found in that decision. The case-law cited in paragraph 150 above prohibits the Court from making such a substitution.

237    Moreover, it should be added that even the economic analyses submitted by the Commission before the Court reveal, at least in one of the foreseeable assumptions based on analysis of the D1 email of 10 November 2005, a contestable share of 10.4%.

238    The Commission states, in that regard, in its written pleadings, that a contestable share ranging between 5.6% and 10.4% in an unbiased analysis of the D1 email of 10 November 2005 was consistent with the figure in the 2004 spreadsheet, namely 7.1%.

239    Such a conclusion cannot be accepted, as the result of the AEC test could vary depending on whether the contestable share used was 7.1% or 10.4%. In particular, in recitals 1255 to 1259 of the contested decision, the forecasted contestable share is then compared with the required share stated in Table 22, in relation to which only the last three quarters involve figures exceeding 10.4%. No objective factor makes it possible to rule out one or other of the possible assumptions in the light of the D1 email of 10 November 2005, regarding the scale of 5.6% to 10.4% as a contestable share or even to conclude that one of them was more likely than the other. Accordingly, there remains a doubt as to the definitive percentage of the contestable share for Dell and, more particularly as to that contestable share having to be set at 7.1%.

240    In the second place, Intel submits, in essence, that the conclusions which may be drawn from the switch of Dell’s requirements to AMD demonstrate that Dell’s contestable share could be greater than 7.1%.

241    The Commission maintains that (i) the switch of Dell’s requirements to AMD in 2006 and 2007 was of only limited bearing when assessing the situation during the relevant period, (ii) it would be necessary at the very least to adapt certain calculation parameters, in particular the level of rebates during 2006, (iii) it carried out, as an alternative, an AEC test in the contested decision which took account of the situation in 2006 and 2007, which supports its findings, and (iv) Annex D.9, which was submitted during the proceedings before the Court, makes it possible for Intel’s claims to be contradicted.

242    In recitals 1241 to 1246 of the contested decision, the Commission examined Intel’s argument that the actual switching rate when Dell decided to switch part of its supplies to AMD after 2006 could be relevant for assessing the contestable share. The Commission took the view, in particular, that, while subsequent switching could be informative as such, it should not be given greater weight than the documents showing contemporaneous estimates. Next, when examining Dell’s supply in the three quarters beginning in October 2006 and ending in June 2007, corrected to take account of the transition timing, in the light of its own arguments concerning the starting point for the one-year time horizon, the Commission estimated AMD’s total share at 8.2% according to Gartner data and between 8.8% and 10.1% according to Intel’s internal estimates, during the first year of Dell switching requirements to AMD. The Commission concluded that, while those estimates were slightly higher than Dell’s estimates during the relevant period, they were not at a level where they could be relied on in order to reduce the accuracy of the Commission’s analysis.

243    The Court observes that the Commission expressly acknowledged, in recital 1245 of the contested decision, that, on the basis of observations drawn from the actual switch of part of Dell’s requirements to AMD, it was possible to calculate a contestable share of more than 7.1%, being between 8.2% and 10.1%.

244    Even if the Commission took the view, in the contested decision, that those estimates were slightly higher than the estimate taken from the 2004 spreadsheet, with the result that it was not necessary to take them into account, the fact remains that the very existence of those estimates is sufficient to demonstrate that the assumption of a 7.1% contestable share was not the only conceivable assumption and casts doubt on the substance of the assessment made by the Commission in the contested decision.

245    Before the Court, first of all, the Commission reiterates the argument set out in recitals 1242 and 1243 of the contested decision that the observations deduced from the switch of part of Dell’s requirements to AMD during 2006 and 2007 have only limited bearing on findings regarding the contestable share during the relevant period.

246    However, it must be noted that, in response to Professor Shapiro’s argument that the calculation of the one-year time horizon for the AEC test could not begin after the date on which the partial switch of Dell’s requirements to AMD began to have consequences, the Commission relied conclusively, in recitals 1221 to 1227 of the contested decision, on observations which could be drawn from events in 2006. The Commission inferred, from a set of facts, that Intel was already aware of the switch of supplier in May 2006 and that it had significantly reduced the rebates between the first and second quarters of the 2007 tax year.

247    Accordingly, the Commission, in the context of assessing the contestable share, itself used the observations deduced from the partial switch of Dell’s requirements to AMD in 2006 and 2007 to contradict Professor Shapiro’s argument concerning the starting point for the one-year time horizon.

248    Consequently, the Commission cannot reasonably maintain, in recitals 1242 and 1243 of the contested decision, that the same observations have only a limited bearing in challenging the relevance of the contestable share being assessed at between 8.2% and 10.1%.

249    Next, the Commission submits that a calculation using the figures for the contestable share derived from 2006 and 2007 ought to include, inter alia, the fact that the level of rebates granted by Intel to Dell reached unprecedented levels in 2006. However, if the Commission took the view that the assessment of the contestable share had to be adapted because of that factor, it ought to have included it in the calculation made in recital 1245 of the contested decision.

250    Moreover, the Commission, referring to recital 1258 of the contested decision, argues that the AEC test incorporated the actual market shares which AMD reached at Dell in 2006 and 2007, as provided by Intel during the investigation, and that the results of that calculation support the contested decision’s finding for the period ending in 2005.

251    However, in recital 1258 of the contested decision, while acknowledging that it was possible that the contestable share may have increased somewhat over time, as consumers became aware of the viability of the alternative proposed by AMD, the Commission commented on how the required share evolved during 2006 and the volume of Dell’s requirements which were switched to AMD in 2007. The Commission did not, at that stage, adapt the contestable share used for 2005 based on the calculations made in recital 1245 of the contested decision.

252    Lastly, in the defence and in the rejoinder, the Commission relies on Annex B.31, which contains an analysis based on the partial switch of Dell’s requirements to AMD in 2006 and 2007, confirming the contested decision’s finding on the capability of the rebates to have a foreclosure effect, and on Annex D.9, which shows that the market share which AMD reached at Dell was lower than that in the reply, and used the new figures taken from the reply to carry out an AEC test.

253    However, the Court cannot take account of those additional analyses, which were submitted for the first time during the procedure before it, in order to substantiate the AEC test contained in the contested decision without substituting its own reasoning for that of the Commission set out in that decision. The case-law cited in paragraph 150 above prohibits the Court from making such a substitution.

254    Consequently, it is apparent from the contested decision that it was possible to determine a contestable share for Dell of between 8.2% and 10.1% on the basis of evidence other than the 2004 spreadsheet. The very existence of those estimates demonstrates that the assumption of a 7.1% contestable share in the case of Dell was not the only conceivable assumption, which leads the Court to doubt the substance of that assumption, relied on by Commission in the contested decision.

255    That finding and the finding already set out in paragraph 234 above regarding the question as to whether Dell’s contestable share had to be assessed solely on the basis of the 2004 spreadsheet containing the figure of 7% for 2005, when taken together, support the existence of doubt as to the assessment of that contestable share used in the contested decision.

256    In the light of all the foregoing, it must be concluded that the evidence put forward by Intel is capable of giving rise to doubt in the mind of the Court as to whether the contestable share for Dell had to be set at 7.1%. Consequently, the Commission has not demonstrated to the requisite legal standard that the assessment of that contestable share is well founded.

(iii) The applicant’s claim regarding the initial part of the relevant period, from December 2002 to October 2003

257    Although the conclusion referred to in paragraph 256 above already invalidates, in itself, the assessment of Dell’s contestable share made in the contested decision, it is necessary, for the sake of completeness, to assess, in the light of Intel’s arguments, the substance of the Commission’s analysis of Dell’s contestable share as regards the initial part of the relevant period, between December 2002 and October 2003.

258    According to Intel, the Commission’s finding that Dell’s contestable share was 7.1% is inconsistent with its conclusion in recital 1281 of the contested decision, arrived at on the basis of a comparison of the contestable share with the market share required for an as-efficient competitor to be able to enter the market without incurring losses (‘the required share’), that, throughout the period from December 2002 to December 2005, Intel’s rebates were capable of having or likely to have anticompetitive foreclosure effects.

259    The Commission disputes Intel’s allegations, maintaining that what were involved were merely interim findings, and refers to recitals 1281 and 1282 of the contested decision, which contain an overall assessment.

260    In that regard, it must be stated that Table 22 illustrates clearly that the contestable share was greater than the required share, for the first four quarters indicated, even if the calculations of the required share and the contestable share made by the Commission are accepted. In accordance with Table 22, during the financial years falling within Dell’s accounting period from the fourth quarter of the 2003 tax year until the third quarter of the 2004 tax year, the required share was a maximum of 6.6%, whereas the contestable share used in the contested decision was 7.1%.

261    In addition, in recital 1256 of the contested decision, the Commission expressly states that ‘in most of the quarters (9 out of 13), the required share is higher than the contestable share’. Therefore, and as Intel maintains, according to the Commission’s own figures, the AEC test concerning Intel’s rebates to Dell led to a positive result during the first four quarters covered by the contested decision.

262    As regards recitals 1281 and 1282 of the contested decision, to which the Commission refers (see paragraph 259 above) in order to argue that the comparison of the required share and the contestable share is only one of three elements drawn upon for the conclusion of the AEC analysis, those recitals indicate that the conclusions reached by the Commission as regards the rebates granted to Dell are inferred from the comparison of the contestable share and the required share, the reinforcing factors and the alternative method of calculation, and that the contested decision is based on cost figures which are most favourable to Intel. However, it is apparent from recital 1213 of the contested decision that Table 22 was used in the comparison of the contestable share with the required share. In addition, for the reasons set out in paragraphs 272 to 282 below, neither the alternative method of calculation nor the reinforcing factors involve consideration of the foreclosure effect of the rebates for the first four quarters covered by the contested decision. Therefore, those three elements of the Commission’s analysis, even taken together, provide no explanation for the fact that the AEC test for Intel’s rebates to Dell had a positive result during the first four quarters covered by the contested decision.

263    Accordingly, there is a contradiction between what is stated in recital 1256 of the contested decision, namely that in at least four quarters of the relevant period Intel passed the AEC test, and the Commission’s findings in recitals 1281 and 1282 of that decision, from which it is apparent that the rebates granted to Dell were capable of having a foreclosure effect throughout the whole of the relevant period.

264    Next, the other parts of the contested decision, to which the Commission refers in order to demonstrate that there was no error in relation to the first four quarters, are also inconclusive as regards the period between December 2002 and October 2003. According to the Commission, recitals 1258 and 1259 of the contested decision show that a strict quarter-by-quarter approach was not relevant.

265    More particularly, in recital 1258 of the contested decision, the Commission states that it is possible that the contestable share increased over time as consumers become increasingly aware of the viability of the AMD alternative. The Commission also notes that, in all calculation hypotheses, the required share increased steadily over the period covered by the contested decision. The Commission also refers to the actual figures derived from the situation prevailing in 2006 when Dell had chosen to begin sourcing from AMD. It relies, in particular, on data provided by Gartner.

266    In recital 1259 of the contested decision, the Commission states that, conversely, it is possible that the contestable share was lower than 7.1% during the period preceding the first quarter of the 2005 tax year, the earliest date on which Dell’s switch of part of its x86 CPU requirements from Intel to AMD could have taken place in the scenario which formed the basis of the 2004 spreadsheet. The Commission concludes that the difference between the required share and the contestable share in the first quarters of the relevant period may be lower than the figures from Table 22 would suggest.

267    As regards that point, the applicant makes the argument that the Commission never adjusted its contestable share assessment for the first four quarters of the relevant period to reflect that increase in AMD’s viability, which, according to the applicant, did not take occur overnight.

268    It must be stated that the Commission did not in any way quantify in the contested decision that assumed increase in the contestable share, taking account of consumers’ changing perception of AMD over time. On the contrary, only the figure of 7.1% is used in the contested decision, even though the 2004 spreadsheet envisaged that figure evolving during the subsequent years under consideration, providing values from which it could be inferred that Dell’s contestable share was 17.3%, 22.5% and 24.2% for the three years following the initial year of partial supply by AMD.

269    Nowhere in the contested decision is it stated definitively that Dell’s contestable share had increased over time, due to an improvement in the perception of AMD’s products, rather, it is simply stated, in recitals 1258 and 1259 of that decision, that that was ‘possible’. Furthermore, even Table 22 assesses solely the changes over time in the required share, on a multiannual basis, but not changes in the contestable share. However, at the 2020 hearing, in response to a question from the Court in that regard, the Commission merely stated that the 7.1% used concerning the contestable share was used consistently for the entire relevant period for ‘technical reasons’, linked to an alleged agreement between Intel and the Commission regarding the use of a one-year period for the AEC analysis. Accordingly, even though recital 1212 of the contested decision sets out four different data points taken from the 2004 spreadsheet, only the 7.1% figure is deemed to be appropriate for the contestable share, in recital 1213 of the contested decision.

270    In those circumstances, the Commission’s arguments do not make it possible to explain or validate, a posteriori, the difference between the results indicated by the Commission for the first four quarters of the relevant period in Table 22, and its conclusion, adopted for the whole of the relevant period, that Intel did not pass the AEC test.

271    It must therefore be held that, since the AEC test result was positive for Intel in the principal calculation for the first four quarters covered by the contested decision, the Commission has not demonstrated solely on the basis of that test that the rebates which Intel granted to Dell were capable of restricting competition throughout the whole of the relevant period.

(2)    The alternative method of calculation

272    In recitals 1266 to 1274 and 1281 of the contested decision, the Commission made an alternative calculation, on the basis of information contained in Dell’s 17 February 2004 presentation, which, according to the Commission, confirmed the conclusions drawn from the principal calculation in the AEC test, namely that the rebates granted by Intel to Dell were capable of foreclosing an as-efficient competitor.

273    Intel disputes the relevance of the alternative calculation. It argues that that assessment relates solely to the 2005 tax year, which falls outside the period for which the contested decision demonstrates that Intel passed the AEC test. In its view, the infringement finding relating to Dell for the period between December 2002 and October 2003 cannot therefore be upheld.

274    The Commission rejects Intel’s arguments and takes the view that the alternative calculation confirms the solution adopted in the contested decision as its principal argument.

275    In that regard, in so far as the contested decision bases its alternative calculation on Dell’s 17 February 2004 presentation, which, as is also apparent from Tables 28 and 29 in recitals 1268 and 1270 of that decision, assesses the period commencing with the 2005 tax year, it cannot be inferred that it makes it possible to explain or, a fortiori, to modify the Commission’s assessments for the period between December 2002 and October 2003. In addition, in so far as the Commission referred, at the 2020 hearing, to footnote 1604 to recital 1264 of the contested decision to argue that the documents used did indeed concern the relevant period, suffice it to note that that footnote concerns the reinforcing factors and not the alternative method of calculation.

276    Consequently, there being no need to give a ruling on the substance of the alternative method, it is sufficient to hold that that method does not demonstrate that Intel’s rebates were capable of having a foreclosure effect throughout the whole of the relevant period.

(3)    The reinforcing factors

277    According to the applicant, the Commission cannot substantiate its analysis by stating that its application of the as-efficient-competitor test is actually conservative, in not taking account of reinforcing factors (see also paragraph 177 above). The Commission, by contrast, maintains, in essence, that it was justified in taking account of the reinforcing factors.

278    It is therefore necessary to analyse whether the various Commission errors in the AEC test for Dell could be corrected by the various elements taken into account as reinforcing factors, as set out in recitals 1260 to 1265 of the contested decision.

279    In that regard, first, it is apparent from recital 1260 of the contested decision that the relevance of the reinforcing factors is that ‘a number of factors have not been fully taken into account in the [preceding] analysis, but if included, would reinforce the assessed capability to foreclose of the rebates’. Accordingly, the sole purpose of the factors in question was to reinforce the principal assessment as to whether there was a foreclosure effect.

280    Second, it is apparent from recital 1261 of the contested decision that the Commission itself took the view that a full assessment of the effects of the reinforcing factors would have required additional assumptions as to how the rebates would be reallocated to other competitors, as well as how more aggressive competition would affect Dell’s revenues.

281    In response to a question put by the Court at the 2020 hearing, the Commission stated that the reinforcing factors had been the subject of a ‘pragmatic assessment’ and that they were ‘sui generis’ elements, included in the structure of the contested decision concerning the AEC test. However, the Commission has nonetheless failed to assert that they were assessed precisely, in figures, in the context of the AEC test. Rather, the Commission states that those elements, irrespective of whether or not they were lawful, constituted ‘additional leverage’ in Intel’s favour, in so far as the rebates lost by Dell would be transferred to competitors and given that those rebate losses could affect other chips purchased from Intel, which ‘were not covered by the contested decision’.

282    It follows from the foregoing that the reinforcing factors were included in the contested decision as elements which were capable of reinforcing the principal assessment regarding the existence of a foreclosure effect caused by the rebates at issue and that those factors were not analysed sufficiently by the Commission as regards the effect that they would have had on the assessment of the foreclosure capability of those rebates. Accordingly, the Court cannot review effectively how those factors were taken into account, nor can consideration of those factors replace the Commission’s principal analysis of the capability of Intel’s rebates to Dell to have a foreclosure effect.

(4)    Conclusion on the AEC test for the rebates granted to Dell

283    It follows from all of the foregoing that the Commission has not established to the requisite legal standard that it is well founded in assuming that Dell’s contestable share for the period under consideration was 7.1%. Since, in recitals 1255 to 1257 of the contested decision, that assumption formed the basis for demonstrating, by comparing the required share and the contestable share, the capability of Intel’s rebates to Dell to have a foreclosure effect, there being no need to analyse the applicant’s complaints regarding the calculation of the conditional portion, it follows that that comparison has not demonstrated that capability to the requisite legal standard.

284    In addition, the reinforcing factors are not, in themselves, capable of demonstrating that Intel’s rebates granted to Dell were capable of having a foreclosure effect and, in any event, were not analysed sufficiently, while the alternative calculation method does not demonstrate that Intel’s rebates were capable of having a foreclosure effect throughout the whole of the relevant period.

285    It must be observed that, in recital 1281 of the contested decision, the Commission stated that the conclusions which it had reached as regards the capability of the rebates granted to Dell to have a foreclosure effect are inferred from the comparison of the contestable share and the required share, the reinforcing factors and the confirmation provided by the alternative calculation method.

286    However, since the comparison of the contestable share and the required share does not demonstrate, to the requisite legal standard, the foreclosure effects and since the reinforcing factors were not analysed sufficiently, the Commission is not in a position to establish, on the basis of those first two elements, the capability of the rebates granted to Dell to have a foreclosure effect. In addition, the third element relied on by the Commission, which consists of an alternative method of calculation, which, according to recital 1281 of the contested decision, confirms the first two elements, cannot in itself substantiate the Commission’s findings, a fortiori since it does not demonstrate that Intel’s rebates were capable of having a foreclosure effect throughout the whole of the relevant period.

287    It is therefore necessary to uphold the applicant’s complaint that the Commission has not established to the requisite legal standard the validity of the conclusion set out in recital 1281 of the contested decision that, over the period between December 2002 and 2005, Intel’s rebates were capable of having or were likely to have an anticompetitive foreclosure effect, since even an as-efficient competitor would have been prevented from supplying Dell for its x86 CPU requirements.

(b)    The alleged errors in the AEC test applied to HP

288    In recital 413 of the contested decision, the Commission stated that HP and the applicant concluded, for the period between November 2002 and May 2005, the HPA agreements concerning corporate desktop computers. In the same recital, the Commission observed that those agreements included an unwritten condition for the grant of rebates to HP (‘the HPA rebates’), namely that HP would source from the applicant at least 95% of its x86 CPU requirements for its corporate desktop computers (‘the quasi-exclusivity condition’). In recital 1406 of the contested decision, the Commission concluded on the basis of the AEC test that those HPA rebates were capable of having anticompetitive foreclosure effects.

289    As regards, more specifically, the periods covered by the agreements which resulted in the HPA rebates, the Commission stated in recitals 338, 341 and 1296 of the contested decision that the first of those agreements (‘the HPA1 agreement’) had been concluded following the merger in May 2002 of HP with Compaq and covered a period from November 2002 to May 2004. As regards the second of the HPA agreements, the Commission noted in recitals 342 and 343 of the contested decision that it covered a period from June 2004 to May 2005.

290    The applicant disputes the Commission’s conclusion that the HPA rebates were capable of having anticompetitive foreclosure effects and submits that, when applied correctly, the AEC test shows that those rebates were not capable of foreclosing an as-efficient competitor.

291    In essence, the applicant claims that the contested decision contains four errors, concerning (i) the contestable share, (ii) the amount of the conditional portion of the rebates, (iii) the infringement period examined, and (iv) the reinforcing factors taken into consideration. The applicant puts forward a fifth argument to the effect that the Commission erred in its assessment of its AAC.

(1)    The period examined by the AEC test

292    The applicant maintains that the Commission did not carry out the AEC test for the entire period covered by the contested decision. The contested decision fails to offer an AEC test in respect of the first 11 months of the period concerned, namely from November 2002 to the third quarter of HP’s 2003 tax year. In addition, although Table 35 in recital 1337 of the contested decision (‘Table 35’) presents a ‘robustness’ analysis which allegedly examines the entirety of the HPA1 agreement, that analysis is based, in the applicant’s view, on incomplete data. The Commission, it argues, made a ‘manifest error of assessment’ in stating that the applicant failed the AEC test for the period covered by the HPA1 agreement, while acknowledging that it was possible, in the absence of consistent data, that the reference period ‘may not fully coincide with the actual contractual duration’ of the HPA1 agreement.

293    Furthermore, the applicant submits that the Commission’s assertion in recital 1014 of the contested decision that the AEC test requires an examination of the contestable share of an OEM’s supplies over a period of at most one year is inconsistent with the analysis in Table 35, which was conducted over a longer period, namely over one and a half years.

294    The Commission contends that the contested decision does contain an AEC analysis for the entire period covered by the HPA1 agreement, which runs from November 2002 to May 2004, and that Annex B.31 demonstrates how that is a relevant period.

295    Next, the Commission maintains that the applicant failed to raise that argument at any point during the administrative procedure, even though the same reference periods were used for all the calculations regarding HP. Moreover, the applicant used those reference periods for its own calculations concerning HP in its response to the Statement of Objections of 2007.

296    Lastly, the Commission sets out the reason why the calculation for the entire duration of the HPA1 agreement’s validity – namely a year and half – does not use contestable shares established for a one-year and a half time horizon, but instead an average over the life of the quarterly contestable shares during the period of validity of the HPA1 agreement established for a one-year horizon. According to the Commission, whatever the point in time in which the as-efficient competitor attempted to enter the HP market, HP would have had to assess that competitor’s proposal for the year commencing at the point of its market entry.

297    In recitals 1334 to 1337 of the contested decision, the Commission set out its calculations regarding HP’s required share.

298    In recitals 1385 to 1387 of the contested decision, the Commission, referring to the figures in recital 1334 of that decision, concluded that the required share was consistently above the contestable share.

299    In recital 1406 of the contested decision, the Commission found that, on the basis of the comparison of the contestable share and the required share conducted in recitals 1385 to 1389 of that decision, it had been concluded that, during the period from November 2002 to May 2005, the rebates granted by Intel to HP were capable of having an anticompetitive foreclosure effect.

300    In the first place, it should be recalled that, according to the case-law, as regards the application of Articles 101 and 102 TFEU, there is no requirement under the law of the European Union that the addressee of a statement of objections must challenge its various matters of fact or of law during the administrative procedure, if it is not to be barred from doing so later at the stage of the judicial proceedings (judgment of 1 July 2010, Knauf Gips v Commission, C‑407/08 P, EU:C:2010:389, paragraph 89).

301    Therefore, the Commission’s argument that the applicant did not challenge, during the administrative procedure, the periods used by the Commission for its calculations cannot succeed.

302    The same applies with regard to the Commission’s argument that the applicant used the periods in question for its own calculations during the administrative procedure. Since certain periods were used by the Commission for its own calculations in the contested decision, they form part of the reasoning of that decision which the applicant may challenge before the Court.

303    In the second place, it must be observed that Table 34, which is found in recital 1334 of the contested decision, setting out the parameters and calculations of the required share (‘Table 34’) covers the period from the fourth quarter of the 2003 tax year to the third quarter of the 2005 tax year, with the result that it does not incorporate any data relating to the months of November and December 2002 or the first three quarters of the 2003 tax year.

304    In addition, the applicant is correct in arguing that the figures for the HPA1 agreement found in the first line of Table 35, which seek to determine the robustness of the Commission’s conclusions by setting out the calculation of the required share for the HPA agreements, are derived from the arithmetic sum or average of the figures in the first three lines of Table 34.

305    Specifically:

–        the number of x86 CPUs purchased by HP set out in Table 35 for the period covered by the HPA1 agreement, namely 7 079 382 units, matches the number of x86 CPUs purchased by HP over the period from the fourth quarter of the 2003 tax year to the second quarter of the 2004 tax year, set out in Table 34 (fourth quarter of the 2003 tax year: 2 416 750 units; first quarter of the 2004 tax year: 2 200 225 units; second quarter of the 2004 tax year: 2 462 407 units);

–        the amount of the rebates received by HP set out in Table 35 for the period covered by the HPA1 agreement, namely USD 97 499 999, matches the rebates received by HP over the period from the fourth quarter of the 2003 tax year to the second quarter of the 2004 tax year, set out in Table 34 (fourth quarter of the 2003 tax year: USD 32 499 999; first quarter of the 2004 tax year: USD 32 500 000; second quarter of the 2004 tax year: USD 32 500 000);

–        the value ‘V’ (namely, the percentage of the total volume of x86 CPU units which HP would have purchased from the applicant while complying with the quasi-exclusivity condition) set out in Table 35 for the period covered by the HPA1 agreement, namely 6 725 413 units, matches, to within one unit and taking into account a typing error, the ‘V’ values for the period from the fourth quarter of the 2003 tax year to the second quarter of the 2004 tax year, set out in Table 34 (fourth quarter of the 2003 tax year: 2 295 913 units; first quarter of the 2004 tax year: 2 090 214 units; second quarter of the 2004 tax year: 2 339 287 units);

–        the ‘ASP of Intel CPUs’ set out in Table 35 for the period covered by the HPA1 agreement, namely 165.15, matches the unweighted arithmetical average of the ASP identified, for the period from the fourth quarter of the 2003 tax year to the second quarter of the 2004 tax year in Table 34 (fourth quarter of the 2003 tax year: 176.19; first quarter of the 2004 tax year: 159.45; second quarter of the 2004 tax year: 159.82).

306    In that regard, it should be noted that the Commission does not maintain that the demonstration set out above is the product of a coincidence and that the various values set out in paragraph 305 above are identical for the missing three quarters and for the following three quarters.

307    Consequently, the foregoing is sufficient to demonstrate that the months of November and December 2002 and the first three quarters of the 2003 tax year were not actually taken into account by the Commission in the calculations which gave rise to the figures set out in Table 35. The calculation of the required share throughout the duration of the HPA1 agreement which produced the results set out in Tables 34 and 35 therefore does not cover the whole of the period between November 2002 and May 2005, in relation to which the Commission took the view that it could demonstrate that there was a foreclosure effect caused by the rebates which Intel granted to HP.

308    In the third place, the arguments put forward by the Commission are not such as to call into question the foregoing conclusion.

309    First of all, in the rejoinder, the Commission maintains that the result of a quarterly calculation does not differ fundamentally from the result of the overall calculation which is alleged to have been made.

310    However, that argument is made in response to the reply, in order to claim that the approach used in the contested decision, based on an average of the quarterly contestable shares, in which the contestable share is calculated over a maximum period of one year, was not incompatible with making a calculation over the entire duration of the validity of the HPA1 agreement. Although the calculations made by the Commission in the contested decision do not take account of the data for the months of November and December 2002 or for the first three quarters of the 2003 tax year over the duration of the HPA1 agreement, it matters little whether those calculations are made on a quarter-by-quarter basis or on an overall basis, since the months of November and December 2002 and the first three quarters of the 2003 tax year will never be taken into account, whatever figures they may involve.

311    Next, in the defence and the rejoinder, the Commission, in support of its arguments, refers to annexes B.31. and D.17 to those documents.

312    As regards the reference in the defence to Annex B.31, it should be recalled that, whilst the body of the application may be supported and supplemented on specific points by references to extracts from documents annexed thereto, a general reference to other documents, even those annexed to the application, cannot make up for the absence of the essential arguments in law which, in accordance with Article 21 of the Statute of the Court of Justice of the European Union and Article 76 of the Rules of Procedure of the General Court, must appear in the application (judgment of 17 September 2007, Microsoft v Commission, T‑201/04, EU:T:2007:289, paragraph 94).

313    Furthermore, it is not for the Court to seek and identify in the annexes the pleas and arguments on which it may consider the action to be based, since the annexes have a purely evidential and instrumental function (judgment of 17 September 2007, Microsoft v Commission, T‑201/04, EU:T:2007:289, paragraph 94).

314    Thus, an annex to the application may be taken into consideration only in so far as it supports or supplements the arguments expressly set out by the applicant in the body of the application and in so far as it is possible for the Court to determine precisely what are the matters contained in the annex that support or supplement those arguments (see, to that effect, judgment of 17 September 2007, Microsoft v Commission, T‑201/04, EU:T:2007:289, paragraph 99).

315    In the present case, in the defence, the Commission simply states that the period analysed in the contested decision, namely the full period covered by the HPA1 agreement, is relevant for the purposes of the AEC test, without developing that argument, and it refers, without providing further information, to the explanations set out in Annex B.31, without it being possible for the Court to determine precisely which elements in that annex might substantiate that undeveloped argument. It follows that that argument is inadmissible, through application, by analogy, of the case-law referred to in paragraphs 312 to 314 above.

316    As regards the rejoinder, the Commission refers to paragraphs 77 to 82 of Annex D.17 to submit that quarterly calculations based on a figure supplied by HP lead to results which are less favourable to Intel than the average results on which the contested decision is based.

317    In so far as the Commission submits, in Annex D.17 to the rejoinder, a calculation for two of the three missing quarters, namely the second and third quarters of the 2003 tax year, it should be noted that those calculations are not apparent from the contested decision and are submitted for the first time during the judicial proceedings. The Court cannot therefore take those additional calculations into account in order to substantiate the AEC test contained in the contested decision without substituting its own reasoning for that of the Commission in that decision. However, the case-law cited in paragraph 150 above prohibits the Court from making such a substitution.

318    Be that as it may, it must be stated that there is nothing to demonstrate the accuracy of the premiss put forward by the Commission that, because the rebates were stable during the period covered by the HPA1 agreement, the results of the required share would be the same for the missing two months and three quarters. In addition, it should be recalled that the required share is calculated by reference to three parameters, namely the amount of the rebates, the volume of HP’s purchases and the ASP. It has not been demonstrated in the contested decision that the value of those last two parameters, for the missing two months and three quarters, was identical to the values identified in the examination of the quarterly periods taken into account. Also, there is no guarantee that the data for the months and quarters not taken into account for the AEC test do not differ from those identified for the quarterly periods analysed.

319    Consequently, it follows from the foregoing that the Commission erred in finding that its calculation of the required share allowed it to draw conclusions concerning the foreclosure effect of the rebates which Intel granted to HP for the entire period between November 2002 and May 2005. The Commission has not demonstrated that that effect was present for the period between November 2002 and September 2003.

320    The fact that the Commission carried out an alternative calculation of the required share in recital 1389 of the contested decision, using the figures set out in recital 1338 of that decision, cannot make up for that error. It is apparent from Tables 36 and 37 that the data concerning the required share in the two alternative scenarios envisaged by the Commission cover the period from the fourth quarter of 2004 to the third quarter of 2005 and the period from the second quarter to the third quarter of 2005 respectively. Accordingly, that alternative calculation also does not cover the whole of the period between November 2002 and May 2005.

(2)    The alleged reinforcing factors

321    In recitals 1390 to 1395 of the contested decision, the Commission stated, in essence, that the AEC test did not take account of two additional considerations, namely, first, that the Commission had used figures which were most favourable to the applicant and, second, that, if HP were to switch its purchases of x86 CPUs to AMD, the applicant could in turn transfer the rebates initially intended for HP to another competitor using its x86 CPUs, such as Dell. That, according to the Commission, exacerbates the disadvantages for HP resulting from a switch of its purchases of x86 CPUs to AMD.

322    The applicant submits, first, that the Commission does not explain why an increase in the rebates granted to HP’s competitors, in order to meet competition, is anticompetitive. Second, it is apparent from the HP document entitled ‘Managing Intel and AMD to maximise value to BPC’ that HP itself had concluded that a measure of that type did not constitute a real risk and that that phenomenon had not been observed in other global business units with a richer mix of AMD products. Third, if HP had received one million x86 CPUs from AMD free of charge, it would have avoided paying USD 163.86 million to the applicant (namely the ASP excluding rebates for one million x86 CPUs). The applicant’s total rebates provided for in the HPA1 agreement amounted to only USD 130 million, with the result that HP would have had to pay almost USD 34 million to purchase the equivalent number of x86 CPUs from the applicant. HP, in the applicant’s view, therefore must have rejected AMD’s offer simply because there was insufficient demand for AMD-based x86 CPU systems and did not take into consideration the potential loss of the rebates granted by the applicant. Fourth, it is also apparent from the HP document entitled ‘Managing Intel and AMD to maximise value to BPC’ that there was uncertainty over market acceptance of AMD by enterprise customers.

323    The Commission asserts, first, that the possibility of a rebate being transferred to HP’s competitors would strengthen the economic incentives so that HP would not be moved to infringe the conditions of the HPA agreements. Second, the HP document entitled ‘Managing Intel and AMD to maximise value to BPC’ does not address transferring rebates to competitors. Third, HP’s decision not to accept AMD’s offer of one million free x86 CPUs is not the result of a mere accounting comparison. Unlike the AEC test, which is purely theoretical, real business decisions are influenced by multiple factors. The Commission argues that the applicant’s calculations are, moreover, flawed, because, while HP was able, under the HPA agreements, to obtain small quantities from AMD, it ultimately acquired 160 000 x86 CPUs. Accordingly, HP did not refuse one million x86 CPUs, but only 840 000 x86 CPUs. With an ASP of USD 163.86 per unit, the saved amount is equal to only USD 137.6 million, which is not significantly different from the USD 130 million of HPA rebates.

324    In that regard, before considering whether or not the Commission erred in its assessment in the contested decision of the reinforcing factor identified therein, consisting of a transfer of the rebates initially granted to HP to its competitors, it must be observed that the contested decision contains no analysis of the effect of that factor on the factors taken into consideration in the AEC test.

325    It has consistently been held that an absence of or an inadequate statement of reasons constitutes an infringement of essential procedural requirements for the purposes of Article 263 TFEU and is a plea involving a matter of public policy which may, and even must, be raised by the EU Courts of their own motion (see judgment of 2 December 2009, Commission v Ireland and Others, C‑89/08 P, EU:C:2009:742, paragraph 34 and the case-law cited).

326    In the light of the foregoing, the Court is required to rule on whether there has been a possible failure to fulfil the obligation to state reasons and to hear the parties for that purpose, which it did at the 2020 hearing.

327    It should be recalled that, according to settled case-law, the scope of the obligation to state reasons depends on the nature of the measure in question and on the context in which it was adopted. The statement of reasons must disclose in a clear and unequivocal fashion the reasoning followed by the institution, in such a way as to enable the persons concerned to ascertain the reasons for the measure so that they can defend their rights and ascertain whether or not the measure is well founded and to enable the EU Courts to carry out their review. It is not necessary for the reasoning to go into all the relevant facts and points of law, since the question whether the statement of reasons meets the requirements of Article 296 TFEU must be assessed with regard not only to its wording but also to its context and to all the legal rules governing the matter in question (see judgment of 18 January 2012, Djebel – SGPS v Commission, T‑422/07, not published, EU:T:2012:11, paragraph 52 and the case-law cited).

328    In the present case, the fact remains that, although the Commission took the view that the transfer of the rebates initially intended for HP to its competitors was a reinforcing factor supporting the conclusions which it drew from the AEC test, the Commission did not specify which of the factors taken into account in that test would be influenced and how. Since the Commission found that that reinforcing factor played a role in the assessment of the foreclosure capability of the rebates at issue, the Commission was required to assess more precisely the effect of the reinforcing factor on the foreclosure capability of the rebates at issue. That is all the more so since the Commission found, in recital 1395 of the contested decision, that that factor prevailed over all the arguments made by the applicant during the administrative procedure on the factors used by the Commission to apply the AEC test in respect of HP.

329    At the 2020 hearing, in response to a question from the Court on (i) its reasoning that the reinforcing factor consisting of a transfer of the rebates initially granted to HP to one of its competitors outweighed all the errors contained in the contested decision and (ii) the reasoning of the contested decision in that regard, the Commission simply stated that no reasonable business partner would have refused AMD’s offer to provide it with it one million x86 CPUs free of charge. Also, HP refused AMD’s offer solely because of the consequences that acceptance would have had on its relationship with the applicant. The Commission considered that it had nothing to add to what was stated in the contested decision.

330    Accordingly, there being no need to give a ruling on its admissibility, it must be held that the argument put forward by the Commission at the 2020 hearing is a simple unsubstantiated assumption which cannot make up for the lack of reasoning in the contested decision as to the influence of the reinforcing factor, consisting of a transfer of the rebates initially granted to HP to one of its competitors, on the Commission’s conclusions stemming from the AEC test.

331    In the light of the foregoing, it must be held that the contested decision, as regards the reinforcing factor consisting of a transfer of the rebates initially granted to HP to one of its competitors, is vitiated by a failure to state reasons.

332    In the light of the foregoing, it should be noted that, in recital 1406 of the contested decision, in order to assert that it had demonstrated the capability of the rebates granted to HP to have a foreclosure effect, the Commission relied on the comparison of the contestable share and the required share, the reinforcing factors and the absence of relevance of Intel’s allegations concerning a ‘new theory’ by the Commission.

333    First of all, it is apparent from recitals 1396 to 1405 of the contested decision that the examination of the absence of relevance of Intel’s allegations concerning a ‘new theory’ by the Commission is not an alternative AEC test, but a challenge to new calculations submitted by Intel in its observations of 28 March 2008, with the result that it cannot be regarded as an examination by the Commission which is designed to establish the capability of the rebates at issue to have a foreclosure effect.

334    Next, it follows from all of the foregoing, as regards the AEC test applied to HP, that the Commission, first, when comparing the contestable share and the required share, has not demonstrated that there were foreclosure effects in the period between 1 November 2002 and 31 September 2003 and, second, has failed to provide reasons, to the requisite legal standard, for examining the reinforcing factors.

335    Consequently, the Commission has not established to the requisite legal standard the validity of the conclusion set out in recital 1406 of the contested decision, that, during the period from November 2002 to May 2005, Intel’s rebate to HP was capable of having an anticompetitive foreclosure effect or was likely to have such an effect, since it has not demonstrated that there were foreclosure effects in the period between 1 November 2002 and 31 September 2003.

(c)    The alleged errors in the AEC test applied to NEC

336    In recitals 451 to 453 to the contested decision, the Commission stated that NEC was one of the ten largest sellers of personal computers and servers worldwide. Until April 2005, NEC’s operations as an OEM were managed by two wholly owned subsidiaries, namely NEC Japan and NEC Computer International (‘NECCI’). NEC Japan managed NEC’s activities in Japan and the Americas, whereas NEC’s operations in the rest of the world were handled by NECCI. NECCI was based in Europe, but also managed NEC’s operations in Asia (with the exception of Japan) through its Asia-Pacific countries branch. In April 2005, the corporate structure was modified and the Asia-Pacific countries branch was hived off from NECCI and transferred back to NEC Corporation.

337    It is also apparent, in particular, from recitals 483, 501 to 502 and 981 of the contested decision that, first, between October 2002 and November 2005, Intel had granted rebates to NEC under an arrangement called the ‘Santa Clara Agreement’ adopted in May 2002 (‘the Santa Clara agreement’), second, the rebates granted under that agreement were de facto conditional on NEC agreeing to purchase from Intel 80% of its x86 CPU requirements worldwide, with that global share being broken down into a 70% requirement for NECCI and a 90% requirement for NEC Japan, and third, in order to prove that they had reached the required market share, NEC and NECCI were required to notify their market shares to Intel on a quarterly basis.

338    Under the Santa Clara agreement, Intel states that it provided NEC with both discounts known as ‘exception to customer authori[s]ed pricing’ (‘ECAP’) and market development funds (‘MDF’). The Commission noted, in recital 466 of the contested decision, that, after 1 July 2003, the structure of Intel’s payments changed and that the MDF had been subsumed into ECAP and had been renamed ‘super ECAP’.

339    The Commission examined the rebates granted by Intel to NEC using the effective price method in the AEC test. According to that method, the Commission calculated the ratio between the total value of the payments granted under the Santa Clara agreement and the value of the Intel business at risk in the fourth quarter of 2002 in order to achieve a measure of the effective price. The Commission then compared that ratio with the existing ratio between Intel’s ASP and AAC and concluded that Intel had priced below its costs because the first ratio was lower than the ratio between the ASP and AAC.

340    The applicant maintains that the Commission’s calculations are vitiated by five errors, each of which is sufficient to overturn its conclusions. Intel asserts, first, that the Commission’s own data show that the rebates granted to NEC are not capable of foreclosing an as-efficient competitor; second, that the Commission erred in calculating the conditional portion of the rebates granted to NEC; third, that the Commission incorrectly calculated the value of the business at risk for Intel; fourth, that the Commission relied on an incorrect value in determining Intel’s AAC and, fifth, that the Commission erred in assuming that the fourth quarter of 2002 was representative of the entire period for which abuse had been found.

341    The Court considers it appropriate to examine first the substance of the argument which seeks to demonstrate that the Commission erred in its calculation of the conditional portion of the rebates.

(1)    The calculation of the conditional portion of the rebates

342    The applicant states that, according to recitals 1408, 1443 and 1444 of the contested decision, all the discounts granted to NEC in the fourth quarter of 2002 were conditional. First, it argues, that argument is not supported by the evidence put forward in the contested decision and is, moreover, contradicted by the unambiguous responses provided by NECCI under Article 18 of Regulation No 1/2003 and by other evidence, according to which the USD 6 million of MDF provided in respect of the fourth quarter of 2002 was the only advantage granted to NEC under the market share commitment provided for in the Santa Clara agreement. According to Intel, the Commission was therefore wrong to take the view that ECAP was conditional. Second, Intel argues that NEC received significant rebates from Intel during the periods prior to the agreement, when Intel’s market share for NEC’s purchases was significantly below 80%. The Commission does not explain why NEC would have lost 100% of its rebates if it had purchased less than 80% of its x86 CPUs from Intel even though it had already done so without suffering such a loss. Third, it is not disputed that Intel granted rebates to NEC even though NEC did not reach the 80% threshold, which, according to the findings in the contested decision, was a prerequisite to obtaining any discount.

343    The Commission disputes Intel’s arguments. First, it states that the contested decision contains no finding that all the rebates granted to NEC were conditional. The contested decision simply states that the conditional portion of the Intel rebates included not only MDF payments but also certain – but not necessarily all – categories of ECAP-type rebates. That finding is, according to the Commission, based on a consistent, precise and firm body of evidence set out in recitals 1412 to 1444 of the contested decision and in Annex B.31 to the defence.

344    The Commission’s view is that Intel’s argument based on the grant of significant rebates during the periods preceding the Santa Clara agreement is not persuasive, inter alia, on the ground, first, that the conditions applicable to the rebates previously granted are unknown and, second, that the data submitted by NECCI show an increase of some 500% in rebates granted by Intel to NECCI following that agreement.

345    Lastly, in so far as the applicant asserts that it granted rebates to NEC even though NEC did not manage to fulfil the 80% market share condition, the Commission adds, in the context of the AEC analysis, that, even if Intel’s claims that AMD’s market share at NEC had ‘routinely’ exceeded the 20% threshold were correct, AMD’s share at NEC never came close to the contestable share (41%).

346    As regards the applicant’s first argument that the Commission erred in taking the view that ECAP was conditional, it is necessary to determine whether the Commission demonstrated in the contested decision that the rebates taken into consideration in the calculation of the effective price of Intel x86 CPUs sold to NEC other than MDF, namely ECAP, were conditional on NEC complying with its obligation to source from Intel a certain percentage of its x86 CPU purchases.

347    In recitals 1415 to 1444 of the contested decision, the Commission assessed the total value of conditional rebates as falling between USD 13 088 100 and USD 16 583 100, of which USD 6 million consisted of MDF and the remainder of ECAP.

348    Consequently, a finding that ECAP is not subject to a specific market share would necessarily call into question the Commission’s calculations, as they appear in the contested decision.

349    It is therefore necessary to assess, on the basis of the evidence relating to the fourth quarter of 2002 upon which the Commission based the AEC test in respect of NEC, whether, during that quarter, payments other than MDF were conditional upon NEC obtaining supplies from Intel up to a certain percentage of market segment share (‘MSS’). At the outset, it must be pointed out that the applicant does not deny that, under the Santa Clara agreement, it provided NEC with both MDF- and ECAP-type rebates. However, the applicant maintains that, unlike the first of those rebates, ECAP was not conditional on the obligation to reach a certain level of MSS.

350    The applicant claims, in essence, that the evidence presented in the contested decision does not support the conclusion that ECAP was conditional in the fourth quarter of 2002, and it draws attention to other documents, from which it claims that it is apparent that the only rebates conditional on NEC’s obligation to reach a certain level of MSS were MDF. The Commission disputes the applicant’s arguments and maintains that the evidence adduced by Intel does not demonstrate that MDF was the only rebate conditional on NEC’s obligation to obtain supplies from Intel up to a certain percentage of MSS.

(i)    The evidence taken into account in the contested decision

351    In the first place, it must be pointed out that the Commission relied, inter alia, in recitals 461 and 464 of the contested decision, on a presentation by NEC of 27 January 2003, entitled ‘NEC/Intel WW Meeting (Purchasing Session)’ and, more specifically, on the fourth page of that presentation, entitled ‘WW: CQ4/Y2002 Achievements’. That page confirms, under the heading ‘Original Plan’, that NEC’s intention was to obtain supplies from Intel for only 59% of its requirements, namely 68% for NEC Japan, the division of NEC active on, inter alia, the Japanese market, and 48% for NECCI. In addition, that page sets out, under the heading ‘Realignment Plans’, first, the market shares envisaged by Intel, namely 70% for NECCI, 90% for NEC Japan and 80% at a worldwide level and, second, certain rebates and other benefits to be granted to NEC by Intel. Those plans deal with, inter alia, MDF, reduced prices (rebates) for x86 CPUs, ‘MultiNational Corporation status’ and a Supply Line Agreement.

352    However, it must be stated that, while that document post-dates both the conclusion of the Santa Clara agreement and the quarter in question, and mentions reduced prices, that is to say ECAP, as being one of the advantages enjoyed by NEC under that agreement and confirms that that ECAP formed part of that agreement, which is not disputed by Intel, it does not follow that ECAP was conditional on a certain level of MSS. That document is therefore, at most, merely circumstantial and must be confirmed by other evidence.

353    In the second place, recitals 462 and 464 of the contested decision are based on an email of 15 May 2002 in which a senior NEC executive informed an NECCI executive that, following a conference call held on the same day with Intel’s executives, NEC would have MultiNational Corporation status, that it would increase its share of purchases of x86 CPUs from Intel to a certain worldwide percentage of its total sales and that Intel would give MDF and ‘aggressive prices’, that is to say reduced prices to NEC for ‘Celeron’ x86 CPUs.

354    However, as was the case for the document of 27 January 2003, that email, which is contemporaneous with the conclusion of the Santa Clara agreement, does not reveal any link between the market shares and the existence, or amount, of ECAP. Even if it were possible to take the view that the reference to ‘aggressive prices’ refers to ECAP, it is apparent only that ECAP forms part of the Santa Clara agreement and is mentioned in the context of the objectives of increasing Intel’s market share in NEC’s purchases of x86 CPUs. It is not expressly stated that ECAP was dependent on NEC meeting those objectives.

355    In the third place, in recital 462 of the contested decision, the Commission refers to an exchange of emails between NEC executives dated 10 May 2002 (‘the NEC email exchange of 10 May 2002’). That exchange describes how NECCI and NEC Japan could reach the levels of MSS requested by Intel and mentions the sums which would be received as MDF.

356    However, the fact remains that that evidence, which predates the conclusion of the Santa Clara agreement and which forms part of the negotiation of that agreement, makes no mention at all of ECAP, as Intel correctly points out, with the result that it cannot support the Commission’s finding regarding the conditionality of ECAP. On the contrary, it appears that that email exchange corroborates Intel’s argument that MDF was the only rebate dependent on Intel’s market share of NEC’s purchases. From the wording of that exchange, it is apparent that NECCI and NEC were to reduce the market share of AMD in their purchases and would receive a certain amount as MDF. It therefore appears that MDF is a consequence of reductions in the market share of AMD and is the only benefit which is directly dependent on the respective levels of both AMD’s and Intel’s market shares within NEC’s purchases.

357    In the fourth place, in recital 464 of the contested decision, the Commission mentions NECCI’s response to question 14 of the 2005 information request made under Article 18 of Regulation No 1/2003 (‘the 2005 information request’). The Commission maintains that it is apparent from that response that ECAP was dependent on the levels of MSS. Intel claims, however, in essence, that what is involved is a reference to ECAP in force after the fourth quarter of 2002.

358    In that regard, it must be pointed out that it is stated at the beginning of the second paragraph of that response that ECAP prices are dependent on an agreement on levels of MSS and not on volumes.

359    As has already been observed in paragraph 967 of the initial judgment, NECCI’s responses provided under Article 18 of Regulation No 1/2003 are particularly reliable evidence, since, first, NECCI would appear to have had no interest in providing incorrect information which could be used by the Commission to establish an infringement of Article 102 TFEU committed by Intel, its unavoidable trading partner, and, second, fines may be imposed under Article 23(1)(a) of Regulation No 1/2003 if the information is incorrect.

360    Nevertheless, taken in context, the Court does not consider that that response can be used as evidence to support the Commission’s findings.

361    First, the 2005 information request was organised so that each reference to a document was followed by one or more questions relating to that document. As the Commission confirmed at the 2020 hearing following a question put by the Court, the response to question 14 concerns the document entitled ‘JH 210’. That refers to a statement made by a seller of NECCI on 22 February 2005. Accordingly, the document entitled ‘JH 210’ is subsequent to that date and, consequently, to the fourth quarter of 2002 and the date on which Intel’s scheme for rebates to NEC was modified, namely 1 July 2003. Consequently, NECCI’s reply to question 14 concerns a document and a statement which are not certain to be directly relevant to what the Commission was seeking to demonstrate, since it appears that they concern a period after 1 July 2003, that is to say, a period during which Intel’s payment structure had changed and MDF had been subsumed into the standard ECAP-type discounts and had been renamed ‘super-ECAP’.

362    Second, in the light of that clarification regarding time and context, it is not certain whether, in mentioning ECAP, the response referred, as a general category of rebates granted by Intel, to ‘super-ECAP’ (also referred to as ‘special ECAP’), which existed from 1 July 2003 and replaced MDF, while being subsumed into the general category of ECAP, or to classic ECAP, which was, however, referred to simply as ‘ECAP’ and which existed both in the fourth quarter of 2002 and after the modification of the system of rebates. As is apparent, in particular, from NECCI’s response to question 20 of the 2005 information request, ‘super-ECAP’, like the MDF which it replaced, was conditional on a certain level of MSS, whereas ECAP was not.

363    In the fifth place, in recital 464 of the contested decision, the Commission examined an internal NEC presentation of 15 May 2002, which is contemporaneous with the negotiation of the Santa Clara agreement. The Commission demonstrated that, in exchange for a certain level of MSS, Intel agreed to a dozen payments to NEC, only two of which were MDF.

364    That two-page document presents two MDF payments to NEC approved by Intel and other price levels for different types of x86 CPUs. However, there is no information concerning obligations regarding NEC’s level of MSS. The second page contains a graphic demonstrating the transition from the original plan to the revised plan, and therefore to the realignment plan leading to the Santa Clara agreement, and sets out the respective objectives for Intel’s market share of NEC’s purchases of x86 CPUs. The transition from the original plan to the realignment plan is graphically presented by an arrow between the two plans, in the middle of which the reference ‘$6M MDF’ appears, namely the MDF payment of USD 6 million. Only that payment appears clearly in that document as consideration for the increase in the level of MSS. It follows that, while the document confirms, as is the case for the first two documents analysed, that ECAP was discussed in the negotiations which led to the Santa Clara agreement, it was only MDF which was dependent on the level of MSS.

365    In the sixth place, in its defence, the Commission refers to page 4 of an internal NEC presentation of 15 April 2002, from which it is apparent that, in return for the increase in Intel’s market share within NEC’s purchases, NEC wished to receive, inter alia, ECAP.

366    It is apparent from that presentation that NEC identified three requests to be made to Intel in order to increase its level of MSS within NEC’s purchases. The requests concerned a ‘Marketing & Engineering Fund’ (which is probably MDF), but also ECAP, and the improvement of the contractual framework with Intel.

367    However, the Court considers that that presentation cannot provide a sound basis for the conclusion reached by the Commission.

368    Although the document examined presents ECAP as part of the consideration for adopting the realignment plan, what is involved is a desire on the part of NEC prior to negotiation with Intel and not a presentation of rebates as they were defined in the Santa Clara agreement following its negotiation.

369    In the seventh place, the Commission refers, in its defence, to an NEC presentation dated 6 May 2002 containing another graphic representation of NEC’s possible transition from the original plan to the realignment plan. The transition is graphically illustrated by an arrow with a comment relating to it, stating that it ‘will depend on more than [USD 6 million] MDF’. In addition, another page of that presentation mentions ‘need ECAP to achieve [the realignment plan]’.

370    However, as is the case for the NEC presentation of 15 April 2002, although the document is contemporaneous with the Santa Clara agreement negotiations (meetings of 6 and 7 May 2002), it does not set out the results of those negotiations, but only NEC’s wishes. Consequently, it must be rejected for the same reasons as those set out in paragraph 368 above.

371    It follows from the foregoing considerations that it is apparent from the documents taken into account by the Commission as a whole that the rebates on the prices of x86 CPUs, including ECAP, were discussed and agreed in the context of the Santa Clara agreement negotiations and that NEC wished to obtain concessions on ECAP as consideration for its MSS level commitment. However, only the NEC presentation of 27 January 2003 can constitute evidence supporting the Commission’s position that the ECAP finally agreed in the context of the Santa Clara agreement was paid, at least in part, as consideration for compliance with the obligation relating to a level of MSS noted in the realignment plan. By contrast, the NEC email exchange of 10 May 2002, NECCI’s response to question 20 of the 2005 information request and the internal NEC presentation of 15 May 2002 tend rather to demonstrate that solely the MDF agreed under the Santa Clara agreement was conditional.

372    Consequently, the Court considers that those documents do not contain sufficient evidence or a set of indicia sufficient to confirm the argument concerning the conditionality of ECAP in the fourth quarter of 2002.

(ii) The evidence adduced by Intel

373    At this point, it is necessary to assess the evidential value of the documents put forward by Intel to call into question the Commission’s finding that both MDF and ECAP were conditional.

374    In the first place, as regards the applicant’s argument that, in essence, no relationship of conditionality between ECAP and a level of MSS is apparent from the response to question 32 of the 2005 information request, it should be noted that the Commission specifically asked NECCI to specify the type of benefit, if any, granted to NECCI in exchange for it meeting the MSS level obligation in the realignment plan. However, in its reply, NECCI mentioned only MDF. In that document, which sets out the outcome of the Santa Clara agreement, MDF is presented as the sole consideration for achieving levels of MSS, with the result that solely MDF is conditional.

375    The Commission casts doubt on the evidential value of that document by stating that it is apparent from the documents annexed to that response that ECAP is conditional. The Commission simply refers to confidential annexes 32.1 to 32.4. However, while it is possible to identify the first two annexes, that is not the case for the latter two annexes. As regards Annex 32.1, this corresponds to the NEC email exchange of 10 May 2002, referred to in recital 462 of the contested decision, which was analysed in paragraphs 355 to 356 above and which has been found to support Intel’s argument. Annex 32.2 corresponds to the document analysed in paragraphs 353 and 354 above which was found not to prove that ECAP was conditional.

376    It must therefore be found that, by simply referring to the annexes mentioned above without providing further details, the Commission has failed to dispute the evidential value of NECCI’s response referred to in paragraph 371 above, to which a great evidential value must be attached, since it is an exhaustive response to a direct question provided under Article 18 of Regulation No 1/2003. NECCI’s response to question 32 of the 2005 information request therefore tends to support the argument that solely the MDF agreed in the context of the Santa Clara agreement was conditional and that ECAP was not.

377    In the second place, in question 21 of the 2005 information request, the Commission in particular requested NECCI to explain, first, whether the ECAP rebates offered to it were dependent on NECCI, NEC Japan and NEC complying, at a worldwide level, with certain levels of MSS provided for in the Santa Clara agreement and, second, the consequences of non-compliance with those obligations for a given quarter.

378    In its reply, NECCI explained, first, that ‘special ECAP’, ‘super-ECAP’ or the MDF granted to it were dependent on compliance with specific levels of MSS both by it and by NEC Japan and NEC at a worldwide level. However, unlike ‘special ECAP’ or ‘super-ECAP’, ECAP is not dependent on a certain level of MSS, but is simply the result of commercial negotiations. Second, at the time when global MDF was applied, if NECCI had not complied with its obligation concerning a certain level of MSS for a particular quarter, it would have received no MDF payment. At the time when the response to the 2005 information request was provided, if NECCI had not complied with its MSS obligation for a specific quarter, that would also have compromised the ‘super-ECAP’ negotiations for the following quarters.

379    That response is clear. A great evidential value must be ascribed to an exhaustive answer to a direct question provided under Article 18 of Regulation No 1/2003.

380    In addition, contrary to what the Commission asserts in its defence, NEC confirms, unambiguously, that the only rebates that were conditional on a particular MSS objective were MDF and ‘special ECAP’ or ‘super-ECAP’. By contrast, ECAP was not conditional on that objective and was determined in the context of business relations. Any penalty for failure to comply with the obligation relating to a level of MSS concerned MDF, ‘special ECAP’ or ‘super-ECAP’, and not classic ECAP. Since, as from 1 July 2003, MDF became ‘special ECAP’ or ‘super-ECAP’, the only ECAP which existed during the fourth quarter of 2002 was classic ECAP. Consequently, NECCI’s response to question 21 of the 2005 information request tends to support the argument that classic ECAP was not conditional on a certain level of MSS.

381    In the third place, by question 6 of its 2007 information request made under Article 18 of Regulation No 1/2003 (‘the 2007 information request’), the Commission, in essence, requested that NECCI specify which funds it had received during a period which also covered the fourth quarter of 2002, in exchange for complying with the MSS level obligation.

382    In its response, NECCI mentions, as regards the period between the third quarter of 2002 and the second quarter of 2003, only MDF as being conditional and states that the market share percentage translated into a number of x86 CPUs to be purchased. Contrary to the Commission’s allegations, that response therefore confirms that, during the period concerned, only MDF was dependent on the condition regarding the levels of MSS. As a response provided by NECCI under Article 18 of Regulation No 1/2003, a great evidential value must be ascribed to that item of evidence.

383    In the fourth place, in order to substantiate its claims, Intel referred to NEC’s minutes dated 8 May 2002 of the meeting held with Intel on 6 and 7 May 2002. As the Commission submits, it is apparent from the second page of that document that, in exchange for accepting the realignment plan, NEC wished to obtain not only MDF but also ECAP and a new contractual framework. That corresponds, moreover, to the documents examined in paragraph 365 above. However, it is apparent from the third page of those minutes that the condition for achieving the level of MSS discussed on the second day of the negotiations consisted of allocating a certain sum as MDF. Furthermore, while Intel appears during the second day of the discussions to have accepted part of NEC’s request concerning ‘MDF/ECAP’, since there is a reference on the third page of those minutes that ‘Intel responded with 50% acceptance for total 12 items of NEC’s ECAP/MDF request’, that reference does not make it possible to determine which part of that request was accepted, namely whether it was the part concerning MDF or the part concerning ECAP. That is all the more the case given that, on the fourth page of those minutes, under the heading ‘Next step’, it is stated that ‘Intel reviews with [M and P] for MDF request/ECAP request’, with the result that it is not possible to identify whether part of NEC’s request was ultimately accepted and, if so, which. It must therefore be noted that those minutes are drafted in summary form and that there is uncertainty as to how they are properly to be understood.

384    It must therefore be held that the evidential value of that document is relatively low, given it does not represent the outcome of the negotiations and that its summary nature makes it uncertain how it is properly to be understood.

385    In the fifth place, in the reply, Intel adduces Annexes C.37 and C.38, stating that they are documents preparatory to the Santa Clara agreements which were drawn up by NEC from which it can be seen that, irrespective of the level of MSS reached, NEC expected to receive the same level of ECAP.

386    However, regardless of whether that evidence adduced by the applicant at the stage of the reply is admissible, first, Annex C.37, at pages 5 and 6, contains tables, figures and a reference to ‘ECAP request’, without it being possible to deduce clearly from it a correlation between the development of Intel’s market share and NEC’s expectations regarding ECAP. Second, Intel refers, in the reply, to ‘Annex C.38 on page 10’. However, Annex C.38 consists of a document comprising eight pages and containing a large amount of information, with the result that it is not possible for the Court to determine precisely the elements in the annex which support the applicant’s argument. Consequently, Annex C.38 cannot be taken into consideration, in accordance with the case-law cited in paragraph 314 above.

387    It thus follows from all of the foregoing considerations that the evidence taken into account in the contested decision does not constitute sufficient evidence or a set of indicia sufficient to demonstrate that ECAP-type rebates, or rebates other than MDF, were conditional on the obligation imposed on NEC to reach a certain level of MSS during the fourth quarter of 2002. In addition, the other items of evidence relied on by Intel tend rather to support the argument that only MDF was conditional.

388    From this it follows that the evidence relied on in the contested decision is not reliable, with the result that it is not capable of substantiating the conclusions drawn from it.

389    Consequently, there being no need to examine Intel’s other arguments, it must be held that the Commission erred in its assessment of the value of the conditional rebates granted by Intel to NEC.

(2)    Use of the fourth quarter of 2002 as a reference

390    The applicant criticises the Commission on the ground that it erred in carrying out the AEC test only for the fourth quarter of 2002 and in finding on that basis alone, in recital 1456 of the contested decision, that the payments granted by Intel to NEC under the Santa Clara agreement were capable of foreclosing or likely to foreclose an as-efficient competitor during the entire period from October 2002 to November 2005. In other words, the applicant argues that the Commission erred in taking the view that the fourth quarter of 2002 was representative of all subsequent periods.

391    In general, it submits, the Commission bears the burden of demonstrating that Intel’s practices were capable of foreclosing an as-efficient competitor throughout the reference period, but it had no basis for assuming that all the figures relevant to such a test, such as, for example, gross prices, discounts or volumes, remained unchanged from 2002 to 2005. That is the case, in particular, as regards the contestable share, in respect of which the Commission itself stated, in recital 1243 of the contested decision, that it could increase over time as consumers become increasingly aware of the viability of the AMD alternative.

392    In particular, first, the USD 6 million of MDF linked to market share expectations did not remain in effect after the first quarter of 2003.

393    Second, in its main observations, Intel states, in essence, that, contrary to what is stated in recital 1410 of the contested decision, the application of the AEC test should not be based on an assessment of whether or not the rebate levels recorded for the quarter in question were significantly varied in subsequent periods, but on their actual levels. Comparatively small changes in discount levels can alter the outcome of the analysis. Intel argues that the Commission stated with regard to the rebates granted to NECCI that a new rebate programme had begun in July 2003, but the Commission never examined whether the changes introduced under that programme had affected any AEC test parameter.

394    Third, Intel states, in those observations, that the lack of precision in the Commission’s prospective analysis is also based on the fact that recital 1410 of the contested decision examines only the discounts granted to NECCI, whereas the infringement finding was in relation to the parent company as a whole, and thus in relation to NEC.

395    The Commission submits, in its supplementary observations, that the arguments put forward by Intel in its main observations are inadmissible on the ground that the applicant was disputing for the first time the ground on which the Commission justified the extrapolation in question in the contested decision.

396    As regards the substance, it submits, first, that recital 1410 of the contested decision lists the reasons why that quarter is representative and the documentary evidence on which that decision relies.

397    Second, the argument that MDF payments did not remain in effect beyond the first quarter of 2003 does not take account of the fact that documents in the file prove that those payments did not disappear, but that they were simply subsumed into other categories of rebate. NECCI also explained that the Santa Clara agreement and the associated conditions remained in force until at least November 2005.

398    Third, the document which Intel puts forward in support of its assertions contains no calculation excluding MDF payments.

399    Fourth, if it were true that Intel’s payments to NEC had undergone significant variations during that period, Intel could easily have provided evidence to that effect during the administrative procedure.

400    Fifth, the analysis in recital 1243 of the contested decision concerns Dell. The Commission submits, in essence, that that analysis cannot be transposed to NEC, since NEC did not obtain supplies exclusively from Intel, unlike Dell, with the result that NEC’s customers were already aware of the value of products with an AMD processor and that NEC’s contestable share during the quarter in which the Commission made its comparison was already substantial, since NEC’s initial plan was to make 41.6% of its purchases from AMD.

401    As regards the admissibility of the arguments put forward by Intel in its main observations, it must be stated that Intel maintained, in paragraphs 473 to 475 of the application, that the Commission’s allegation regarding the ability to extrapolate the results of its analysis of the fourth quarter of 2002 until 2005 had no basis. There is no evidence of any stability in gross prices, discounts or volumes. In that regard, Intel refers expressly to paragraphs 454 to 473 of Annex A.8 to the application, which set out its response to the arguments concerning that extrapolation made in the Statement of Objections of 2007. In paragraph 467 of Annex A.8 to the application, Intel submits that the Commission provided no details as to the levels of ‘super-ECAP’. In paragraph 468 of that annex, Intel distinguishes between the rebates received by the parent company NEC and those received by NECCI, stating that certain types of rebate were granted only to NECCI. In particular, Intel denies that the ‘super-ECAP’ nomenclature was used beyond the third quarter of 2003. In paragraphs 470 and 471 of that annex, Intel states that the quantities of x86 CPUs purchased were the subject of constant renegotiation, with the result that nothing was fixed in the data for the fourth quarter of 2002.

402    It follows that, contrary to the Commission’s assertions, in the application the applicant disputed numerous items of evidence regarding the data relating to the reference quarter being extrapolated to the entire period covered by the contested decision. The applicant stated that the difference between the discounts granted to NEC and to NECCI lies in different competitive constraints, from which it follows that it was impossible to make any presumption that the rebates granted were stable as between Intel and its customer. The applicant also states that the Commission has no data regarding the levels of ECAP payments for NEC and that the quantities of x86 CPUs purchased were anything but stable during that period.

403    Consequently, it must be held that the arguments put forward by Intel in its main observations are admissible since they relate to arguments submitted in the context of the application.

404    As regards the assessment of the substance of those arguments, it must be observed that the economic data used by the Commission in recitals 1410 to 1455 of the contested decision to carry out the AEC analysis of the rebates granted by Intel to NEC include, inter alia, the total quantity of x86 CPUs purchased, the net and gross prices of the different types of x86 CPUs, the types and amounts of the rebates granted and Intel’s costs.

405    First, the parameters of the alleged stability on which the Commission bases its ability to extrapolate, which are set out in recital 1410(a) to (c) of the contested decision, concern only the level of rebates (MDF, ECAP and total rebates), the prolongation of those levels during the following quarters and Intel’s AAC, but the Commission conducts no analysis at all, as Intel in essence submits, of the volumes and types of x86 CPU sold or of their net or gross prices.

406    Second, the table mentioned in recital 1410(a) of the contested decision, and which is taken from NECCI’s response to question 9 of the 2007 information request made under Article 18 of Regulation No 1/2003, concerns only the rebates granted to NECCI, whereas the infringement of Article 102 TFEU was established in regard to its parent company NEC. There is nothing to suggest that the rebates granted to NEC Japan or to NEC overall were stable during the entire period concerned by the infringement finding.

407    Third, a review of the table mentioned above demonstrates that, contrary to what is stated in recital 1410(a) of the contested decision, the rebates received by NECCI were not stable during the entire period concerned. There is a difference of 461.3% between the lowest payment, in the second quarter of 2003 (USD 3.3 million) and the highest payment in the third quarter of 2005 (USD 15.224 million). There is also a significant difference between the fourth quarter of 2002 (USD 7.945 million), and the second quarter of 2003 (USD 3.3 million), namely a reduction of 58.4%.

408    Fourth, as NECCI stated in the response to question 9 of the 2007 information request, the system for granting rebates changed as from the third quarter of 2003. Instead of a single sum, ‘super-ECAP’ was included in the quarterly price setting. As Intel submits, there is nothing to indicate that that new system did not involve quantitative differences vis-à-vis the previous system.

409    Fifth, although the Commission submits that the Santa Clara agreement remained in force until 2005, it does not follow that the situation prevailing in the fourth quarter of 2002 under that agreement persisted throughout the period of the alleged infringement. It follows in particular from paragraph 2 of NECCI’s response to question 9 of the 2007 information request that, after 1 July 2003, NECCI requested rebates on a monthly basis, based on the quantities purchased and the price difference between the pricing proposed to customers and the ECAP or ‘super-ECAP’. However, nothing permits the inference that the Commission examined whether the amendments introduced by that new programme had affected one of the parameters of the AEC test.

410    It follows from all of the foregoing that, first, the parameters set out in recital 1410(a) to (c) of the contested decision do not include all the economic data used by the Commission to carry out the AEC analysis of the rebates granted by Intel to NEC and, second, that, contrary to what is apparent from recital 1410(a) and (b) of the contested decision, the evidence placed on the file shows that Intel’s payments to NECCI were subject to significant variations after the fourth quarter of 2002 and that the rebate system changed as from the third quarter of 2003. It follows that Intel is correct in maintaining that the Commission erred in taking the view that, given the parameters set out in recital 1410(a) to (c) of the contested decision, it could rely on data relating to the fourth quarter of 2002 in order to draw conclusions regarding the capability of Intel’s practices to foreclose an as-efficient competitor between the fourth quarter of 2002 and November 2005.

411    Consequently, there being no need to consider the applicant’s arguments referred to in paragraph 340 above, that, first, the Commission’s own data showed that the rebates granted to NEC were not capable of foreclosing an as-efficient competitor and, second, the Commission incorrectly calculated the value of the business at risk for Intel, it must be concluded that, as regards the AEC test carried out for NEC, the Commission made two errors of assessment, first, by using an exaggerated value for the conditional rebates and, second, by extrapolating the results which it obtained for the fourth quarter of 2002 to the entire period of the alleged infringement. It must be held that, in the light of those two errors, the basic parameters of the Commission’s assessment of the AEC test are incorrect. Since those errors relate to the fourth quarter of 2002, which was taken as a reference for the entire period under consideration for the rebates granted to NEC, they affect the whole of the period examined in the contested decision as regards NEC. It follows that the Commission has not established to the requisite legal standard the validity of the conclusion set out in recital 1456 of the contested decision, according to which the payments granted by Intel to NEC under the Santa Clara agreement were capable of foreclosing or likely to foreclose an as-efficient competitor.

(d)    The alleged errors in the AEC test applied to Lenovo

(1)    General overview of the part of the contested decision devoted to Lenovo

412    The Commission conducted the AEC test for Lenovo in recitals 1457 to 1508 of the contested decision. The Commission analysed, first of all, the extent and nature of the rebates, on the basis of the 2007 MoU.

413    The Commission then calculated the ASP, the costs and the required number of x86 CPU units.

414    Finally, the Commission assessed the contestable number of x86 CPU units. In its principal calculation, it limited its assessment to the notebook segment (see recitals 1473 to 1478 of the contested decision), whereas, in its alternative calculations, it responded to Intel’s assertions that the contestable number of x86 CPU units should also include the desktop segment (see recitals 1479 to 1508 of the contested decision). Those alternative calculations are subdivided, first, into a response from the Commission to Intel’s allegations regarding the overall contestable number of x86 CPU units and, second, into a confirmatory calculation, made on the basis of a comparison with data from a document entitled ‘Statement of Work April 2006’, adopted pursuant to an agreement between AMD and Lenovo.

(2)    The conditional portion of the rebates

415    In recital 1461 of the contested decision, the Commission stated that the amount of the rebates in question was set out in the 2007 MoU, which provided for a financial support of USD 180 million for 2007, in the form of quarterly payments.

416    In recital 1462 of the contested decision, it was stated that the payments under the 2007 MoU were incremental to payments which Intel continued to make under other previously agreed financial support programmes, separately from the 2007 MoU. The Commission therefore took the view that they had to be attributed entirely to the outcome of the agreement on the 2007 MoU. All payments and favourable commercial terms provided for in the 2007 MoU were conditional on Lenovo cancelling its plans for notebook computers with AMD-based x86 CPUs.

417    In recital 1463 of the contested decision, it was stated that, in the submission of 5 February 2009, Intel had put forward the argument that the relevant value for the size of the rebates was only USD 138 million. That is explained by the fact that, out of the financial support for Lenovo of USD 180 million provided for in the 2007 MoU, only USD 135 million were awarded in the form of cash. The remainder of the financial support was awarded in the form of non-cash advantages, namely the extension of Intel’s standard one-year warranty and the offer of improved use of an Intel hub facility in China. The Commission stated that Intel had submitted that, whilst the value of those two non-cash contributions for Lenovo was USD 20 million and USD 24 million respectively, their cost for Intel was much lower, namely USD 1.7 million and USD 1.3 million respectively. Intel had submitted that, for the purposes of the as-efficient-competitor analysis, those elements had to be assessed not by reference to their value for Lenovo, but by reference to their economic cost to Intel. Intel reached the amount of USD 138 million by adding the USD 1.7 million and USD 1.3 million costs to the USD 135 million cash funding.

418    In recital 1464 of the contested decision, the Commission stated that, before discussing the validity of Intel’s argument on the appropriate measure to be used in the as-efficient-competitor analysis, it had noted the disparity between the alleged economic cost of the contributions for Intel and their value to Lenovo. The ratio between the value to Lenovo and the alleged economic costs for Intel was 1176% (20:1.7) for the warranty extension and 1846% (24:1.3) for the hub facility. The Commission stated that Intel had provided certain calculations carried out for the purpose of the 5 February 2009 submission related to the Supplementary Statement of Objections of 2008 in support of its assertion regarding the economic cost of the contributions, but that it had failed to explain the reason for the stark discrepancy between those costs and their value to Lenovo.

419    In recital 1465 of the contested decision, the Commission noted that, without prejudice to the observation mentioned above, Intel’s argument that the appropriate measure to be used in the as-efficient-competitor analysis was not the value of such items to Lenovo but their economic cost for Intel was based on a misunderstanding of the principles of that analysis.

420    In that regard, in recital 1466 of the contested decision, the Commission stated that ‘the as-efficient-competitor analysis assesses the price at which a competitor which is as efficient as the dominant company – but which is not dominant – would have to offer its products in order to compensate the customer for the loss of the conditional benefits granted by the dominant company and which would result from that customer’s switching the contestable share of its supply needs away from the dominant company to the hypothetical as-efficient competitor’.

421    Lastly, in recital 1467 of the contested decision, the Commission stated that it was clear from the foregoing that the relevant measure was the loss for the customer, as that was the loss which the as-efficient competitor would have to compensate, and not the economic costs to the dominant company, in the event that the two figures diverged. That difference was illustrated, according to the Commission, by the case of the supply hub. As a dominant undertaking, Intel already had a supply hub in China, in respect of which it claimed that it needed to make only marginal improvements, with an economic value of USD 1.3 million, in order to be in a position to offer Lenovo a full benefit value of USD 24 million. However, the Commission stated that a competitor as efficient as the dominant company, but which was not dominant, would normally not have put in place such a facility at that stage. In order to compensate Lenovo for the loss of the benefit linked to improved use of Intel’s supply hub, the as-efficient competitor would therefore have to have awarded Lenovo a monetary payment equivalent to the economic value of the improved supply hub facility for Lenovo.

422    The applicant makes a general assertion that the 2007 MoU did not support the conclusion that its rebates would have resulted in it foreclosing an as-efficient competitor. The Commission’s analysis, first, overstates the allegedly conditional portion of the discount, second, it understates the contestable share and, third, it overstates Intel’s costs. As regards, more specifically, the conditional portion, the applicant states that, so far as the rebates under the 2007 MoU are concerned, the contested decision concludes in recitals 1461 and 1474 to 1477 that conditional rebates of USD 180 million were granted for a contestable share of only 0.9 to 1.1 million notebook computers. However, according to the applicant, the amount of conditional rebates was only USD 138 million.

423    According to the applicant, the Commission’s methodology for taking account of those non-cash advantages is incorrect, on the ground that, for the purposes of the AEC analysis, the conditional rebate ought to take account of the costs incurred by the applicant in providing such rebates, and not of the value which they represent to Lenovo. The supplemental Shapiro-Hayes report of 28 January 2009 (‘the Supplemental Shapiro-Hayes report’) calculated the cost to Intel of the two non-cash advantages at approximately USD 3 million. Of the USD 3 million, USD 1 680 073 (rounded to USD 1.7 million) correspond to the warranty extension and another USD 1 256 948 (rounded to USD 1.3 million) to Intel’s cost of offering Lenovo a supply hub.

424    The Commission disputes all of the applicant’s arguments. The Commission submits that, in order to assess the conditional portion of the rebates, the contested decision assessed the incremental funding awarded by Intel to Lenovo in 2007 under the 2007 MoU. That approach is not disputed. According to the Commission, the decision established that that incremental funding amounted to USD 180 million, based on the examination of slides prepared by Intel during the negotiation of the 2007 MoU. The Commission’s view is that it was correct to rely on the value of the non-cash advantages granted to Lenovo, instead of taking as a basis the cost for Intel of those advantages. According to the Commission, the as-efficient-competitor analysis involves an assessment, in essence, of the compensation which a hypothetical as-efficient competitor would have to offer to Lenovo for the loss of Intel’s rebates. In order to be incentivised to choose the as-efficient competitor, Lenovo would expect to be compensated for its own losses, and not for Intel’s losses.

425    Moreover, the Commission submits, relying on Annex B.31 to the defence, that Intel has adduced no sound evidence that there was a divergence of views between Lenovo and it on the value of the non-cash advantages granted to Lenovo, nor, a fortiori, that Lenovo calculated a different value for those non-cash advantages. It also states that documents contemporaneous with the facts set out in the file prove that Lenovo considered those non-cash advantages to be a significant help for it and that Lenovo had requested them from Intel as from the beginning of the negotiations.

426    According to the Commission, it is incorrect to assert, as the applicant does, that the value of the non-cash advantage for the purposes of the as-efficient-competitor analysis is the cost to the dominant undertaking of those advantages. The reply attempts, according to the Commission, to circumvent the error which vitiates the reasoning in the application, claiming that ‘by definition, an as-efficient competitor could provide the same non-cash benefit to Lenovo at the same cost as Intel’. That ignores the fact that the as-efficient competitor is smaller than Intel. The Commission refers to recital 1467 of the contested decision, which explained that the as-efficient competitor would normally not have a supply hub in China at that stage. It would therefore, according to the Commission, have to provide cash compensation for the loss of the advantages granted by Intel to Lenovo.

427    On that point, the Commission contends that the reply merely responds, first, by stating that an as-efficient competitor would necessarily have a supply hub in China and, second, by alleging that AMD had one. In that regard, the Commission considers that the first response provided in the reply is a mere assertion. There is, according to the Commission, no reason why a competitor, even one who is as efficient, would necessarily have a supply hub in China. As regards the second response provided in the reply, the as-efficient-competitor analysis is concerned, according to the Commission, with a hypothetical competitor, not AMD. In any event, the document to which Intel refers merely states that AMD had ‘facilities’ in China, which does not demonstrate the existence of a supply hub, let alone one which is equivalent to that of Intel.

428    According to the Commission, paragraphs 22 to 37 of Annex D.39 to the rejoinder show that, even if it were accepted that the as-efficient competitor had a supply hub in China, its cost in making that hub available to Lenovo would be significantly higher than the cost for Intel of providing that advantage. The same is true of the warranty extension. If, as Intel claims, the cost of the two non-cash advantages amounts to USD 3 million for Intel, then the provision of the same benefits to Lenovo would cost an as-efficient competitor at least USD 38 million. That amount is calculated using two of Intel’s assumptions which the Commission disputes, namely, first, that the as-efficient competitor would have a supply hub in China and, second, that Intel’s cost in awarding the non-cash advantages was USD 3 million.

429    According to the Commission, in any event, the key argument in the application and in the reply that the cost of the two non-cash advantages for Intel is USD 3 million misrepresents Intel’s own evidence. Paragraphs 38 to 44 of Annex D.39 to the rejoinder demonstrate, according to the Commission, that contemporaneous internal Intel documents show that the applicant calculated that, in reality, the cost of the two non-cash advantages would be equal to, or even higher than, their value to Lenovo. Their cumulated cost for Intel was found to be USD 47 million, and not USD 3 million as Intel alleges.

430    Before considering the parties’ arguments regarding the two non-cash advantages, it should be noted that the applicant does not deny having mentioned the figures of USD 20 million for the warranty extension and USD 24 million for the supply hub, in a presentation which it prepared for Lenovo. The applicant claims, however, that those figures ought to be replaced, for the purposes of the AEC test, by USD 1.7 million and USD 1.3 million for each of them, in order to reflect its costs and not the benefit to Lenovo. The Commission attributed USD 44 million of the USD 180 million of conditional discounts to the non-cash advantages on the basis of the value which those services represented for Lenovo. From a reading of recital 1465 of the contested decision, it is apparent that it cannot be accepted that the Commission took into consideration, in that decision, Intel’s calculations estimating its costs of offering the USD 3 million of non-cash advantages or that it analysed that figure.

431    The Commission’s approach consists, in essence, of taking the view that, even if it is accepted that an as-efficient competitor may, in principle, offer non-cash advantages, the fact remains that making available a supply hub or a warranty extension is more expensive for the competitor than the dominant undertaking, in particular where the value of the non-cash advantages is compared with the contestable share. The Commission also maintains that Intel adduced no sound evidence that there was a divergence of views between Lenovo and Intel on the value of the non-cash advantages granted.

432    The applicant criticises that analysis by the Commission. According to the applicant, the Supplemental Shapiro-Hayes report and the Salop-Hayes report demonstrate that that methodology is incorrect and that an appropriate analysis of the as-efficient competitor takes into account the cost to Intel of providing those non-cash advantages. The applicant refers to the Salop-Hayes report, which states as follows:

‘For the as-efficient-competitor test, the conditional discount would include the cost to Intel of providing those benefits, not the value to Lenovo. The focus of the as-efficient-competitor test is to determine whether Intel’s incremental revenues on the contestable share exceed its incremental costs of providing that volume, taking into account the reduction in Intel’s profits from the conditional discounts. The reduction in Intel’s profits involves the cost to Intel of the non-cash benefits.’

433    In that regard, the foundation of the AEC test applied by the Commission in the present case is set out, in particular, in recitals 1003 and 1004 of the contested decision.

434    In recital 1003 of the contested decision, the Commission explains the logic inherent in the AEC test, stating that ‘in essence, this examines whether Intel itself, in view of its own costs and the effect of the rebate, would be able to enter the market at a more limited scale without incurring losses’.

435    In recital 1004 of the contested decision, the Commission states that the as-efficient-competitor analysis is a purely hypothetical exercise in the sense that it attempts to analyse whether a competitor which is as efficient as Intel, in terms of producing and delivering x86 CPUs that provide the same value to customers as Intel, but which would not have as broad a sales base as Intel, would be foreclosed from entering. That analysis is, in principle, independent of whether or not AMD was actually able to enter.

436    It follows from the foregoing that the hypothetical competitor whose ability to enter the market is assessed notwithstanding Intel’s pricing practices is an as-efficient competitor, namely an operator capable of supplying x86 CPUs under the same conditions as Intel. As is apparent from recital 1003 of the contested decision, the AEC test amounts, in essence, to examining whether Intel itself could have entered the market despite the system of rebates at issue. It is apparent from recital 1004 of that decision that, in principle, the only difference between the situation of the hypothetical competitor and the actual situation of Intel on the market is that that hypothetical competitor does not have an equivalent sales base. In the light of the clarifications provided in recital 1005 of the contested decision, that reference to the lack of equivalent sales base must be interpreted as meaning that, due to Intel’s status as an unavoidable trading partner, the hypothetical as-efficient competitor is likely to win from Intel only the contestable share of customer requirements for x86 CPUs.

437    As the applicant correctly observes, when the Commission assessed, in the contested decision, the value of the non-cash advantages offered by the applicant in the context of examining the breadth of the rebates granted to Lenovo, its reasoning was not conducted on the basis of the hypothetical competitor being capable of selling x86 CPUs to Lenovo while also offering non-cash advantages to Lenovo on the same terms as Intel.

438    In recital 1466 of the contested decision, the Commission expressed the view that it was a question of assessing the price that an as-efficient competitor, which is not the dominant company, would itself have had to pay to Lenovo to compensate for the loss of the non-cash advantages offered by Intel to Lenovo, such as an extension to the hub or a warranty extension. In recital 1467 of that decision, in order to justify that approach, the Commission relied on the example of the supply hub. The Commission found that, unlike Intel, which had a supply hub in China, to which certain adaptations could be made in order to offer a non-cash advantage to Lenovo, a competitor as efficient as the dominant undertaking, but which was not dominant and which was therefore of smaller scale, would not normally have had such a facility at that stage.

439    Thus, the Commission proceeded on the basis of an assumption which was contrary to the foundation of the AEC test set out in recitals 1003 and 1004 of the contested decision, which is based on the principle that the hypothetical competitor is as efficient as Intel, in particular from the perspective of the costs of extending a hub or a warranty. The Commission in fact conducted its reasoning by reference to a less efficient competitor, which is not, however, the relevant economic operator for assessing whether the rebate at issue is capable of having a foreclosure effect.

440    None of the arguments put forward by the Commission is such as to invalidate that finding.

441    The Commission – which, admittedly, refers to the fact that Intel had evaluated the amount of benefit for Lenovo at a high value (USD 20 million and USD 24 million respectively) – provides no response, in the contested decision, to the question of what the cost would have been for an as-efficient competitor if it had had to provide access to a supply hub or simply modify its own existing hub so as to extend it to an OEM, as Intel offered to Lenovo. The same logic applies to costs associated with a warranty extension.

442    In that regard, the parties stated, in response to the questions put by the Court at the 2020 hearing, that economies of scale ought not to be taken into consideration as a differentiating factor, but that the costs of an as-efficient competitor ought to be regarded as being the same as for Intel. However, such explanations on the part of the Commission contradict the approach adopted in recitals 1466 and 1467 of the contested decision, which take account of the size of the as-efficient competitor to emphasise, inter alia, that a hub comparable to that of Intel would not be in place at that stage.

443    In addition, in so far as the Commission referred, before the Court, to the actual size of an as-efficient competitor’s hub (see paragraph 426 above, in fine), it must be stated, as Intel submitted at the 2020 hearing, that that factor was not analysed in the contested decision. The same applies to the quantified assessments, submitted by the Commission for the first time in Annex D.39 to the rejoinder, aiming to evaluate the actual costs to Intel in relation to the non-cash advantages (see paragraphs 429 and 430 above).

444    The Court, however, cannot take account of those additional analyses, which were submitted during the procedure before it to substantiate the AEC test in the contested decision without substituting its own reasoning for that of the Commission set out in that decision. The case-law cited in paragraph 150 above prohibits the Court from making such a substitution.

445    As regards the Commission’s assertions in recital 1464 of the contested decision, concerning the alleged stark discrepancy between the economic costs put forward by Intel for the grant of non-cash advantages and their value to Lenovo, it must be noted that, irrespective of the fact that the value to Lenovo is not decisive for the AEC test analysis, as is apparent from the minutes of a statement of 2 June 2009 by L10, [confidential], Lenovo did not accept that the negotiations with Intel concerned an exact value for the non-cash advantages. L10 considered, in essence, that the approach calculated in USD in relation to those advantages could be radically different from the sum presented by Intel. In his view, in essence, that undertaking attempted to give itself credit for factors in respect of which it did not calculate a monetary value, such as distribution via a hub. Intel attempted to persuade it that those factors had an economic value, even though they were, rather, an operational benefit. L10 stated that he had not given any credit from a monetary perspective to those non-cash advantages. Lastly, as regards the Commission’s reference to the L10 email of 12 January 2006, which acknowledged the importance of the non-cash advantages, it should be noted that they are not quantified in dollars in that email.

446    It is also apparent from a series of emails dated 26 November 2006 to 28 November 2008, entitled ‘RE: Intel Meet Comp Response November 27 06.ppt’ that Intel’s negotiating tactics made various references to exaggerated advantages, in particular by positioning elements which it intended to provide to the business partner in any event as an advantage. In those circumstances, the Commission cannot draw the inference, even if only implicitly, as is the case in recital 1464 of the contested decision, from those data alone concerning the negotiations on non-cash advantages, that the actual costs, as stated by Intel, were minimised. Similarly, it is necessary to reject as ineffective the Commission’s claim, referred to in paragraph 614 of the defence and referring to Annex B.31 to the defence, that Intel had not demonstrated that there were divergences of views between Lenovo and itself regarding the value of the non-cash advantages granted. The question is what costs were necessary in order to offer them, and not Lenovo’s perception of their value.

447    In addition, it is not sufficient to rely, as the Commission did in recital 1464 of the contested decision and, subsequently, in paragraph 614 of the defence, referring to paragraph 416 of Annex B.31 to the defence, on the argument that Intel failed to explain the stark discrepancy between its alleged costs of USD 3 million and the amount of USD 44 million to Lenovo. It was for the Commission to assess, directly in the contested decision and not in calculations submitted for the first time before the Court, what the costs of an as-efficient competitor would have been if it had had to offer to an OEM such as Lenovo non-cash advantages equivalent to those proposed by Intel (see also paragraph 444 above).

448    Furthermore, in so far as the Commission carried out, for the first time before the Court, in paragraph 326 of the rejoinder, referring, by way of illustration, to Annex D.39 to the rejoinder, calculations of the costs for the situation in which account was to be taken of the as-efficient competitor having a supply hub in China, it must be noted that the result reached by the Commission regarding the costs differs, in any event, from the result set out in the contested decision, irrespective of the fact that those calculations were made out of time and do not form part of the reasons for the contested decision, which applied a different test. First, as is apparent from paragraph 36 of Annex D.39 to the rejoinder, the cost to an as-efficient competitor is USD 20 690 000, and not USD 24 million as stated in recital 1463 of the contested decision in relation to the supply hub. Second, as regards the warranty extension, the cost of which for an as-efficient competitor is also calculated for the first time at USD 17 473 664 in paragraph 30 of Annex D.39 to the rejoinder, that differs from the USD 20 million stated in the contested decision.

449    Lastly, the Commission’s assertion in paragraph 327 of the rejoinder, referring to paragraphs 38 to 44 of Annex D.39 to the rejoinder, that the applicant’s key argument that the cost of the two non-cash advantages, which amounted to USD 3 million for Intel, contradicts Intel’s own evidence, cannot succeed.

450    As regards Intel’s internal documents with references D.41 and D.42 annexed to the rejoinder, from which it is apparent, according to the Commission, that Intel estimated the costs of non-cash advantages at USD 47 million rather than at USD 3 million, these were not mentioned in the contested decision and do not therefore form part of the reasons for that decision. On reading recital 1465 of the contested decision, it appears inconceivable that the Commission took them into consideration in its principal analysis as set out in that decision, in so far as it states in the decision that ‘Intel’s argument that the appropriate measure to be used in the as-efficient-competitor analysis is not the value of such items to Lenovo but their economic cost to Intel is based on a misunderstanding of the principles of the analysis’.

451    In any event, even if the Commission’s reference to the documents mentioned in paragraph 450 above had been admissible, it could not have been inferred from those documents that Intel had incorrectly minimised its costs by stating that the two non-cash advantages corresponded to USD 1.7 million and USD 1.3 million respectively. The documents to which the Commission refers relate to a situation in which negotiations with Lenovo were underway and in which Intel sought to demonstrate the significance of its commercial proposals, by presenting them in a favourable light to Lenovo (see also paragraphs 445 and 446 above). A specific analysis of those documents, carried out subject to what has already been established, purely as a precautionary measure, demonstrates that they lack clarity and accordingly do not make it possible to uphold the Commission’s position.

452    Accordingly, first, in the document entitled ‘Intel chart entitled “2006 vs. 2007 Trend”’, the advantage relating to the enlargement of the supply hub falls within the item ‘Incremental 07 Spending’ and a reference including the term ‘billing impact’ is made in that chart. However, the figure of USD 24 million, concerning the hub, is included in the column entitled ‘Contra’ and not in the column entitled ‘Expense’. That indicates that what was involved was Intel’s estimate of the equivalent value of the use of the hub, as explained in paragraph 71 of the Supplemental Shapiro-Hayes report and illustrated in Annex 10 to that report, and not Intel’s costs of such a hub or of modifying it. Similarly, the cost of the warranty extension is calculated in paragraph 70 of the Supplemental Shapiro-Hayes report and in Annex 9 thereto at USD 1.7 million. In those circumstances, it is not necessary to give a ruling on Dr Hayes’ allegations made at the 2020 hearing that, because of the limited number of x86 CPU failures, increasing the warranty from one to three years did not involve significant incremental costs.

453    Second, while the tables set out in Annex D.42 to the rejoinder enable a link to be established between the costs for Intel and the advantages for Lenovo, those tables do not show the overall cost of modifying a supply hub, quantified in the contested decision at USD 24 million. In any event, it cannot be ruled out that the aim of that document may have been to present the proposal in a favourable light during the negotiations with Lenovo.

454    Accordingly, in the light of the errors of assessment made by the Commission, it is not necessary to assess certain additional Intel arguments regarding whether AMD actually had a hub in China, since AMD’s situation was not, in any event, decisive for the AEC test.

455    Consequently, it must be concluded that the Commission erred in making a quantified assessment of the non-cash advantages offered by Intel to Lenovo, by using the amounts of USD 20 million and USD 24 million respectively, from which it estimated the amount of the rebates at USD 180 million. That amount of USD 180 million is therefore itself vitiated by an error.

456    In the light of the foregoing, it must be observed that, in recital 1507 of the contested decision, the Commission stated that the conclusions which it had reached as regards the capability of the rebates granted to Lenovo to have foreclosure effects were based on the comparison between the required number of units and the contestable number of units established in recital 1478 of that decision as well as on the considerations in recitals 1479 to 1506, which set out an alternative test for the required share on the combined segments of desktop computers and notebook computers. However, as is apparent from recitals 1472, 1478 and 1503 to 1506 of the contested decision, in the context of both the comparison mentioned above and the alternative test, the Commission took into account a conditional portion of USD 180 million for its analyses regarding the definition of the required share, for the purposes of comparing the required share with the contestable share of the x86 CPU units. The error in the quantified assessment of the non-cash advantages offered by Intel to Lenovo therefore affected all the component parts of the examination of the rebates granted to that OEM.

457    Consequently, there being no need to assess the substance of Intel’s arguments regarding the contestable number of units to be taken into consideration, it must be found that the Commission has not established to the requisite legal standard the validity of the conclusion set out in recital 1507 of the contested decision that in 2007 Intel’s rebates were capable of having or likely to have anticompetitive foreclosure effects, since even an as-efficient competitor would have been prevented from supplying Lenovo’s notebook x86 CPU requirements.

(e)    The alleged errors in the AEC test applied to MSH

458    The applicant submits that the AEC analysis concerning MSH in the contested decision, apart from the fact that it overstated Intel’s AAC, contains two errors concerning, first, the ‘double conditional rebate’ method (‘the double rebate method’) and, second, the conditional portion of the payments. In the applicant’s view, correcting one of those errors would demonstrate that MSH passed the AEC test.

459    The Court considers it appropriate to examine first the substance of the line of argument aimed at demonstrating that the Commission erred in applying the double rebate method.

460    The applicant disputes, in essence, the relevance of the figures used to apply that method and the inferences drawn from it by the Commission.

461    For its part, the Commission considers that all the arguments must be rejected since no error vitiates the application of the double rebate method.

462    The Commission asserts, first, that, in order to be able to sell computers of a particular brand to MSH, an as-efficient competitor would have to ensure not only that MSH was ready to purchase computers based on its CPUs, but also, and most importantly, that OEMs were ready to manufacture those computers. Therefore, in the Commission’s view, Intel’s practices at different levels of the supply chain could have a cumulative effect.

463    Second, the Commission maintains that, in order to demonstrate that Intel’s payments to MSH were capable of having an anticompetitive foreclosure effect when taken together with Intel’s practice towards an OEM, it suffices to show that capability of effect by reference to a representative example of a conditional payment granted by Intel to one OEM without having to repeat the same exercise for each OEM.

464    Third, the contested decision analyses the cumulation of Intel’s payments to MSH with Intel’s naked restrictions, in particular with respect to AMD-based x86 CPU Lenovo notebooks for the period from June to December 2006.

465    Furthermore, the Commission maintains that Annex B.31 to the defence analyses Intel’s other arguments in detail. The Commission argues, in essence, that that annex established that (i) the contested decision provides proper justification that the rebates granted to NEC for the relevant quarter are representative of the entire relevant period; (ii) NECCI could not plausibly have provided the entire contestable share of MSH; and (iii) the contested decision does not rely on the assumption that 100% of Intel’s rebates to NEC were conditional.

466    In that regard, it must be noted, as the applicant also does, that the Commission initially found, in recital 1565 of the contested decision, that Table 58 in recital 1564 of that decision showed that, using the normal calculation method, Intel did not pass the AEC test for 1997, 1998 and 2000. As the applicant, in essence, submits, the Commission therefore acknowledged, at least implicitly, that, if the normal calculation method were used, the effective price resulting from Intel’s conditional payments to MSH was far above the AAC during the entire period of the alleged infringement, namely from 2002 to 2007.

467    Subsequently, as is apparent from recitals 1561 and 1566 of the contested decision, the Commission nevertheless adapted the AEC test by taking the view that, when Intel provided a conditional rebate to an OEM, an as-efficient competitor would have had to provide two payments: one to ensure that it captured the OEM’s contestable share and the other to ensure that it captured MSH’s contestable share. In taking account of that double rebate, the Commission came to the conclusion, in recital 1568 of the contested decision, that Intel did not pass the AEC test during the entire period in question, except for 2004.

468    It therefore follows from the recitals of the contested decision mentioned above that it is appropriate to proceed on the assumption that Intel passed the AEC test using the normal calculation method and that it was only by taking into consideration the existence of a double rebate that the Commission, using its own figures, was able to prove that Intel’s payments to MSH were capable of causing anticompetitive foreclosure during the entire period in question, except for 2004.

469    As regards the assessment of those facts, it should be noted, as a preliminary point, that the applicant does not contest the double rebate method as such. The applicant acknowledges, in essence, that, in order to be able to sell computers of a particular brand to MSH, an as-efficient competitor had to ensure not only that MSH was ready to purchase computers based on its CPUs, but also, and most importantly, that OEMs were ready to manufacture those computers. Therefore, Intel’s practices at different levels of the supply chain could have had a cumulative effect.

470    By contrast, the applicant disputes the figures used by the Commission to carry out its calculations. As the applicant states, the contested decision calculates the amount of the double rebate by assuming that each OEM which supplied MSH benefited from a conditional rebate equivalent to the total rebates offered to NEC in the fourth quarter of 2002 and that such OEMS would have lost the entirety of that rebate if MSH began selling computers with AMD-based x86 CPUs. On the assumption that 100% of the rebates provided to MSH were conditional, the Commission concluded that, for the whole of the period in question, except 2004, Intel’s rebates would have foreclosed an as-efficient competitor.

471    The Court considers that that analysis contains two flaws, each of which is capable of invalidating the results of the AEC test for MSH based on the discounts granted by Intel to NEC in the fourth quarter of 2002.

472    First, as the applicant submits, the Commission assumes, in recitals 1566 and 1567 of the contested decision, that the rebates granted to NEC are an appropriate proxy for conditional rebates on all Intel-based computers sourced by MSH from any OEM. That assumption is not substantiated in any way.

473    Intel states, without being contradicted by the Commission, that MSH purchased only 4% of its computer requirements from NEC during the period from 2002 to 2007 and that, apart from NEC, the main OEM suppliers of computers to MSH between 2002 and 2007 were Fujitsu, Acer, HP, Compaq, Toshiba and Medion. At the very least, the Commission’s position is necessarily based on the assumption that MSH purchased computers from OEMs other than NEC.

474    However, the Commission does not claim, nor does it demonstrate, that in the segment of computers for private individuals Intel granted conditional rebates to any of the other OEMs from which MSH made its purchases, on conditions comparable to the rebates relating to computers purchased from NEC.

475    It is therefore apparent that the contested decision based its analysis of the dual rebate method on the rebates granted by Intel to NEC over the course of a single quarter which represented only part of MSH’s purchases. Therefore, as the applicant submits, it must be held that the Commission’s assumption that all of MSH’s suppliers were subject to substantial conditional rebates identical to those enjoyed by NEC is unfounded and, in any event, entirely unsubstantiated.

476    That conclusion is, moreover, borne out by the wording of recital 1566 of the contested decision, in which the Commission simply states, in order to illustrate the double rebate method, that the ‘[Supplementary Statement of Objections of 2008] took the example of NEC as [an OEM representative of that situation]’, and by the wording of recital 1567 of that decision, in which the Commission states that ‘section 4.2.3.4 assessed the Intel conditional rebates to NEC in the fourth quarter of 2002 (as that is the only quarter where sufficient data are available for the Commission to perform an analysis of the capability of the rebates to foreclose an as-efficient competitor)’. It therefore follows from those recitals of the contested decision that the Commission appears to have relied on the example of NEC and on one single quarter not simply because of its relevance, but because it was the only quarter for which it had been able to obtain information in order to carry out the AEC analysis for MSH.

477    The Commission claims, in that regard, that it is sufficient to refer to a single representative example, since the AEC test is used solely to demonstrate the anticompetitive capability – and not the actual effects – of a commercial practice. However, the Court considers that, when the Commission chooses a quantitative approach to demonstrate that capability, it must ensure that the data used are reliable and must at least explain how such data may be extrapolated. The Commission, however, has in no way proven that NEC’s figures were ‘representative’ for all OEMs.

478    Second, and in any event, as the applicant submits, the Commission’s analysis assumes that NEC and all the other OEM suppliers of MSH benefited, between 1997 and 2007, from conditional rebates identical to the rebate received by NEC for a single quarter. That implies, therefore, that, even if they were representative for all OEMs, the rebates granted to NEC for the fourth quarter of 2002 were stable over a 10-year period. However, first, the Commission has in no way proven that that was the case. The only justification upon which the Commission appears to rely is that set out in recital 1567 of the contested decision, according to which the only data available to it were data concerning NEC’s rebates in the fourth quarter of 2002. However, as the applicant points out, the fact of not being able to obtain additional evidence does not permit the Commission to found its conclusions on assumed facts. Second, it must be recalled that, as is apparent from paragraphs 404 to 411 above, it has been demonstrated that, as regards NEC, the Commission made an error of assessment in extrapolating the results which it had reached for the fourth quarter of 2002 for the whole of the alleged infringement period.

479    Therefore, without there being any need to give a ruling on the other arguments put forward by the parties as referred to in paragraphs 458 to 465 above, it must be held that the Commission erred in finding that Intel’s conditional rebates to NEC in the fourth quarter of 2002 constituted sufficient data to carry out the AEC test for MSH over the entire infringement period.

480    Since the Commission has not demonstrated that the conditions for extrapolation were satisfied, it must therefore be held, without there being any need to give a ruling on the second argument concerning the conditional portion of the payments (see paragraph 458 above), that the applicant is justified in maintaining that the application of the AEC test concerning MSH is vitiated by an error of assessment which applies to the entire period examined.

481    In the light of the foregoing, the Commission has not established to the requisite legal standard the validity of the conclusion, set out in recital 1573 of the contested decision, that, on the basis of the considerations set out in recitals 1559 to 1572 of that decision, during the period from the last quarter of 1997 to 12 February 2008, the Intel payments to MSH were capable of having or likely to have anticompetitive foreclosure effects, either in themselves or as a reinforcing factor of Intel’s conduct vis-à-vis other actors on the market, since even an as-efficient competitor would have been prevented from entering the relevant part of the market.

(f)    Conclusions on the AEC test

482    In the light of all the considerations set out in paragraphs 179 to 480 above, there being no need at all to assess the applicant’s various claims concerning the cost analysis, the Court must accept the applicant’s argument that the AEC analysis carried out by the Commission in the contested decision is vitiated by errors.

C.      The argument that the contested decision did not properly analyse and take account of the criteria referred to in paragraph 139 of the judgment on the appeal

483    According to the applicant and ACT, the Commission’s findings in the contested decision regarding the foreclosure capability of Intel’s rebates do not take proper account of all the criteria set by the Court of Justice in paragraph 139 of the judgment on the appeal. The fact of failing to take account of even one of those criteria should, they argue, result in the General Court annulling the contested decision.

484    The applicant and ACT maintain that, of those five criteria, at least three were not examined appropriately. They submit that, although the contested decision contains an analysis of the first and third criteria referred to in paragraph 139 of the judgment on the appeal, namely the extent of Intel’s dominant position on the relevant market and the conditions and arrangements for granting Intel’s rebates, it fails, in any event, to analyse the criteria relating to the share of the market covered, the duration and amount of the rebates or the existence of a strategy designed to exclude from the market competitors that are at least as efficient as the dominant undertaking.

1.      The coverage rate

485    In recital 1577 of the contested decision, in Section 4.2.4 which relates to the strategic importance of the OEMs which benefited from Intel’s rebates, the Commission stated, in essence, that, because of their market share, their strong presence in the more profitable segment of the market and their ability to legitimise a new processor in the market, some OEMs, in the present case Dell and HP, were strategically more important than others in providing x86 CPU manufacturers with access to the market. The Commission also took the view, in recital 1597 of the contested decision, that the OEMs targeted by Intel’s conduct held a significant part of the market and that, moreover, they were strategically more important than other OEMs, which had had a more significant impact on the overall market than would have corresponded to their aggregate market share alone. The Commission concluded that the coverage of the abusive practices had to be regarded as significant.

486    The applicant and ACT submit, in essence, that, by simply stating in recital 1597 of the contested decision that the OEMs targeted by Intel’s conduct held a significant part of the market and that they were strategically the most important, which had a more significant impact on the overall market than would have corresponded to their aggregate market share alone, the contested decision did not properly take account of the criterion of the coverage rate for the purpose of analysing whether Intel’s rebates and payments were capable of leading to foreclosure.

487    Furthermore, the applicant states that that finding in the contested decision was set out following the conclusion, in recital 1001 of that decision, that Intel’s rebates and payments fulfilled the criteria to be regarded as abusive, whereas the judgment on the appeal requires that the Commission analyse the market coverage before making a finding of abuse. In addition, the applicant and ACT argue that the evidence relied on by the Commission was not sufficient for the view to be taken that the market share covered by Intel’s conduct was significant.

488    The Commission disputes the substance of the arguments put forward by the applicant.

489    First, the Commission observes that market coverage was considered in Section 4.2.4, of the contested decision, in the context of the strategic importance of the OEMs which received rebates from Intel. The Commission dwells in particular on the fact that, although paragraph 139 of the judgment on the appeal simply identifies market coverage as a factor, this must be applied in the context of each case and, in the present case, the strategic importance of the part of the market covered is to be taken into consideration in assessing that factor as demonstrating the capability of Intel’s loyalty rebates to foreclose competition. Similarly, it is necessary to take account of the fact that Intel was an unavoidable trading partner for the OEMs, which gave Intel significant leverage over its customers, since it would have been unrealistic for them to switch to an all AMD or majority AMD product line-up.

490    Second, the Commission maintains that, as regards the share of the market covered, the applicant no longer relies on the assertion, set out in paragraph 115 of the application, that the market coverage of Intel’s practices did not exceed 2% in any year, but appears to accept the fact that the Court found, in paragraph 194 of the initial judgment, that market coverage was approximately 14% on average during the period in which the infringement was committed and submits that it could be inferred from certain evidence that the market shares of the OEMs concerned by the rebates at issue exceeded 25%.

491    Third, as regards the complaint raised by Intel in its observations that the finding in recital 1597 of the contested decision was set out after the conclusion had been reached, in recital 1001 of that decision, that those rebates and payments fulfilled the criteria to be regarded as abusive (see paragraph 487 above), the Commission takes the view that the applicant mischaracterises the contested decision. Recital 1001 of that decision relies on the judgment of 13 February 1979, Hoffmann-La Roche v Commission (85/76, EU:C:1979:36), according to which loyalty rebates infringe Article 102 TFEU. However, as set out in recital 1597 of the contested decision, the Commission states that the subsequent analysis shows that targeting such strategically important OEMs has a more significant impact on the overall market than would correspond to their aggregate market share alone. Consequently, the coverage of the abusive practices has to be regarded as ‘significant’ and recital 1616 of the contested decision reaches the overall conclusion that the loyalty induced by the rebates had complementary effects which significantly diminished the ability of other competitors to compete and sell their products on the merits of their x86 CPUs.

492    It should be recalled that it is apparent from paragraph 139 of the judgment on the appeal that the share of the market covered by the contested practice is one of the criteria which the Commission must take into account for the purposes of assessing the foreclosure capability of rebates and conditional payments (see paragraphs 119 and 125 above).

493    In the first place, in the circumstances of the present case, it cannot be ruled out that Section 4.2.4 of the contested decision, concerning the strategic importance of the OEMs which benefited from Intel’s rebates, might be relevant in examining the coverage rate. That section deals with certain factors which are a priori relevant to examining the foreclosure capability of a system of rebates, such as targeting certain pricing practices at the most profitable segments of the market or using, to the detriment of a competitor, the largest market operators’ power to legitimise a product.

494    The fact remains that, contrary to the Commission’s assertions, and regardless of whether the finding in recital 1597 of the contested decision was set out after the conclusion had been reached in recital 1001 of that decision that Intel’s rebates and payments fulfilled the criteria to be regarded as abusive, the content of Section 4.2.4 of the contested decision, concerning the strategic importance of OEMs which benefited from Intel’s rebates, and in particular recital 1597 of that decision upon which the Commission relies in order to find that the share of the market covered had been examined, cannot be interpreted as constituting in itself sufficient examination, in the circumstances of the present case, of the share of the market covered by the contested practice, within the meaning of paragraph 139 of the judgment on the appeal.

495    Irrespective of the fact that the Commission relied on the market shares of certain OEMs and on the assumption that the Commission could legitimately confine itself to relying on the market shares of certain OEMs rather than examining the share of the market covered by the contested practice, as mentioned in paragraph 139 of the judgment on the appeal, recitals 1578 to 1580 of the contested decision take account solely of Dell’s and HP’s market shares, and exclude the other OEMs concerned by the contested practice, as pointed out by the applicant and ACT. It should be added that the market shares taken into account cover only the period from the first quarter of 2003 to the final quarter of 2005. Therefore, not only does that decision cover only one part of the entire period covered by that decision, that is to say, from October 2002 to December 2007, but it also disregards the period from 2006 to 2007, during which Lenovo and MSH were concerned. Lastly, as the applicant and ACT point out, it is apparent from recitals 1578 to 1580 of the contested decision that the figures for market shares relied on by the Commission take account of Dell’s and HP’s worldwide market shares in all segments, despite the fact that the sole contested practice as regards HP relates to corporate desktops, as stated in Article 1(b) of the contested decision.

496    In the second place, in its main observations, the Commission relies on the Court’s finding, in paragraph 194 of the initial judgment, that the market coverage was approximately 14% on average during the period of the infringement and submits that it could be inferred from certain evidence that the market shares of the OEMs concerned by the rebates at issue exceeded 25%. The Commission also states that ‘Intel’s complaint … that the Commission relied on HP’s market share across all segments is unfounded; the [contested] decision does not rely on any specific figure as regards HP and the [initial] judgment’s reliance on the average coverage of 14%, contrary to Intel’s assertion … does not take account of the segment-specific loyalty rebate with HP’.

497    However, it is necessary to reject the Commission’s argument based on the fact that the Court found, in paragraph 194 of the initial judgment, that market coverage was approximately 14% on average during the period in which the infringement was committed, which the applicant did not dispute, or that the market shares of the OEMs concerned by the rebates exceeded 25%.

498    It must be stated that the rates of 14% or 25% do not appear anywhere in the contested decision subsequent to an examination of the coverage rate. Consequently, for the purposes of reviewing the legality of the contested decision, as regards the share of the market covered by the contested practice, the Court is not in a position to place reliance on those rates, even if they are derived from evidence in the file, since they were not included in the contested decision and, by definition, the Commission could not have acted on the basis of that evidence.

499    Consequently, there being no need to give a ruling on the Commission’s arguments regarding HP’s market share, it must be held that the Commission failed to determine the share of the market covered by the practice at issue, contrary to the requirement placed on it pursuant to paragraph 139 of the judgment on the appeal. It should be added that that is, moreover, contrary to the Commission’s own guidelines on the analysis of cases falling within the scope of Article 102 TFEU, and in particular contrary to paragraph 20 of the Guidance on the Commission’s enforcement priorities in applying Article [102 TFEU] to abusive exclusionary conduct by dominant undertakings (OJ 2009 C 45, p. 7).

500    In the light of the foregoing, it must therefore be concluded that the applicant and ACT are correct in maintaining that the contested decision is vitiated by errors in that it did not consider properly the criterion relating to the share of the market covered by the contested practice.

2.      Duration and amount of the rebates

501    In the application and in its main observations, Intel criticises the failure to analyse, in the contested decision, the importance, first, of the duration of the conditional rebates and payments offered and, second, of the amounts of those rebates and payments. Intel maintains, in particular, that it is not possible, in order to assess the foreclosure effects of the rebates at issue on an as-efficient competitor, to accumulate short-term agreements concluded with the OEMs and MSH. In its view, in order to do so, it is necessary to take into consideration the duration of each of those agreements.

502    The Commission submits that the conditions for granting the rebates and various payments granted by Intel were analysed, for each OEM, in Section VII.4.2.2 of the contested decision. Those analyses concerned the nature and modus operandi of the exclusivity or quasi-exclusivity conditions to which the payments and rebates were subject, the amounts of the rebates and, lastly, the decisive nature of the conditionality of the payments and rebates for each of the OEMs, and for MSH, when they assessed the possibility of obtaining part of their supplies of x86 CPUs from AMD. In particular, the Commission states in the defence that the brevity of the notice of termination period for certain agreements, including those with HP, did not change the harmful effects on competition. Accordingly, if Intel had terminated the HPA agreements following contravention by HP of its quasi-exclusivity obligation, HP would have lost the rebates over the entire remaining duration of the agreement and, at least potentially, over the duration of the extension of that agreement.

503    In its main observations, the Commission submits, in essence, that the applicant has not challenged the passage in the initial judgment which is relevant in the light of the judgment on the appeal, namely the findings in paragraph 195 of the initial judgment, in which the Court considered the implications of the duration of the rebate agreements for their foreclosure capability. The Commission therefore submits that it is necessary to regard as definitive the findings in paragraph 195 of the initial judgment, to the effect that the duration of Intel’s agreements did not impair their ability to foreclose competition.

504    The Commission, in its main observations, also submits, with regard to paragraph 195 of the initial judgment, that, even if Intel were permitted to renew its challenge to the contested decision by reference to the assessment of the significance of the duration of its agreements, there is no reason to depart from the conclusion reached in the initial judgment. First, according to the Commission, if, as found in the contested decision, Intel fails the AEC test, Intel’s insistence on the OEMs being able to withdraw from the loyalty rebate agreements is illogical. An as-efficient competitor simply could not compete. Second, even if Intel were to pass the AEC test, it is inherent to an as-efficient competitor’s bid for the OEMs’ business in such circumstances that it would have to accept a much lower level of profitability on those sales than Intel. Third, the Commission reiterates the allegation that the overall duration of an Intel loyalty rebate scheme would be a factor in how long an as-efficient competitor would have to accept reduced profitability in ‘capturing an OEM’s custom from Intel’ on those sales. Accordingly, for HP, any competitor wishing to displace Intel would have to be prepared to offer terms which would offset the loss of Intel’s rebates over at least the full term of the HPA1 agreement. Moreover, the Commission maintains that each set of agreements with the OEMs ran for a period sufficient for Intel’s actions to be capable of foreclosing competition, since those agreements focused on the most profitable periods for sales of x86 CPU processors, early in the life cycle of a new design. The Commission also asserts that the duration of Intel’s practices cannot be isolated from their timing, since they sought to overcome Intel’s inability to produce a timely technical response to AMD’s 64-bit x86 CPUs.

505    During the 2020 hearing, the Commission submitted a document to the Court concerning the recitals of the contested decision which, in its view, evaluated the various criteria as set out in paragraph 139 of the judgment on the appeal, including duration.

506    In the first place, it is necessary to reject the argument raised by the Commission in its main observations that the complaints regarding the duration and amount of the conditional rebates and payments are inadmissible. It is sufficient to note that the applicant’s arguments put forward in its main and supplementary observations in that regard relate clearly to those put forward in paragraphs 102 and 111 to 114 of the application. Consequently, in accordance with the case-law cited in paragraph 106 above, those complaints are admissible.

507    In the second place, it is apparent from paragraph 139 of the judgment on the appeal that the analysis of the duration and amount of the conditional rebates and payments, which are the object of the contested practice, is one of the criteria which must be taken into account in assessing the foreclosure capability of those practices.

508    First, it is true that, in the contested decision, the Commission examined, on several occasions, factors relating to the duration of the rebates.

509    First of all, recitals 1013 to 1035 of the contested decision concern the time horizon for the AEC test. The Commission found, in particular, in recitals 1015 and 1017 of the contested decision, that, in certain circumstances, there could be quarterly adjustments to rebate practices and stated, in recitals 1017 to 1028 of the contested decision, that, because of the fact that the relevant market was very fast-moving, innovation in the relevant sector made it difficult or even impossible to make long-term predictions. Similarly, recitals 1025 to 1027 of the contested decision contain a reference to the duration of the contracts and to the fact that it was necessary regularly to ‘refresh’ product cycles.

510    Next, recitals 201 and 202 of the contested decision indicate that the Commission considered that some of the relevant negotiations between Intel and the OEMs were conducted on a quarterly basis. Those negotiations also concerned a relatively short period, which could allow an as-efficient competitor to offer its own x86 CPUs to those OEMs more easily. Similarly, in recitals 965 to 968 of the contested decision, the Commission examined Intel’s argument that the 30-day termination notice provision in the HPA agreements gave HP greater freedom to compare its offers with those of AMD, and indicated that Intel’s status as an unavoidable trading partner and the effects of its rebates resulted in that argument being rejected. At the 2020 hearing, the Commission stated that, in some cases, as regards HP, there had been several renewals of the agreements with Intel on a monthly basis. As regards Dell, the Commission stated, in recital 1227 of the contested decision, that, due to the absence of any written contract with Intel, as regards the rebates granted under the Meet Competition Programme, those rebates were the subject of ‘constant’ oral renegotiation, such that Intel had great flexibility to change discounts.

511    The fact remains that, first, the purpose of recitals 1013 to 1035 of the contested decision was solely to define the time horizon on which the OEMs based their decisions regarding their supply requirements for x86 CPUs as an assumption underlying the calculation of the contestable share of the rebates for each of the OEMs concerned. The Commission concluded that, in relation to the AEC test, it would proceed on the assumption that the relevant time horizon was one year.

512    Consequently, the time horizon was used in that regard to determine the methodology for calculating an OEM’s contestable share, which then had to be compared with other factors in the AEC test in order to assess whether the rebates at issue could have a foreclosure effect. Such an examination does not therefore constitute an analysis of the duration of the rebates as a factor which is capable, in itself, of demonstrating their ability to have a foreclosure effect.

513    Second, it is apparent from recitals 201, 202, 965 to 968 and 1227 of the contested decision that the Commission examined the duration and form of the OEMs’ commitments to Intel giving a right to rebates as factors capable of encouraging or hindering a new competitor’s entry into the market, having regard in particular to the temporal scope of those commitments or Intel’s ability to pay or adjust its rebates within a short time frame.

514    However, although those aspects of the time horizon appeared to it to be relevant, the Commission examined them only in a haphazard and limited manner, in recitals 201, 202, 965 to 968 and 1227 of the contested decision. It did not carry out a thorough and exhaustive examination for all OEMs of those aspects in so far as they were capable of determining or strengthening the capability of Intel’s pricing practices at issue to have a foreclosure effect.

515    It is apparent from the foregoing that the Commission did not examine the duration of the rebates as a factor intrinsically relevant to determining the capability of Intel’s pricing practices at issue to have a foreclosure effect.

516    Second, the Commission asserts, in essence, that, even if the AEC test did not demonstrate the capability of the rebates at issue to have a foreclosure effect, it is the total period during which the applicant applied the rebates and exclusivity payments to OEMs which should be considered and that, in so far as the rebates lasted one year for Lenovo and several years for the other OEMs and for MSH, the conclusion must be drawn that a competitor of Intel in the x86 CPU market would have had to accept a drop in profitability and a much lower level of profitability on those sales as compared with Intel. In the Commission’s view, those considerations are apparent from paragraphs 93 and 195 of the initial judgment and are therefore definitive.

517    In that regard, first, it is apparent from paragraph 81 above that the operative part of the judgment on the appeal sets aside the entirety of the initial judgment. Consequently, the Court must, following the referral back, carry out a fresh examination of the parties’ arguments concerning the duration of the rebates, without being bound by paragraphs 93 and 195 of the initial judgment, which it does not accept for its own account.

518    Second, it is apparent from paragraphs 138 and 139 of the judgment on the appeal that, in the case where the undertaking concerned submits, during the administrative procedure, on the basis of supporting evidence, that its conduct was not capable of restricting competition and, in particular, was not capable of producing the alleged foreclosure effects, the Commission is required to assess all of the criteria referred to in paragraph 139 of that judgment, and not solely the criterion relating to the duration of the rebates set out in that paragraph. Accordingly, the sole reference to the period during which the rebates were granted to the OEMs and MSH is not sufficient, in itself, notwithstanding the conclusions which may be drawn from the AEC test, to justify definitive conclusions as to the foreclosure effects thereby produced.

519    Third, the Commission cannot succeed in its assertion that the duration of Intel’s practices cannot be isolated from their timing, since they sought to overcome Intel’s inability to produce a timely technical response to AMD’s 64-bit x86 CPUs. For the same reasons as those set out in paragraph 518 above, that argument, even if it appears as such in the contested decision, is not in itself sufficient to justify definitive findings as to the foreclosure effects thereby produced.

520    There being no need to rule on the applicant’s arguments concerning the amounts of the rebates, it is apparent from the foregoing that the Commission erred in failing to examine, in the contested decision, the duration of the rebates as evidence making it possible to determine the capacity of Intel’s pricing practices at issue to have a foreclosure effect.

3.      Conclusions on taking account of the criteria mentioned in paragraph 139 of the judgment on the appeal

521    In the light of all the considerations set out in paragraphs 485 to 520 above, there being no need to analyse the applicant’s complaints regarding the criteria concerning the amount of the rebates and the strategy aiming to exclude competitors from the market, it must be concluded that the applicant is justified in maintaining that the Commission’s analysis in the contested decision of the criteria mentioned in paragraph 139 of the judgment on the appeal is vitiated by a number of errors. The Commission did not consider properly the criterion relating to the share of the market covered by the contested practice and did not analyse correctly the duration of the rebates.

D.      Conclusion on the application for annulment of the contested decision

522    It is apparent from paragraphs 124 to 126 above that, although a system of rebates set up by an undertaking in a dominant position on the market may be characterised as a restriction of competition, since, given its nature, it may be presumed to have restrictive effects on competition, the fact remains that what is involved is, in that regard, a mere presumption and not a per se infringement of Article 102 TFEU, which would relieve the Commission in all cases of the obligation to examine whether there were anticompetitive effects. Where an undertaking in a dominant position submits, during the administrative procedure, on the basis of supporting evidence, that its conduct was not capable of restricting competition and, in particular, was not capable of producing the alleged foreclosure effects, the Commission must analyse the foreclosure capability of the system of rebates by applying the five criteria listed in paragraph 139 of the judgment on the appeal. In addition, where the Commission has carried out an AEC test, that test is one of the factors which the Commission must take into account in assessing whether the system of rebates is capable of restricting competition.

523    In the present case, the applicant submitted, during the administrative procedure, on the basis of supporting evidence, that the rebates at issue were not capable of producing the alleged foreclosure effects. In recitals 1002 to 1573 of the contested decision, the Commission carried out an AEC test and, in the light of the results of that test, concluded, in recitals 1574 and 1575 of that decision, that Intel’s rebates and payments at issue were capable of having or likely to have anticompetitive foreclosure effects, since even an as-efficient competitor would have been prevented from supplying Dell, HP, NEC and Lenovo for their x86 CPU requirements or from having MSH sell computers equipped with its x86 CPUs.

524    However, it follows from all of the foregoing that, first, the AEC test carried out in the contested decision is vitiated by errors and, second, as regards the criteria mentioned in paragraph 139 of the judgment on the appeal, the Commission did not consider properly the criterion relating to the share of the market covered by the contested practice and did not analyse correctly the duration of the rebates.

525    It should be noted, as regards the rebates granted to HP, that it has been held, in paragraph 334 above, that the Commission did not establish to the requisite legal standard its finding that, during the period from November 2002 to May 2005, the rebate which Intel granted to HP was capable of having or was likely to have an anticompetitive foreclosure effect, since it has not demonstrated that those effects were present for the period between 1 November 2002 and 31 September 2003. Even if it were necessary to infer that the AEC test could be regarded as conclusive for part of the period from November 2002 to May 2005, that could not demonstrate to the requisite legal standard the foreclosure effect of the rebates granted to HP, since the Commission did not consider properly the criterion relating to the share of the market covered by the contested practice and did not analyse correctly the duration of the rebates.

526    Therefore, the Commission is not in a position to determine that the applicant’s rebates and payments at issue were capable of having or likely to have anticompetitive effects and that they therefore constituted an infringement of Article 102 TFEU.

527    Consequently, the Court considers that the grounds of the contested decision are not capable of serving as a basis for Article 1(a) to (e) of that decision.

528    Furthermore, in reply to a question from the Court of 2 April 2012, seeking to ascertain, as regards a possible adjustment to the amount of the fine should the contested decision be the subject of annulment in part, the relative value of the infringements consisting of the exclusivity payments compared with the infringements consisting of the naked restrictions, the Commission, in a reply lodged on 8 May 2012, replied solely in relation to the gravity of the infringements, arguing that it had assessed the conduct in question as a whole and had come to the view that those infringements complemented and mutually reinforced one another.

529    Since the Court is not in a position to identify the amount of the fine relating solely to the naked restrictions, Article 2 of the contested decision must therefore also be annulled.

530    Article 3 of the contested decision must be annulled in so far as it concerns the exclusivity rebates.

531    The action is dismissed as to the remainder, having regard, in particular, to the findings in the initial judgment which the Court accepts, as set out in paragraphs 96 to 98 above.

 The head of claim seeking the annulment of the fine or a reduction in the amount of the fine

532    In the light of the foregoing, it is not necessary to give a ruling on the second head of claim, put forward in the alternative.

 Costs

533    In so far as, in the judgment on the appeal, the Court of Justice set aside the initial judgment and reserved the costs, it is for the General Court, in accordance with Article 219 of the Rules of Procedure, to decide, in the present judgment, on all the costs relating to the proceedings brought before it, namely the proceedings in Cases T‑286/09 and T‑286/09 RENV, and on the costs relating to the appeal proceedings, namely the proceedings in Case C‑413/14 P.

534    Under Article 134 of the Rules of Procedure, the unsuccessful party is to be ordered to pay the costs if they have been applied for in the successful party’s pleadings. Where there is more than one unsuccessful party, the Court is to decide how the costs are to be shared. Where each party succeeds on some and fails on other heads, the parties are to bear their own costs. However, if it appears justified in the circumstances of the case, the Court may order that one party, in addition to bearing its own costs, pay a proportion of the costs of the other party.

535    Under Article 138(3) of the Rules of Procedure, the Court may order an intervener other than those referred to in Article 138(1) and (2) to bear his or her own costs.

536    Since the Commission has been unsuccessful in part, it must be ordered to pay, in addition to its own costs relating to the proceedings before the General Court in Cases T‑286/09 and T‑286/09 RENV and to the appeal proceedings before the Court of Justice in Case C‑413/14 P, two thirds of the costs incurred by the applicant and by ACT in those proceedings, while the applicant and ACT are each to bear one third of their own costs.

537    UFC is to bear its own costs relating to the proceedings brought before the General Court in Cases T‑286/09 and T‑286/09 RENV and to the appeal proceedings before the Court of Justice in Case C‑413/14 P.

On those grounds,

THE GENERAL COURT (Fourth Chamber, Extended Composition)

hereby:

1.      Annuls Article 1(a) to (e) and Article 2 of Commission Decision C(2009) 3726 final of 13 May 2009 relating to a proceeding under Article [102 TFEU] and Article 54 of the EEA Agreement (Case COMP/C3/37.990 – Intel);

2.      Annuls Article 3 of Decision C(2009) 3726 final solely in so far as it concerns Article 1(a) to (e) of that decision;

3.      Dismisses the action as to the remainder;

4.      Orders the European Commission to pay, in addition to its own costs relating to the proceedings before the General Court in Cases T286/09 and T286/09 RENV and to the appeal proceedings before the Court of Justice in Case C413/14 P, two thirds of the costs incurred by Intel Corporation, Inc., and by Association for Competitive Technology, Inc., in those proceedings, while Intel Corporation and Association for Competitive Technology are each to bear one third of their own costs;

5.      Orders Union fédérale des consommateurs – Que choisir (UFC – Que choisir) to bear its own costs relating to the proceedings brought before the General Court in Cases T286/09 and T286/09 RENV and to the appeal proceedings before the Court of Justice in Case C413/14 P.

Kanninen

Schwarcz

Iliopoulos

Delivered in open court in Luxembourg on 26 January 2022.

E. Coulon

 

M. van der Woude

Registrar

 

President


Table of contents


Background to the dispute

Administrative procedure

The contested decision

Relevant market

Dominance

Abuse and fine

Operative Part

Procedure before the General Court and the Court of Justice

Procedure and forms of order sought following the referral back

Law

Arguments of the parties relating to the subject matter of the proceedings following the referral back

The Commission’s arguments concerning the admissibility of certain arguments contained in the main observations of the applicant and ACT

Substance

The heads of claim for annulment of the contested decision

I. The method defined by the Court of Justice for assessing whether a system of rebates has the capacity to restrict competition

II. The principles arising from the judgment on the appeal

III. The substance of the arguments raised by the applicant and ACT

A. The argument that the contested decision is founded on an incorrect legal analysis

B. The argument that the contested decision must be annulled on the ground that it contains an AEC analysis which is vitiated by numerous errors

1. The scope of the Court’s review

2. General considerations concerning the AEC analysis

3. The burden of proof and the standard of proof required

4. The substance of the arguments that the contested decision is vitiated by numerous errors in the AEC test

(a) General arguments concerning the alleged errors in the AEC test applied to Dell

(1) The assessment of the contestable share

(i) The arguments concerning the principle of legal certainty

(ii) The assessment of the contestable share as 7.1%

(iii) The applicant’s claim regarding the initial part of the relevant period, from December 2002 to October 2003

(2) The alternative method of calculation

(3) The reinforcing factors

(4) Conclusion on the AEC test for the rebates granted to Dell

(b) The alleged errors in the AEC test applied to HP

(1) The period examined by the AEC test

(2) The alleged reinforcing factors

(c) The alleged errors in the AEC test applied to NEC

(1) The calculation of the conditional portion of the rebates

(i) The evidence taken into account in the contested decision

(ii) The evidence adduced by Intel

(2) Use of the fourth quarter of 2002 as a reference

(d) The alleged errors in the AEC test applied to Lenovo

(1) General overview of the part of the contested decision devoted to Lenovo

(2) The conditional portion of the rebates

(e) The alleged errors in the AEC test applied to MSH

(f) Conclusions on the AEC test

C. The argument that the contested decision did not properly analyse and take account of the criteria referred to in paragraph 139 of the judgment on the appeal

1. The coverage rate

2. Duration and amount of the rebates

3. Conclusions on taking account of the criteria mentioned in paragraph 139 of the judgment on the appeal

D. Conclusion on the application for annulment of the contested decision

The head of claim seeking the annulment of the fine or a reduction in the amount of the fine

Costs


*      Language of the case: English.


1      Confidential information omitted.