Language of document : ECLI:EU:T:2021:252

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

12 May 2021 (*)

(State aid – Aid implemented by Luxembourg in favour of Amazon – Decision declaring the aid incompatible with the internal market and unlawful and ordering its recovery – Tax ruling – Transfer pricing – Selective tax advantage – Transfer pricing arrangement – Functional analysis)

In Cases T‑816/17 and T‑318/18,

Grand Duchy of Luxembourg, represented by T. Uri, acting as Agent, and by D. Waelbroeck, A. Steichen and J. Bracker, lawyers,

applicant in Case T‑816/17,

supported by

Ireland, represented by J. Quaney and A. Joyce, acting as Agents, and by P. Gallagher, Senior Counsel, B. Doherty, Barrister, and S. Kingston, Senior Counsel,

intervener in Case T‑816/17,

Amazon EU Sàrl, established in Luxembourg (Luxembourg),

Amazon.com, Inc., established in Seattle, Washington (United States),

represented by D. Paemen, M. Petite and A. Tombiński, lawyers,

applicants in Case T‑318/18,

v

European Commission, represented, in Case T‑816/17, by P. Stancanelli, P.‑J. Loewenthal and F. Tomat, acting as Agents, and by M. Chammas, lawyer, and, in Case T‑318/18, by P.-J. Loewenthal and F. Tomat,

defendant,

ACTIONS pursuant to Article 263 TFEU for the annulment of Commission Decision (EU) 2018/859 of 4 October 2017 on State aid SA.38944 (2014/C) (ex 2014/NN) implemented by Luxembourg to Amazon (OJ 2018 L 153, p. 1),

THE GENERAL COURT (Seventh Chamber, Extended Composition),

composed of M. van der Woude, President, V. Tomljenović (Rapporteur) and A. Marcoulli, Judges,

Registrar: S. Spyropoulos, Administrator,

having regard to the written part of the procedure and further to the hearing on 5 and 6 March 2020,

gives the following

Judgment

I.      Background to the dispute

1        Amazon.com, Inc., which has its registered office in the United States, and the companies under its control (together, ‘the Amazon group’) carry on online activities, including online retail transactions and the provision of various online services. To that end, the Amazon group manages several internet sites in various languages of the European Union, including amazon.de, amazon.fr, amazon.it and amazon.es.

2        Prior to May 2006, the Amazon group’s European business was managed from the United States. In particular, retail and service activities on the EU websites were carried out by two entities established in the United States, namely Amazon.com International Sales, Inc. (‘AIS’) and Amazon International Marketplace (‘AIM’), as well as by others established in France, Germany and the United Kingdom.

3        In 2003, a restructuring of the Amazon group’s business in Europe was planned. That restructuring, which was in fact carried out in 2006 (‘the 2006 restructuring’), was structured around the formation of two companies established in Luxembourg (Luxembourg). More specifically, the companies in question were, first, Amazon Europe Holding Technologies SCS (‘LuxSCS’), a Luxembourg limited partnership, the partners of which were United States companies, and, secondly, Amazon EU Sàrl (‘LuxOpCo’), which, like LuxSCS, had its registered office in Luxembourg.

4        LuxSCS first concluded several agreements with certain Amazon group entities established in the United States, namely:

–        licence and assignment agreements for pre-existing intellectual property (together, ‘the Buy-In Agreement’) with Amazon Technologies, Inc. (‘ATI’), an Amazon group entity established in the United States;

–        a cost-sharing agreement (‘the CSA’) concluded in 2005 with ATI and A9.com, Inc. (‘A9’), an Amazon group entity established in the United States. Under the Buy-In Agreement and the CSA, LuxSCS obtained the right to exploit certain intellectual property rights and ‘derivative works’ thereof, which were owned and further developed by A9 and ATI. The intangible assets covered by the CSA consisted essentially of three categories of intellectual property, namely technology, customer data and trade marks. Under the CSA and the Buy-In Agreement, LuxSCS could also sub-license the intangible assets, in particular with a view to operating the EU websites. In return for those rights, LuxSCS was required to pay Buy-In payments and its annual share of the costs related to the CSA development programme.

5        Secondly, LuxSCS entered into a licence agreement with LuxOpCo, which took effect on 30 April 2006, relating to the abovementioned intangible assets (‘the Licence Agreement’). Under that agreement, LuxOpCo obtained the right to use the intangible assets in exchange for the payment of a royalty to LuxSCS (‘the royalty’).

6        Lastly, LuxSCS concluded an agreement for the licensing and assignment of intellectual property rights with Amazon.co.uk Ltd, Amazon.fr SARL and Amazon.de GmbH, under which LuxSCS received certain trade marks and the intellectual property rights in respect of the EU websites.

7        In 2014, the Amazon group underwent a second restructuring and the contractual arrangement between LuxSCS and LuxOpCo was no longer applicable.

A.      The tax ruling at issue

8        In preparation for the 2006 restructuring, Amazon.com and a tax adviser, by letters of 23 and 31 October 2003, requested the Luxembourg tax administration to issue a tax ruling confirming the treatment of LuxOpCo and LuxSCS for the purposes of Luxembourg corporate income tax.

9        By its letter of 23 October 2003, Amazon.com requested approval for the method of calculating the rate of the royalty that LuxOpCo was to pay to LuxSCS from 30 April 2006. That request by Amazon.com was based on a transfer pricing report prepared by its tax advisors (‘the 2003 transfer pricing report’). The authors of that report proposed, in essence, a transfer pricing arrangement which, in their view, enabled the determination of the corporate income tax liability which LuxOpCo was required to pay in Luxembourg. More specifically, by the letter of 23 October 2003, Amazon.com had requested confirmation that the transfer pricing arrangement determining the rate of the annual royalty that LuxOpCo was to pay to LuxSCS under the Licence Agreement, as set out in the 2003 transfer pricing report, would result in an ‘appropriate and acceptable profit’ for LuxOpCo with respect to the transfer pricing policy and Article 56 and Article 164(3) of the loi du 4 décembre 1967 concernant l’impot sur le revenue, telle que modifiée (Law of 4 December 1967 on income tax, as amended) (‘the LIR’). The method of calculating the royalty payable by LuxOpCo to LuxSCS set out in the letter of 23 October 2003 was described as follows:

‘1.      Compute and allocate to LuxOpCo the “LuxOpCo Return”, which is equal to the lesser of (a) [confidential]% (1) of LuxOpCo’s total EU Operating Expenses for the year and (b) total EU Operating Profit attributable to the European Web Sites for such year;

2.      The License Fee shall be equal to EU Operating Profit minus the LuxOpCo Return, provided that the License Fee shall not be less than zero;

3.      The Royalty Rate for the year shall be equal to the License Fee divided by total EU Revenue for the year;

4.      Notwithstanding the foregoing, the amount of the LuxOpCo Return for any year shall not be less than 0.45% of EU Revenue, nor greater than 0.55% of EU Revenue;

5.      (a) In the event that the LuxOpCo Return determined under step (1) would be less than 0.45% of EU Revenues, the LuxOpCo Return shall be adjusted to equal the lesser of (i) 0.45% of Revenue or EU Operating Profit or (ii) EU Operating Profit;

(b) In the event that the LuxOpCo Return determined under step (1) would be greater than 0.55% of EU Revenues, the LuxOpCo Return shall be adjusted to equal the lesser of (i) 0.55% of EU Revenues or (ii) EU Operating Profit.’

10      By letter of 31 October 2003, drafted by another tax adviser, Amazon.com requested confirmation of the tax treatment of LuxSCS, of its partners established in the United States and of the dividends received by LuxOpCo as part of that structure. The letter explained that LuxSCS, as a ‘Société en Commandite Simple’, did not have a separate tax personality from that of its partners and that, as a result, it was not subject to corporate income tax or net wealth tax in Luxembourg.

11      On 6 November 2003, the Administration des contributions directes du Grand-Duché de Luxembourg (‘the Luxembourg tax administration’ or ‘the Luxembourg tax authorities’) sent Amazon.com a letter (‘the tax ruling at issue’) which reads, in part, as follows:

‘Sir,

After having made myself acquainted with the letter of [O]ctober 31, 2003, directed to me by [your tax advisor] just as with your letter of [O]ctob[er] 23, 2003 and dealing with your position regarding Luxembourg tax treatment within the framework of your future activities, I am pleased to inform you that I may approve the contents of the two letters. …’

12      At the request of Amazon.com, the Luxembourg tax administration extended the validity of the tax ruling at issue in 2010 and effectively applied it until June 2014, when the European structure of the Amazon group was modified. Thus, the tax ruling at issue was applied from 2006 to 2014 (‘the relevant period’).

B.      The administrative procedure before the Commission

13      On 24 June 2014, the European Commission requested that the Grand Duchy of Luxembourg provide information on the tax rulings granted to the Amazon group. On 7 October 2014, it published the decision to initiate a formal investigation procedure under Article 108(2) TFEU.

14      During the investigation thus initiated, the Commission requested a variety of information from the Grand Duchy of Luxembourg and Amazon.com. Among the responses to those requests for information, Amazon.com submitted a copy of an opinion of the United States Tax Court of 23 March 2017 (‘the opinion of the United States Tax Court’) which had been delivered on an application brought by the Internal Revenue Service (the United States federal government tax collection agency; IRS) concerning the amount of payments related to the agreements referred to in paragraph 4 above.

15      In addition, Amazon.com submitted to the Commission a new transfer pricing report drawn up by a tax adviser, the purpose of which was to verify ex post whether the royalty paid by LuxOpCo to LuxSCS in accordance with the tax ruling at issue complied with the arm’s length principle (‘the 2017 transfer pricing report’).

C.      The contested decision

16      On 4 October 2017, the Commission adopted Decision (EU) 2018/859 on State aid SA.38944 (2014/C) (ex 2014/NN) implemented by Luxembourg to Amazon (OJ 2018 L 153, p. 1; ‘the contested decision’).

17      Article 1 of that decision reads, in part, as follows:

18      ‘The [tax ruling at issue], by virtue of which Luxembourg endorsed a transfer pricing arrangement … that allowed [LuxOpCo] to assess its corporate income tax liability in Luxembourg from 2006 to 2014 and the subsequent acceptance of the yearly corporate income tax declaration based thereon constitutes [State] aid …’

1.      Presentation of the factual and legal context

19      In Section 2 of the contested decision, entitled ‘Factual and Legal Background’, the Commission presents, inter alia, the Amazon group, the tax ruling at issue, the relevant national legal framework and guidance on transfer pricing.

(a)    Presentation of the Amazon group

20      As regards the presentation of the Amazon group, the Commission described the Amazon group’s retail and service businesses, as well as the composition of that group in so far as that was relevant to the contested decision.

21      The European structure of the Amazon group during the relevant period was presented by the Commission in the following schematic form:

Image not found

22      First, with regard to LuxSCS, the Commission pointed out that that company did not have any physical presence or employees in Luxembourg. According to the Commission, during the relevant period, LuxSCS functioned solely as an intangible assets holding company for the Amazon group’s European operations, for which LuxOpCo was responsible as the principal operator. It stated, however, that LuxSCS had also granted intra-group loans to several entities in the Amazon group. The Commission also noted that LuxSCS was a party to several intra-group agreements with ATI, A9 and LuxOpCo (see paragraphs 3 and 5 above).

23      Secondly, as regards LuxOpCo, the Commission placed particular emphasis on the fact that, during the relevant period, LuxOpCo was a wholly owned subsidiary of LuxSCS.

24      According to the Commission, as from the 2006 restructuring of the Amazon group’s European operations, LuxOpCo functioned as the Amazon group’s headquarters in Europe and the principal operator of the Amazon group’s European online retail and service business as carried out through the EU websites. The Commission stated that, in that capacity, LuxOpCo had to manage decision-making related to the retail and service businesses carried out through the EU websites, along with managing key physical components of the retail business. In addition, as the seller of record of the Amazon group’s inventory in Europe, LuxOpCo was also responsible for managing inventory on the EU websites. It held title to that inventory and bore the risks and losses. The Commission further stated that LuxOpCo had recorded revenue in its accounts both from product sales and from order fulfilment. Lastly, LuxOpCo also performed treasury management functions for the Amazon group’s European operations.

25      Next, the Commission indicated that LuxOpCo had held shares in Amazon Services Europe (‘ASE’) and Amazon Media Europe (‘AMEU’), two Amazon group entities resident in Luxembourg, and also in the subsidiaries of Amazon.com established in the United Kingdom, France and Germany (‘the EU local affiliates’), which performed various intra-group services in support of LuxOpCo’s business. During the relevant period, ASE operated the Amazon group’s EU third-party seller business, ‘Marketplace’. AMEU operated the Amazon group’s EU digital business, such as, for example, the sale of MP3s and eBooks. The EU local affiliates supplied services relating to the operation of the EU websites.

26      The Commission added that, during the relevant period, ASE and AMEU, both Luxembourg resident companies, formed a fiscal unity with LuxOpCo for Luxembourg tax purposes in which LuxOpCo operated as the parent of the unity. Those three entities therefore constituted a single taxpayer.

27      Lastly, in addition to the Licence Agreement concluded between LuxOpCo and LuxSCS, the Commission described in detail certain other intra-group agreements to which LuxOpCo was a party during the relevant period, namely certain service agreements concluded on 1 May 2006 with the EU local affiliates and intellectual property licence agreements concluded on 30 April 2006 with ASE and AMEU, under which ASE and AMEU were granted non-exclusive sub-licences to the intangible assets.

(b)    Presentation of the tax ruling at issue

28      Having examined the structure of the Amazon group, the Commission describes the tax ruling at issue.

29      In that regard, first, it referred to the letters of 23 and 31 October 2003, mentioned in paragraphs 8 to 10 above.

30      Secondly, the Commission explained the content of the 2003 transfer pricing report on the basis of which the method for establishing the royalty was proposed.

31      The Commission began by indicating that the 2003 transfer pricing report provided a functional analysis of LuxSCS and LuxOpCo, according to which LuxSCS’s principal activities were limited to those of an intangible assets holding company and a participant in the ongoing development of the intangible assets through the CSA. LuxOpCo was described in that report as managing strategic decision-making related to the EU websites’ retail and service businesses, and also managing the key physical components of the retail business.

32      Next, the Commission stated that the 2003 transfer pricing report contained a section dealing with the selection of the most appropriate transfer pricing method for determining whether the royalty rate complied with the arm’s length principle. Two methods were considered in the report: one based on the comparable uncontrolled price method (‘the CUP method’) and another based on the residual profit split method.

33      Applying the CUP method, the 2003 transfer pricing report calculated an arm’s length range for the royalty rate of 10.6 to 13.6% on the basis of a comparison with an agreement between Amazon.com and a retailer in the United States, namely the [confidential] agreement.

34      Applying the residual profit split method, the 2003 transfer pricing report estimated the return associated with LuxOpCo’s ‘routine functions in its role as the European operating company’ based on a mark-up on costs to be incurred by LuxOpCo. To that end, the ‘net cost plus mark-up’ was considered the profit level indicator for the purpose of determining the arm’s length remuneration for the anticipated functions of LuxOpCo. It was proposed that a mark-up of [confidential] be applied to LuxOpCo’s adjusted operating costs. The Commission observed that, according to the 2003 transfer pricing report, the difference between that return and LuxOpCo’s operating profit constituted the residual profit, which was wholly attributable to the use of the intangible assets licensed from LuxSCS. The Commission also stated that, on the basis of that calculation, the authors of the 2003 transfer pricing report had concluded that a royalty rate in a range of 10.1% to 12.3% of LuxOpCo’s net revenues would be consistent with the arm’s length standard under the guidelines of the Organisation for Economic Cooperation and Development (OECD).

35      Lastly, the Commission stated that the authors of the 2003 transfer pricing report had considered that the results converged and had indicated that the arm’s length range for the royalty rate from LuxOpCo to LuxSCS was 10.1% to 12.3% of LuxOpCo’s sales. However, the authors of the 2003 transfer pricing report considered that the residual profit split analysis was more reliable and should therefore be selected.

36      Thirdly, in Section 2.2.5 of the contested decision, entitled ‘Consequences of the contested tax ruling’, the Commission stated that, by the tax ruling at issue, the Luxembourg tax administration had accepted that the arrangement for determining the level of the royalty, which in turn had determined LuxOpCo’s annual taxable income in Luxembourg, was at arm’s length. It added that the tax ruling at issue was relied upon by LuxOpCo when filing its annual tax declarations.

(c)    Description of the relevant national legal framework

37      As regards the relevant national legal framework, the Commission cited Article 164(3) of the LIR. According to that provision, ‘taxable income comprises hidden profit distributions’ and ‘a hidden profit distribution arises in particular when a shareholder, a stockholder or an interested party receives either directly or indirectly benefits from a company or an association which he normally would not have received if he had not been a shareholder, a stockholder or an interested party’. In that context, the Commission explained, in particular, that, during the relevant period, Article 164(3) of the LIR was interpreted by the Luxembourg tax administration as enshrining the arm’s length principle in Luxembourg tax law.

(d)    Presentation of the OECD transfer pricing framework

38      In recitals 244 to 249 of the contested decision, the Commission presented the OECD transfer pricing framework. According to the Commission, ‘transfer prices’, as understood by the OECD in the guidelines published by that organisation in 1995, 2010 and 2017, are the prices at which a company transfers physical goods or intangible property or provides services to its associated companies. According to the arm’s length principle, as applied in corporate taxation, national tax administrations should only accept the transfer prices agreed between associated group companies for intra-group transactions if those prices reflect what would have been agreed in uncontrolled transactions, that is to say, transactions between independent companies negotiating under comparable circumstances on the market. In addition, the Commission stated that the arm’s length principle is based on the separate entity approach, according to which the members of a corporate group are treated as separate entities for tax purposes.

39      The Commission also indicated that the 1995, 2010 and 2017 OECD Guidelines list five methods for establishing arm’s length pricing for intra-group transactions. Only three of them were relevant for the contested decision, namely the CUP method, the transactional net margin method (‘the TNMM’) and the transactional profit split method. The Commission described those methods in recitals 250 to 256 of the contested decision.

2.      Assessment of the tax ruling at issue

40      In recital 154 of the contested decision, the Commission indicated that the tax ruling at issue accepted that the transfer pricing arrangement for the purposes of determining the level of the annual royalty to be paid by LuxOpCo to LuxSCS under the Licence Agreement, which in turn determined LuxOpCo’s annual taxable income in Luxembourg, was at arm’s length. In recital 155 of the contested decision, it also stated that LuxOpCo had relied on the tax ruling at issue during the relevant period to determine its annual corporate income tax liability in Luxembourg for the purpose of filing its annual tax declarations. As is clear from Article 1 of the contested decision and despite some imprecision in recitals 605 and 606 of that decision, according to the Commission, the State aid therefore consisted, in the present case, of the tax ruling at issue read in conjunction with the acceptance of LuxOpCo’s annual tax declarations (as opposed to the tax ruling at issue as such).

41      Section 9 of the contested decision, entitled ‘Assessment of the contested measure’, was intended to demonstrate that the tax ruling in question and the acceptance of LuxOpCo’s annual tax declarations, taken together, did indeed constitute State aid.

42      In order to do so, after noting the conditions for the existence of State aid laid down in Article 107(1) TFEU, the Commission stated that the first condition for the existence of State aid, namely that there must be an intervention by the State or through State resources, was satisfied in the present case. In that respect, it noted, first, that the tax ruling at issue was imputable to the Grand Duchy of Luxembourg. Secondly, it considered that the tax ruling at issue resulted in a lowering of LuxOpCo’s tax liability in Luxembourg as compared to the tax paid by similarly situated corporate taxpayers. The tax ruling at issue therefore gave rise to loss of State resources, since it led the Grand Duchy of Luxembourg to waive tax revenue that it would otherwise have been entitled to collect from LuxOpCo.

43      As regards the second and fourth conditions for the existence of State aid, the Commission considered, first, that tax ruling at issue had to be regarded as affecting intra-EU trade, since LuxOpCo was part of the Amazon group operating in several Member States and it had carried out retail business through the EU websites. The Commission added that, by granting ‘favourable tax treatment to Amazon’, the Grand Duchy of Luxembourg had potentially drawn investments away from Member States which did not offer such favourable tax treatment to companies forming part of multinational group. Secondly, the Commission indicated that, in so far as the tax ruling at issue had relieved LuxOpCo of corporate income taxes that it would normally have been obliged to pay, that ruling constituted operating aid. Thus, the tax ruling at issue, by freeing up financial resources for LuxOpCo which the latter could use to invest in its business operations, distorted competition on the market.

44      As regards the third condition for the existence of State aid, the Commission explained that, where a tax ruling endorses a result that does not reflect in a reliable manner what would result from a normal application of the ordinary tax system, without justification, that ruling will confer a selective advantage on its addressee in so far as that selective treatment results in a lowering of that taxpayer’s tax liability as compared to companies in a similar factual and legal situation. The Commission also considered that, in the present case, the tax ruling at issue had conferred a selective advantage on LuxOpCo by lowering its corporate income tax liability in Luxembourg.

(a)    Analysis of the existence of an advantage

45      In Section 9.2 of the contested decision, entitled ‘Advantage’, the Commission discussed the reasons it considered that the tax ruling at issue conferred an economic advantage on LuxOpCo.

46      As a preliminary point, the Commission stated that, in respect of tax measures, an advantage for the purposes of Article 107 TFEU may be granted by reducing an undertaking’s taxable base or the amount of tax due from it. In recital 402 of the contested decision, it noted that, according to the Court of Justice, in order to decide whether a method of assessment of taxable income confers an advantage on its beneficiary, it is necessary to compare that method with the ordinary tax system, based on the difference between profits and outgoings of an undertaking carrying on its activities in conditions of free competition. Consequently, according to the Commission, ‘a tax ruling that enables a taxpayer to employ transfer prices in its intra-group transactions that do not resemble prices which would be charged in conditions of free competition between independent undertakings negotiating under comparable circumstances at arm’s length confers an advantage on that taxpayer, in so far as it results in a reduction of the company’s taxable income and thus its taxable base under the ordinary corporate income tax system’.

47      In the light of those considerations, the Commission concluded, in recital 406 of the contested decision, that, in order to establish that the tax ruling at issue confers an economic advantage on LuxOpCo, it had to demonstrate that the transfer pricing arrangement endorsed in the tax ruling at issue produced an outcome that departed from a reliable approximation of a market-based outcome, resulting in a reduction of LuxOpCo’s taxable basis for corporate income tax purposes. According to the Commission, the tax ruling at issue had produced such an outcome.

48      That conclusion is based on a primary finding and three subsidiary findings.

(1)    The primary finding of an advantage

49      In Section 9.2.1 of the contested decision, entitled ‘Primary finding of an economic advantage’, the Commission considered that, by endorsing a transfer pricing arrangement that attributed a remuneration to LuxOpCo solely for ‘routine’ functions and that attributed the entire profit generated by LuxOpCo in excess of that remuneration to LuxSCS in the form of a royalty payment, the tax ruling at issue had produced an outcome that departed from a reliable approximation of a market-based outcome.

50      In essence, by its primary line of reasoning, the Commission considered that the functional analysis of LuxOpCo and LuxSCS set out by the authors of the 2003 transfer pricing report and ultimately accepted by the Luxembourg tax administration was incorrect and could not result in an arm’s length outcome. On the contrary, the Luxembourg tax administration should have concluded that LuxSCS did not perform any ‘unique and valuable’ functions in relation to the intangible assets for which it merely held the legal title.

51      For the purposes of its demonstration, the Commission examined the functions performed, the assets used and the risks borne by LuxSCS and LuxOpCo.

52      Next, as is apparent from recital 519 of the contested decision, on the basis of its functional analysis of LuxOpCo and LuxSCS, the Commission examined the choice of the most appropriate transfer pricing method in this case.

53      As regards the CUP method, the Commission considered, following an analysis based on five comparability criteria set out in the OECD Guidelines, that the application of that method, as set out in the 2003 transfer pricing report, had produced an exaggerated result which had exposed LuxOpCo to the risk of incurring losses.

54      According to the Commission, in the present case, the TNMM was the most appropriate transfer pricing method to determine the royalty to be paid by LuxOpCo under the Licence Agreement. It considered that the party performing unique and valuable functions was LuxOpCo and not LuxSCS. Consequently, the tested party for the purpose of applying the TNMM should have been LuxSCS rather than LuxOpCo.

55      Lastly, in Section 9.2.1.4 of the contested decision, the Commission carried out its own application of the TNMM in this case.

56      In its view, LuxSCS should have been the tested entity. The Luxembourg tax administration should not have accepted Amazon.com’s claim that the mere legal ownership of the intangible assets constituted a ‘unique contribution’ for which LuxSCS should receive a remuneration consisting of almost all profits derived from LuxOpCo’s business activities. In that regard, the Commission referred, inter alia, to its own functional analysis of LuxSCS and LuxOpCo (Section 9.2.1. of the contested decision).

57      As regards the choice of profit level indicator, the Commission considered that, since LuxSCS did not record any sales or assume any risk in connection with the intangible assets, the relevant profit level indicator should have been a mark-up on total relevant costs (recital 550 of the contested decision).

58      As regards the cost base to which a mark-up should be applied in the present case, the Commission considered in essence that LuxSCS merely fulfilled an intermediary function, passing on the costs incurred in relation to the Buy-In Agreement and the CSA to LuxOpCo and transferring a portion of the royalty payments (the Licence Fee) it received from LuxOpCo under the Licence Agreement to A9 and ATI in the amount of those costs (recital 551 of the contested decision).

59      In the light of those findings, in recital 555 of the contested decision, the Commission concluded that LuxSCS’s remuneration should have had two components. The first component should, according to the Commission, have corresponded to the re-charging to LuxOpCo of the costs related to the Buy-In Agreement and the CSA, to which no mark-up should have been applied. The second component should have consisted, according to the Commission, of a mark-up on a cost base consisting solely of the costs incurred for the external services acquired in order to maintain its legal ownership of the intangible assets, to the extent that those costs actually represented actual functions carried out by LuxSCS. According to the Commission, that level of remuneration would have ensured an outcome in line with the arm’s length principle since it would have appropriately reflected LuxSCS’s contributions to the Licence Agreement.

60      As regards the determination of an appropriate mark-up, the Commission noted that, although such an exercise normally required a comparability analysis, it was not possible in the present case to perform a reliable analysis.

61      Instead of a comparability analysis, the Commission considered that it could rely on the conclusions in the 2010 report of the Joint Transfer Pricing Forum (‘the JTPF report’). The Joint Transfer Pricing Forum is a group of experts formed by the Commission in 2002 in order to assist the Commission on transfer pricing matters. According to that report, a mark-up for ‘low value adding intra-group services’ in the range of 3 to 10% was observed by the national tax administrations of the Member States participating in the Joint Transfer Pricing Forum. The mark-up most often observed in practice was 5% on the costs of ‘providing such services’. The Commission therefore considered it appropriate to apply such a mark-up to the external costs incurred by LuxSCS for the maintenance of its legal ownership of the intangible assets.

62      In concluding its primary finding of an economic advantage for the purpose of Article 107(1) TFEU, the Commission indicated that an ‘arm’s length remuneration’ for LuxSCS under the Licence Agreement should have been equal to the sum of the Buy-In and CSA costs incurred by LuxSCS, without a mark-up, plus any relevant costs incurred directly by LuxSCS, to which a mark-up of 5% had to be applied, to the extent that those costs reflected actual functions performed by LuxSCS. In the Commission’s view, that remuneration reflected what an independent party in a position similar to that of LuxOpCo would have been willing to pay for the rights and obligations assumed under the Licence Agreement. In addition, according to the Commission, that level of remuneration would have been sufficient to enable LuxSCS to cover its payment obligations under the Buy-In Agreement and the CSA (recitals 559 and 560 of the contested decision).

63      According to the Commission, since the level of LuxSCS’s remuneration calculated by the Commission was lower than the level of LuxSCS’s remuneration resulting from the transfer pricing arrangement endorsed by the tax ruling at issue, that ruling conferred an advantage on LuxOpCo in the form of a reduction of its taxable base for the purposes of Luxembourg corporate income tax as compared to the income of companies whose taxable profit reflected prices negotiated at arm’s length on the market (recital 561 of the contested decision).

(2)    The subsidiary findings of an advantage

64      In Section 9.2.2 of the contested decision, entitled ‘Subsidiary finding of an economic advantage’, the Commission sets out its subsidiary finding of an advantage, according to which, even if the Luxembourg tax administration were right to have accepted the analysis of LuxSCS’s functions set out in the 2003 transfer pricing report, the transfer pricing arrangement endorsed by the tax ruling at issue was, in any event, based on inappropriate methodological choices that produced an outcome which departed from a reliable approximation of a market-based outcome. The Commission specified that the purpose of its assessment in Section 9.2.2 of the contested decision was not to determine a precise arm’s length remuneration for LuxOpCo but rather to demonstrate that the tax ruling at issue had conferred an economic advantage, since the endorsed transfer pricing arrangement was based on three inappropriate methodological choices which resulted in a lowering of LuxOpCo’s taxable income as compared to companies whose taxable profit reflected prices negotiated at arm’s length on the market.

65      In that context, the Commission made three separate subsidiary findings.

66      In its first subsidiary finding, the Commission stated that LuxOpCo had been inaccurately considered to perform only ‘routine’ management functions and that the profit split method, together with the contribution analysis, should have been applied.

67      In its second subsidiary finding, the Commission found that the choice of operating expenses as a profit level indicator was incorrect.

68      In its third subsidiary finding concerning the advantage, the Commission considered that the inclusion of a ceiling of 0.55% of EU turnover was not appropriate.

(b)    Selectivity of the measure

69      In Section 9.3 of the contested decision, entitled ‘Selectivity’, the Commission set out the reasons why it considered that the measure at issue was selective.

(c)    Identification of the beneficiary of the aid

70      In Section 9.5 of the contested decision, entitled ‘Beneficiary of the aid’, the Commission considered that any favourable tax treatment granted to LuxOpCo had also benefited the Amazon group as a whole by providing it with additional resources, with the result that the group had to be regarded as a single unit benefiting from the contested aid measure.

71      In Section 10 of the contested decision, entitled ‘Recovery’, the Commission stated that, since the aid measure was granted every year in which LuxOpCo’s annual tax declaration was accepted by the Luxembourg tax administration, the Amazon group could not object to the recovery of that aid on the basis of the limitation period. In recitals 639 to 645 of the contested decision, the Commission set out the methodology for recovery.

II.    Procedure and forms of order sought

A.      The procedure in Case T816/17

72      By application lodged at the Registry of the General Court on 14 December 2017, the Grand Duchy of Luxembourg brought the action in Case T‑816/17.

1.      The composition of the formation of the Court and priority treatment

73      By decision of 12 April 2018, the President of the Seventh Chamber of the General Court decided to grant Case T‑816/17 priority over others pursuant to Article 67(2) of the Rules of Procedure of the General Court.

74      By a document lodged at the Court Registry on 11 May 2018, the Grand Duchy of Luxembourg requested that Case T‑816/17 be heard and determined by the Seventh Chamber of the General Court, sitting in extended composition.

75      In accordance with Article 28(5) of the Rules of Procedure, Case T‑816/17 was allocated to the Seventh Chamber (Extended Composition).

76      Since a member of the Seventh Chamber (Extended Composition) of the General Court was unable to sit in the present case, the President of the General Court, by decision of 21 June 2018, designated the Vice-President of the General Court to complete the formation of the Court. Following the appointment of a member of the formation of the Court as a judge at the Court of Justice on 6 October 2020, the most junior Judge for the purpose of Article 8 of the Rules of Procedure abstained from taking part in the deliberations and the present judgment was deliberated by the three judges whose signatures it bears, in accordance with Article 22 of those rules.

2.      The intervention

77      By a document lodged at the Court Registry on 16 April 2018, Ireland sought leave to intervene in Case T‑816/17 in support of the form of order sought by the Grand Duchy of Luxembourg.

78      By order of 29 May 2018, the President of the Seventh Chamber (Extended Composition) granted Ireland leave to intervene.

3.      Applications for confidential treatment

79      By a document lodged at the Court Registry on 14 May 2018, the Grand Duchy of Luxembourg requested that certain data in its application, in some of the annexes thereto and in the defence be treated as confidential vis-à-vis Ireland.

80      By a document lodged at the Court Registry on 6 June 2018, the Grand Duchy of Luxembourg requested that part of the reply be treated as confidential vis-à-vis Ireland.

81      By a document lodged at the Court Registry on 13 September 2018, the Grand Duchy of Luxembourg requested that part of the rejoinder be treated as confidential vis-à-vis Ireland.

82      Following its admission as an intervener, Ireland received only non-confidential versions of the procedural documents covered by the requests for confidential treatment made in its regard by the Grand Duchy of Luxembourg and raised no objection to those requests.

4.      Forms of order sought

83      The Grand Duchy of Luxembourg claims that the Court should:

–        annul the contested decision;

–        in the alternative, annul the contested decision to the extent that it orders the recovery of the aid;

–        order the Commission to pay the costs.

84      The Commission contends that the Court should:

–        dismiss the action as unfounded;

–        order the Grand Duchy of Luxembourg to pay the costs.

85      Ireland claims that the Court should annul the contested decision, in whole or in part, as specified in the form of order sought by the Grand Duchy of Luxembourg.

B.      The procedure in Case T318/18

86      By application lodged at the Court Registry on 22 May 2018, Amazon EU Sàrl and Amazon.com (together ‘Amazon’) brought the action in Case T‑318/18.

1.      The composition of the formation of the Court and priority treatment

87      By decision of 9 July 2018, the President of the Seventh Chamber of the General Court decided to grant Case T‑318/18 priority over others pursuant to Article 67(2) of the Rules of Procedure.

88      On a proposal from the Seventh Chamber of the General Court, the Court decided on 11 July 2018, pursuant to Article 28 of the Rules of Procedure, to refer Case T‑318/18 to a Chamber sitting in extended composition.

89      As a Member of the Seventh Chamber (Extended Composition) was unable to sit in the present case, by decision of 19 July 2018, the President of the General Court designated the Vice-President of the General Court to complete the formation of the Court. Following the appointment of a member of the formation of the Court as a judge at the Court of Justice on 6 October 2020, the most junior Judge for the purpose of Article 8 of the Rules of Procedure abstained from taking part in the deliberations and the present judgment was deliberated by the three judges whose signatures it bears, in accordance with Article 22 of those rules.

2.      Applications for confidential treatment

90      By a document lodged at the Court Registry on 12 July 2018, Amazon requested that part of the application and certain documents annexed thereto be treated as confidential vis-à-vis the public.

3.      Forms of order sought

91      Amazon claims that the Court should:

–        annul Articles 1 to 4 of the contested decision;

–        in the alternative, annul Articles 2 to 4 of the contested decision;

–        order the Commission to pay the costs.

92      The Commission contends that the Court should:

–        dismiss the action;

–        order Amazon to pay the costs in Case T‑318/18.

C.      Joinder of the cases and the oral part of the procedure

93      By documents lodged at the Court Registry on 7 August 2018 and 25 April 2019, the Grand Duchy of Luxembourg applied for Cases T‑816/17 and T‑318/18 to be joined for the purposes of the oral part of the procedure and of the decision closing the proceedings.

94      By documents lodged at the Registry on 10 August 2018 and 21 May 2019, Amazon applied for Cases T‑816/17 and T‑318/18 to be joined for the purposes of the oral part of the procedure and of the decision closing the proceedings.

95      By decision of 14 September 2018, the President of the Seventh Chamber (Extended Composition) of the General Court decided not to join, at that stage of the procedure, Cases T‑816/17 and T‑318/18.

96      By order of 3 October 2019, the President of the Seventh Chamber (Extended Composition) of the General Court decided to join Cases T‑816/17 and T‑318/18 for the purposes of the oral part of the procedure.

97      Acting on a proposal from the Judge-Rapporteur, the Court decided to open the oral part of the procedure and, in connection with the measures of organisation of procedure provided for in Article 89 of the Rules of Procedure, requested the parties to answer written questions. The parties responded to that measure of organisation of procedure within the prescribed period.

98      The parties presented oral argument and responded to the oral questions put by the Court at the hearing held on 5 and 6 March 2020. The parties also presented their views regarding a possible joinder of Cases T‑816/17 and T‑318/18 for the purposes of the decision closing the proceedings, a formal note of which was made by the Court in the minutes of the hearing. The Grand Duchy of Luxembourg, Amazon and Ireland stated that they had no objections to such joinder. The Commission stated that it was not in favour of any joinder of the cases for the purposes of the decision closing the proceedings.

III. Law

99      The actions brought in Cases T‑816/17 and T‑318/18 seek the annulment of the contested decision to the extent that it classifies the tax ruling at issue as State aid for the purposes of Article 107(1) TFEU and to the extent that it orders the recovery of sums that were not collected by the Grand Duchy of Luxembourg from LuxOpCo by way of corporate income tax.

A.      Joinder of Cases T816/17 and T318/18 for the purposes of the decision closing the proceedings

100    In accordance with Article 19(2) of Rules of Procedure, the President of the Seventh Chamber (Extended Composition) of the General Court referred the decision as to whether Cases T‑816/17 and T‑318/18 should be joined for the purposes of the decision closing the proceedings, which fell within his remit, to the Seventh Chamber (Extended Composition) of the General Court.

101    The parties having been heard at the hearing with respect to a possible joinder of the cases, it is appropriate for Cases T‑816/17 and T‑318/18 to be joined for the purposes of the decision which closes the proceedings, on account of the connection between them.

B.      Pleas in law and arguments relied on

102    In support of their actions, the Grand Duchy of Luxembourg and Amazon raise five and nine pleas, respectively, most of which overlap. In its statement in intervention, Ireland sets out its views on four of the five pleas raised by the Grand Duchy of Luxembourg. In essence, the pleas raised by the Grand Duchy of Luxembourg and by Amazon may be presented as follows.

103    In the first place, in the first plea in Case T‑816/17 and in the first to fourth pleas in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon dispute, in essence, the Commission’s primary finding that there was an advantage in favour of LuxOpCo for the purpose of Article 107(1) TFEU.

104    In the second place, in the third complaint in the second part of the first plea in Case T‑816/17 and in the fifth plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon dispute the Commission’s subsidiary findings concerning the existence of a tax advantage in favour of LuxOpCo within the meaning of Article 107(1) TFEU.

105    In the third place, in the second plea in Case T‑816/17 and in the sixth and seventh pleas in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon dispute the Commission’s primary and subsidiary findings in respect of the selectivity of the tax ruling at issue.

106    In the fourth place, in the third plea in Case T‑816/17, the Grand Duchy of Luxembourg submits that the Commission infringed the Member States’ exclusive competence in the area of direct taxation.

107    In the fifth place, in the fourth plea in Case T‑816/17 and in the eighth plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon maintain that the Commission infringed their rights of defence.

108    In the sixth place, in the second part of the first plea and the first complaint in the second part of the second plea in Case T‑816/17 and in the eighth plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon dispute that the 2017 version of the OECD Guidelines, as used by the Commission in adopting the contested decision, is relevant in the present case.

109    In the seventh place, in the fifth plea, relied on in support of the form of order sought in the alternative in Case T‑816/17 and in the ninth plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon call into question the merits of the Commission’s reasoning concerning the recovery of the aid ordered by that institution.

110    In its statement in intervention, Ireland alleges, first, infringement of Article 107 TFEU in that the Commission has not established the existence of an advantage in favour of LuxOpCo; secondly, infringement of Article 107 TFEU in that the Commission failed to prove the selectivity of the measure; thirdly, infringement of Articles 4 and 5 TEU in that the Commission engaged in disguised tax harmonisation; and, fourthly, infringement of the principle of legal certainty in so far as the contested decision orders recovery of the aid.

111    In order to give a useful response to the pleas put forward by the main parties and to the arguments raised by Ireland in its statement in intervention, it is necessary, first of all, to set out certain questions of law which are applicable to all the complaints and pleas in law relied on by the parties (paragraphs 112 to 129 below).

1.      Preliminary remarks

112    According to settled case-law, while direct taxation, as EU law currently stands, falls within the competence of the Member States, they must nonetheless exercise that competence consistently with EU law (see judgment of 12 July 2012, Commission v Spain, C‑269/09, EU:C:2012:439, paragraph 47 and the case-law cited). Thus, intervention by the Member States in matters of direct taxation, even if it relates to issues that have not been harmonised in the European Union, is not excluded from the scope of the rules on the monitoring of State aid (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 142).

113    Accordingly, the Commission may classify a tax measure as State aid so long as the conditions for such a classification are satisfied (see, to that effect, judgments of 2 July 1974, Italy v Commission, 173/73, EU:C:1974:71, paragraph 28, and of 22 June 2006, Belgium and Forum 187 v Commission, C‑182/03 and C‑217/03, EU:C:2006:416, paragraph 81). Member States must exercise their competence in respect of taxation in accordance with EU law (judgment of 3 June 2010, Commission v Spain, C‑487/08, EU:C:2010:310, paragraph 37). Accordingly, they must refrain from adopting any measure, in that context, liable to constitute State aid incompatible with the internal market (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 143).

(a)    Determination of the conditions for the application of Article 107(1) TFEU in the context of national tax measures

114    A measure by which the public authorities grant certain undertakings favourable tax treatment which, although not involving a transfer of State resources, places the recipients in a more favourable financial situation than that of other taxpayers constitutes State aid for the purposes of Article 107(1) TFEU (judgment of 15 March 1994, Banco Exterior de España, C‑387/92, EU:C:1994:100, paragraph 14; see also judgments of 8 September 2011, Paint Graphos and Others, C‑78/08 to C‑80/08, EU:C:2011:550, paragraph 46 and the case-law cited, and of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 145 and the case-law cited).

115    In the case of tax measures, the very existence of an advantage may be established only when compared with ‘normal’ taxation (judgment of 6 September 2006, Portugal v Commission, C‑88/03, EU:C:2006:511, paragraph 56). Accordingly, such a measure confers an economic advantage on its recipient if it mitigates the burdens normally included in the budget of an undertaking and, as a result, without being a subsidy in the strict meaning of the word, is similar in character and has the same effect (judgment of 9 October 2014, Ministerio de Defensa and Navantia, C‑522/13, EU:C:2014:2262, paragraph 22, and of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 146).

116    Consequently, in order to determine whether there is a tax advantage, the position of the recipient as a result of the application of the measure at issue must be compared with his or her position in the absence of the measure at issue (see, to that effect, judgment of 26 April 2018, Cellnex Telecom and Telecom Castilla-La Mancha v Commission, C‑91/17 P and C‑92/17 P, not published, EU:C:2018:284, paragraph 114), and under the normal rules of taxation (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 147).

117    In determining the fiscal position of an integrated company which is part of a group of undertakings, it must be noted at the outset that the pricing of intra-group transactions carried out by that company is not determined under market conditions. That pricing is agreed to by companies belonging to the same group, and is therefore not subject to market forces (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 148).

118    Where national tax law does not make a distinction between integrated undertakings and stand-alone undertakings for the purposes of their liability to corporate income tax, that law is intended to tax the profit arising from the economic activity of such an integrated undertaking as though it had arisen from transactions carried out at market prices. In those circumstances, it must be held that, when examining, pursuant to the power conferred on it by Article 107(1) TFEU, a fiscal measure granted to such an integrated company, the Commission may compare the fiscal burden of such an integrated undertaking resulting from the application of that fiscal measure with the fiscal burden resulting from the application of the normal rules of taxation under national law of an undertaking, placed in a comparable factual situation, carrying on its activities under market conditions (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 149).

119    Furthermore, those findings are supported by the judgment of 22 June 2006, Belgium and Forum 187 v Commission (C‑182/03 and C‑217/03, EU:C:2006:416), concerning Belgian tax law, which provided for integrated companies and stand-alone companies to be treated on equal terms. The Court of Justice recognised in paragraph 95 of that judgment the need to compare a regime of derogating aid with the ‘ordinary tax system, based on the difference between profits and outgoings of an undertaking carrying on its activities in conditions of free competition’ (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 150).

120    In that context, although, through that fiscal measure granted to an integrated company, national authorities have accepted a certain level of pricing for an intra-group transaction, Article 107(1) TFEU allows the Commission to check whether that pricing corresponds to pricing under market conditions, in order to determine whether there is, as a result, any mitigation of the burdens normally included in the budget of the undertaking concerned, thus conferring on that undertaking an advantage within the meaning of that article.

121    It should also be noted that when the Commission uses the arm’s length principle to check whether the taxable profit of an integrated undertaking pursuant to a tax measure corresponds to a reliable approximation of a taxable profit generated under market conditions, the Commission can identify an advantage within the meaning of Article 107(1) TFEU only if the variation between the two comparables goes beyond the inaccuracies inherent in the methodology used to obtain that approximation (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 152).

122    Even though the Commission cannot be formally bound by the OECD Guidelines, the fact remains that those guidelines are based on important work carried out by groups of renowned experts, that they reflect the international consensus achieved with regard to transfer pricing and that they thus have a certain practical significance in the interpretation of issues relating to transfer pricing (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 155).

123    In that context, it should be noted that, if the Commission detects a methodological error in the tax measure under consideration, it cannot be concluded that mere non-compliance with methodological requirements necessarily leads to a reduction in the tax burden. It is further necessary for the Commission to demonstrate that the methodological errors that it identified in the tax ruling concerned do not allow a reliable approximation of an arm’s length outcome to be reached and that they led to a reduction in the taxable profit compared with the tax burden resulting from the application of normal taxation rules under national law to an undertaking placed in a comparable factual situation to the company concerned and carrying out its activities under market conditions. Thus, the mere finding of a methodological error does not in itself suffice, in principle, to demonstrate that a tax ruling conferred an advantage on a specific company and, thus, to establish that there is State aid within the meaning of Article 107 TFEU (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 201).

124    As the Grand Duchy of Luxembourg essentially submits, according to settled case-law, Article 107(1) TFEU defines a measure which reduces the burdens normally imposed on an undertaking in relation to its effects (see judgment of 22 December 2008, British Aggregates v Commission, C‑487/06 P, EU:C:2008:757, paragraph 85 and the case-law cited). The existence of State aid cannot be presumed or inferred from a calculation error which has no impact on the result.

(b)    The burden of proof

125    It must be borne in mind that, in its review of State aid, the Commission must, in principle, provide proof in the contested decision of the existence of the aid (see, to that effect, judgments of 12 September 2007, Olympiaki Aeroporia Ypiresies v Commission, T‑68/03, EU:T:2007:253, paragraph 34, and of 25 June 2015, SACE and Sace BT v Commission, T‑305/13, EU:T:2015:435, paragraph 95). In that context, the Commission is required to conduct a diligent and impartial examination of the measures at issue, so that it has at its disposal, when adopting a final decision establishing the existence and, as the case may be, the incompatibility or unlawfulness of the aid, the most complete and reliable information possible (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 194; see also, to that effect, judgments of 2 September 2010, Commission v Scott, C‑290/07 P, EU:C:2010:480, paragraph 90, and of 3 April 2014, France v Commission, C‑559/12 P, EU:C:2014:217, paragraph 63).

126    It follows that it was for the Commission to show, in the contested decision, that the requirements for a finding of State aid, within the meaning of Article 107(1) TFEU, were met. In that regard, it must be held that, while it is common ground that the Member State has a margin of appreciation in the approval of transfer pricing, that margin of appreciation cannot lead to the Commission being deprived of its power to check that the transfer pricing in question does not lead to the grant of a selective advantage within the meaning of Article 107(1) TFEU. In that context, the Commission must take into account the fact that the arm’s length principle allows it to verify whether the transfer pricing accepted by a Member State corresponds to a reliable approximation of a market-based outcome and whether any variation that may be identified in the course of that examination does not go beyond the inaccuracies inherent in the methodology used to obtain that approximation (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 196).

(c)    The intensity of the review to be conducted by the Court

127    With regard to the intensity of the review to be conducted by the Court in the present case, it should be noted that, as is clear from Article 263 TFEU, the object of an action for annulment is to review the legality of the acts adopted by the EU institutions named therein. Consequently, the analysis of the pleas in law raised in such an action has neither the object nor the effect of replacing a full investigation of the case in the context of an administrative procedure (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 197; see also, to that effect, judgment of 2 September 2010, Commission v Deutsche Post, C‑399/08 P, EU:C:2010:481, paragraph 84).

128    In the field of State aid, it must be recalled that State aid, as defined in the FEU Treaty, is a legal concept which must be interpreted on the basis of objective factors. For that reason, the Courts of the European Union must, in principle, having regard both to the specific features of the case before them and to the technical or complex nature of the Commission’s assessments, carry out a comprehensive review as to whether a measure falls within the scope of Article 107(1) TFEU (judgments of 4 September 2014, SNCM and France v Corsica Ferries France, C‑533/12 P and C‑536/12 P, EU:C:2014:2142, paragraph 15; of 30 November 2016, Commission v France and Orange, C‑486/15 P, EU:C:2016:912, paragraph 87; and of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 198).

129    As to whether a method for determining transfer pricing of an integrated company complies with the arm’s length principle, it should be borne in mind that, as has already been indicated above, when using that tool in carrying out its assessment under Article 107(1) TFEU, the Commission must take into account its approximate nature. The purpose of the Court’s review is therefore to verify whether the errors identified in the contested decision, and on the basis of which the Commission found there to be an advantage, go beyond the inaccuracies inherent in the application of a method designed to obtain a reliable approximation of a market-based outcome (judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 199).

2.      Pleas and arguments challenging the primary finding of an advantage

130    As noted in paragraph 103 above, by the second part of the first plea in Case T‑816/17 and by the first to fourth pleas in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon submit that the Commission infringed Article 107(1) TFEU when it concluded that there was an advantage for LuxOpCo in the context of the primary finding of an advantage in Section 9.2.1 of the contested decision (recitals 409 to 561 of the contested decision). More specifically, by those pleas and arguments, the Grand Duchy of Luxembourg and Amazon seek to challenge the Commission’s reasoning in recitals 394, 395 and 401 to 579 of the contested decision, according to which the implementation of the tax ruling at issue during the relevant period led to a reduction in LuxOpCo’s remuneration, and thus in its tax burden, compared to that which it should have received in the absence of that decision, if it had been treated like any other taxpayer company in a comparable situation. By their arguments concerning the primary finding of an advantage, the Grand Duchy of Luxembourg and Amazon seek to call into question, inter alia, the Commission’s finding that LuxSCS should have been considered to be the tested party in the context of the application of the TNMM. They also seek to dispute the accuracy of the Commission’s application of the TNMM to LuxOpCo.

131    As pointed out in paragraph 110 above, in its statement in intervention, Ireland puts forward arguments in support of the first plea raised by the Grand Duchy of Luxembourg.

132    In that context, Ireland expresses its views on numerous questions of law concerning the interpretation of the concept of the ‘arm’s length principle’, as applied by the Commission in the present case and in certain recent State aid cases in tax matters. In particular, Ireland submits that the case-law of the EU judicature, namely the judgment of 22 June 2006, Belgium and Forum 187 v Commission (C‑182/03 and C‑217/03, EU:C:2006:416), ‘does not say that the Member States are obliged to apply [the arm’s length principle]’. According to Ireland, that case-law also provides no basis for the obligation imposed on Luxembourg to apply (the Commission’s version of) the arm’s length principle in Luxembourg national law. Lastly, Ireland submits that, in the judgment of 22 June 2006, Belgium and Forum 187 v Commission (C‑182/03 and C‑217/03, EU:C:2006:416), the Court of Justice did not identify an arm’s length principle specific to EU law regardless of what is provided in national law.

(a)    Admissibility of certain arguments put forward by Ireland concerning the existence of an advantage

133    The Commission contends that the arguments put forward by Ireland in support of the first plea raised by the Grand Duchy of Luxembourg are inadmissible. According to the Commission, Ireland’s arguments claim that the Commission misinterpreted the concept of advantage for the purpose of Article 107(1) TFEU by using an inappropriate criterion, namely a ‘sui generis’ arm’s length principle, whereas, in fact, by its first plea, the Grand Duchy of Luxembourg submits rather that the Commission misapplied the arm’s length principle.

134    In that respect, it should be noted that, although the third paragraph of Article 40 of the Statute of the Court of Justice of the European Union and Article 142(3) and Article 145(2)(b) of the Rules of Procedure of the General Court do not preclude an intervener from submitting arguments that are new or differ from those of the party he or she supports, as otherwise his or her intervention would be limited to restating the arguments put forward in the application, it cannot be held that those provisions allow him or her to alter or distort the context of the dispute defined by the application by raising new pleas in law (see judgment of 20 September 2019, Le Port de Bruxelles and Région de Bruxelles-Capitale v Commission, T‑674/17, not published, EU:T:2019:651, paragraph 44 and the case-law cited).

135    In other words, those provisions give the intervener the right to set out, independently, arguments as well as pleas, in so far as they support the form of order sought by one of the main parties and are not entirely unconnected with the issues underlying the dispute, as established by the applicant and defendant, which would result in a change in the subject matter of the dispute (see judgment of 20 September 2019, Le Port de Bruxelles and Région de Bruxelles-Capitale v Commission, T‑674/17, not published, EU:T:2019:651, paragraph 45 and case-law cited).

136    In the present case, it should be noted that Ireland’s arguments concern, in essence, the legal basis relied on by the Commission as regards the obligation imposed on the Grand Duchy of Luxembourg to apply the arm’s length principle. Ireland therefore calls into question the sources of law of that principle, as applied by the Commission in the contested decision. In addition, Ireland’s arguments relate to the interpretation of the content of that principle and not to its application by means of a transfer pricing method.

137    It is common ground that the arm’s length principle, as applicable in the present case, may be inferred from Article 164(3) of the LIR. That is apparent, in particular, from recital 241 of the contested decision, and that conclusion has not been called into question by the parties. The first plea raised by the Grand Duchy of Luxembourg does not relate to the question of the source of law of that principle or to questions of interpretation of that principle. In fact, by its first plea, the Grand Duchy of Luxembourg alleges errors in the Commission’s application of certain methods for determining transfer pricing in its reasoning concerning the existence of an advantage, bearing in mind that those methods make it possible to determine, ultimately, whether a royalty corresponds to an arm’s length outcome.

138    It follows that the arguments put forward by Ireland in support of the first plea of the Grand Duchy of Luxembourg are not connected with the considerations on which its first plea is based. Accordingly, they must be rejected as inadmissible.

(b)    The merits of the pleas in law and arguments of the Grand Duchy of Luxembourg and Amazon concerning the primary finding of an advantage

139    In addition to the matters set out in paragraph 130 above, it should be noted that, in the context of the first part of the first plea in Case T‑816/17 and the first plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon call into question the merits of the Commission’s refusal to apply the CUP method in the context of an ex post analysis carried out on the basis of comparable agreements submitted to the Commission by Amazon.com.

140    In the first and second complaints in the second part of the first plea in Case T‑816/17 and the second plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon submit that the functional analysis performed by the Commission as part of its application of the TNMM is incorrect in that it found that LuxSCS was the least complex party, and that the Commission’s application of the TNMM is based on incorrect methodological choices.

141    In the second part of the first plea in Case T‑816/17 and the third plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon maintain that, in its primary analysis, the Commission selected from among the testimony in the United States proceedings mentioned in paragraph 14 above in an arbitrary and biased fashion.

142    In the third part of the first plea in Case T‑816/17 and the sixth part of the second plea and the fourth plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon assert that the result obtained by the Commission in the contested decision deviates from a reliable approximation of an arm’s length outcome.

143    Thus, in essence, the arguments put forward by the Grand Duchy of Luxembourg and Amazon against the primary finding of an advantage seek to challenge, first, the Commission’s rejection of the use of the CUP method and, secondly, the Commission’s application of the TNMM.

144    As regards the arguments challenging the Commission’s rejection of the use of the CUP method, it should be noted that it is common ground that the tax ruling at issue did not apply that method. Even if that method was examined in the 2003 transfer pricing report, provided to the tax authorities in support of the request for a tax ruling, it was not used in the letter of 23 October 2003 by which Amazon sought approval of the method of calculating the royalty (see paragraph 9 above). As is apparent in particular from recital 542 of the contested decision, in its analysis aimed at demonstrating the existence of an advantage for the purpose of Article 107(1) TFEU, the Commission relied solely on the TNMM. However, the Commission’s assessments intended to rule out the applicability of the CUP method (recitals 521 to 538 of the contested decision) are not capable of demonstrating the existence of the first condition laid down in Article 107(1) TFEU. Given that it is for the Commission to demonstrate the existence of an advantage (see paragraphs 125 to 126 above) and in view of the fact that the Commission’s assessments intended to rule out the applicability of the CUP method do not allow such a demonstration, it is not useful to discuss the applicants’ arguments and pleas concerning the CUP method.

145    As regards the arguments challenging the merits of the Commission’s assessments concerning the application of the TNMM by that institution (see paragraphs 146 to 297 below), it will be necessary, first, to identify the relevant version of the OECD Guidelines on transfer pricing (see paragraphs 146 to 155 below). Secondly, it will be necessary to ascertain whether the Grand Duchy of Luxembourg and Amazon are entitled to claim that the Commission made errors when applying the TNMM in the contested decision, which invalidated its primary finding relating to the advantage (see paragraphs 156 to 297 below).

(1)    Temporal relevance of certain OECD Guidelines used by the Commission for the purposes of applying the TNMM

146    In order to demonstrate the existence of an advantage, the Commission, in Section 9.2 of the contested decision, applied a series of OECD Guidelines on transfer pricing in different versions thereof.

147    In the first complaint in the second part of its second plea, the Grand Duchy of Luxembourg submits, in essence, that, in the present case, it is the economic context and regulatory framework prevailing in 2003 which must be taken into account. Leaving aside the fact that, at the time of the adoption of the tax ruling at issue, in 2003, as at the time of its last extension in 2010, the OECD Guidelines constituted merely indicative, non-binding guidance for the Luxembourg authorities, the only OECD guidelines available at the time of the adoption of the tax ruling were the 1995 OECD Guidelines. In the contested decision, the Commission referred, however, to the 2010 and 2017 versions of the OECD Guidelines, which amounts to an inappropriate application ratione temporis of the reference framework, which should be determined on the basis of the facts and the pricing methods which existed at the time of the adoption of the measures at issue.

148    Amazon adds that the 2010 and 2017 OECD Guidelines made several significant changes compared to their 1995 version, such as the introduction of the method for analysing the Development, Enhancement, Maintenance, Protection and Exploitation functions (‘the DEMPE functions’). Amazon disputes, in particular, the relevance of the Commission’s application of that method, since it did not appear until after the date of adoption of the tax ruling at issue, that is to say, in the 2017 version of the OECD Guidelines.

149    The Commission disputes those arguments.

150    It submits, first, that the contested decision does not apply the OECD Guidelines as if they constituted binding standards, but rather as a tool to assist it in applying the test laid down by the Court of Justice in paragraph 95 of the judgment of 22 June 2006, Belgium and Forum 187 v Commission (C‑182/03 and C‑217/03, EU:C:2006:416). According to the Commission, contrary to what the Grand Duchy of Luxembourg appears to argue, the Luxembourg tax administration regularly relied on those guidelines in interpreting the arm’s length principle, with the result that the OECD Guidelines remain relevant in the present case.

151    The Commission then adds that all the findings made in the contested decision are based on the 1995 OECD Guidelines and that the 2010 and 2017 OECD Guidelines are referred to only when those subsequent versions clarify the 1995 version of the Guidelines without, however, amending them.

152    In the present case, it is apparent from a number of footnotes to the contested decision that the Commission based, if only in part, its assessments relating to the existence of an advantage within the meaning of Article 107(1) TFEU not only on the 1995 version of the OECD Guidelines, but also on the 2010 and 2017 versions of those guidelines. It must be noted that the1995, 2010 and 2017 versions of the OECD Guidelines differ on several points, and to various extents. Those differences range from mere clarifications which have no impact on the substance of the earlier versions to entirely new elaborations, namely recommendations which were not contained, including implicitly, in the earlier versions. One of the entirely new elaborations of the OECD Guidelines which appeared only in the 2017 version is the DEMPE functional analysis method (see paragraph 148 above). In the context of the primary finding of an economic advantage, the Commission focused inter alia on that method of analysis.

153    In that respect, it should be borne in mind that, as is apparent from Article 1 and – implicitly – from inter alia recitals 394 and 620 of the contested decision, the measure at issue, as identified by the Commission, is the tax ruling at issue and the subsequent acceptance of LuxOpCo’s annual corporate income tax declarations based on that ruling. During the relevant period, LuxOpCo made its tax declarations on the basis of the method of calculation endorsed in the tax ruling at issue and that decision was extended in 2006 and in 2010.

154    In the light of those factors, it must be held that the Commission was entitled to base its findings concerning the existence of an advantage on the guidelines – which are, moreover, merely a non-binding tool – set out in the 1995 OECD Guidelines. However, in so far as the Commission relied on the 2010 version of the OECD Guidelines, that version is not relevant, unless it merely provides a useful clarification, without further elaboration, of the guidelines already set out in 1995. As for the 2017 version of the OECD Guidelines, since they were published after the relevant period and in so far as the recommendations contained therein have evolved significantly in relation to the 1995 version of the OECD Guidelines, those guidelines are not relevant in the present case.

155    In particular, the DEMPE functional analysis method cannot be regarded as relevant from a temporal point of view, in the present case, since it is a tool which was set out only in the 2017 version of the OECD Guidelines.

(2)    Alleged errors committed by the Commission in the application of the TNMM in the contested decision

156    As noted in paragraph 9 above, the Grand Duchy of Luxembourg and Amazon dispute a whole series of assessments made by the Commission concerning the application of the TNMM in the context of the primary finding of an advantage.

157    It will be recalled that the TNMM is an indirect transfer pricing method. As described in paragraph 3.26 of the 1995 version of the OECD Guidelines, that method consists of determining, relative to an appropriate base, the net profit made by a taxpayer in a controlled transaction or controlled transactions which are closely linked or continuous. To determine that appropriate base, it is necessary to choose a profit level indicator, such as costs, sales or assets. The net profit indicator obtained by the taxpayer in a controlled transaction must be determined by reference to the net profit indicator earned by the same taxpayer or an independent undertaking in comparable uncontrolled transactions.

158    As is apparent from paragraph 3.26 of the 1995 version of the OECD Guidelines, the TNMM involves identifying a party to the transaction for which a profit level indicator is tested, for example a mark-up on costs. That party is referred to as the ‘tested party’. That is the party whose so-called ‘arm’s length’ mark-up must be determined. As a general rule, the tested party is the party to which a transfer pricing method can be applied in the most reliable manner and for which the most reliable comparables can be found.

159    The tested party is chosen on the basis of a functional analysis of the parties to the intra-group transaction. According to paragraph 3.43 of the 1995 version of the OECD Transfer Pricing Guidelines, the tested party will most often be the party which has the less complex functional analysis. According to an understanding already existing at the time when the 1995 Guidelines applied, the functional analysis usually involves examining the functions performed by an entity, the assets held and the risks borne.

160    In addition, it should be noted that the TNMM is considered to be an appropriate method for testing the arm’s length remuneration of a party which makes no unique or valuable contributions in connection with the transaction which is the subject of the transfer pricing analysis.

161    In the present case, the Grand Duchy of Luxembourg and Amazon do not dispute, as such, the Commission’s choice of the TNMM. They dispute only that the Commission applied that method correctly. First, the Grand Duchy of Luxembourg and Amazon challenge the Commission’s functional analysis and the choice of LuxSCS as the tested party for the purposes of applying the TNMM; secondly, the calculation of LuxSCS’s remuneration, namely the choice of profit level indicator and the mark-up rate used by the Commission in applying the TNMM; and, thirdly, the reliability of the result obtained.

(i)    The functional analysis and the Commission’s choice of LuxSCS as the tested party

162    Recitals 409 to 561 of the contested decision, namely those relating to the primary finding of an advantage, seek to show, in essence, that, in the present case, the Luxembourg tax authorities should have applied the TNMM by using LuxSCS as tested party since it is, in the light of the functional analysis carried out by the Commission, the ‘least complex’ party. It is also apparent from those recitals that, according to the Commission, if the Luxembourg tax authorities had applied the TNMM by using LuxSCS as the tested party, LuxOpCo’s remuneration would have been higher than the remuneration determined pursuant to the tax ruling at issue. According to the Commission, consequently, the application of the TNMM by using LuxSCS as the tested party would have resulted in a lower royalty for LuxSCS and thus a higher remuneration for LuxOpCo.

163    In the second part of the first plea in Case T‑816/17 and the second plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon dispute the Commission’s functional analysis. They claim, inter alia, that the functions performed by LuxSCS as well as the assets used and the risks assumed by it were minimised by the Commission. In their view, LuxSCS held the intangible assets and performed unique and valuable functions and could not, therefore, be used as the tested entity for the purposes of the Commission’s application of the TNMM.

164    In that context, it must be pointed out that, by their arguments concerning the primary finding of an advantage, the Grand Duchy of Luxembourg and Amazon do not call into question the merits of the Commission’s choice of the TNMM as being the appropriate method for assessing the arm’s length nature of the royalty. When calling into question the Commission’s findings concerning the functional analysis of LuxSCS in Section 9.2.1.1 of the contested decision, the Grand Duchy of Luxembourg and Amazon therefore seek, in essence, to challenge the Commission’s assertion that LuxSCS should have been regarded by the Luxembourg tax authorities as being the ‘least complex’ party and therefore the tested party in the context of the application of the TNMM.

165    In order to respond to those arguments of the Grand Duchy of Luxembourg and Amazon that the Commission was not entitled to conclude that the Luxembourg tax authorities should have applied the TNMM by using LuxSCS as the tested party, it is not necessary to verify the validity of the functional analysis of LuxOpCo. Since the Commission sought to apply the TNMM by using LuxSCS as the tested party, it is sufficient to verify the validity of the functional analysis of LuxSCS, as set out in Section 9.2.1.1 of the contested decision, and to ascertain whether, in the light of that analysis, it was possible to apply the TNMM to LuxSCS in a sufficiently reliable manner.

166    In that regard, it should be noted as a preliminary point that, according to paragraph 3.43 of the 1995 version of the OECD Guidelines, the party to which the TNMM is applied ‘should be the enterprise for which reliable data on the most closely comparable transactions can be identified’, that this ‘will often entail selecting the associated enterprise that is the least complex of the enterprises involved in the controlled transaction and that does not own valuable intangible property or unique assets’ and that, ‘however, the choice may be restricted by limited data availability’. According to that paragraph, in other words, although, as a general rule, the entity for which there is the most reliable evidence for the purposes of identifying comparables is often the ‘least complex’ entity, the purpose of the application of the TNMM is not necessarily to make that application dependent on the identification of the ‘least complex’ entity. Rather, what matters in the application of that method is, first, to have identified the party for which the most reliable data can be found and, secondly, whether the TNMM can be applied reliably to that party.

167    As a result of what has just been stated in paragraph 166 above, and as is apparent, in particular, from paragraphs 3.26, 3.28, 3.29, 3.34 and 3.43 of the 1995 version of the OECD Guidelines, the application of the TNMM necessarily entails finding reliable data for the comparison with the tested party. Thus, the entirety of the assessments concerning the functional analysis, the assessment of functions, the considerations relating to assets and risks borne, and all of the considerations concerning the ‘unique and valuable’ nature of the assets used, are merely criteria to be taken into account in the choice of the tested party in order to ensure a reliable result.

168    It is in the light of those considerations that the complaints challenging the functional analysis of LuxSCS carried out by the Commission and its conclusion that LuxSCS should have been the tested entity must be examined.

169    In that regard, it should be borne in mind that, in Section 9.2.1.1.1 of the contested decision (recitals 419 to 429 of that decision), the Commission described the functions performed by LuxSCS in the context of the controlled transaction.

170    In essence, as summarised in recital 418 of the contested decision, the Commission’s analysis is based on the following three main assertions. First, the Commission found that LuxSCS had not performed ‘active’ functions in connection with the development, enhancement, management and exploitation of the intangible assets, that it was not entitled to do so because of the exclusive licence granted to LuxOpCo, nor did it have the capacity to do so. Next, the Commission stated that, in its view, LuxSCS had not used assets in relation to those intangible assets, but had merely passively held legal ownership of the intangible assets and a licence to them under the CSA. Lastly, it noted that LuxSCS had neither assumed nor controlled the risks associated with those activities, nor did it have the operational and financial capacity to do so.

171    In recital 429 of the contested decision, the Commission concluded that, during the relevant period, the only functions that could actually be regarded as having been performed by LuxSCS were functions linked to the maintenance of its ‘legal ownership’ of the intangible assets, even though those functions were performed under LuxOpCo’s control. In fact, as is apparent from recitals 418 and 430 of the contested decision, LuxSCS merely held the intangible assets ‘passively’.

172    Next, in Section 9.2.1.1.2 of the contested decision, entitled ‘Assets used by LuxSCS’, and in particular in recital 430 of that decision, the Commission, in essence, again asserted that LuxSCS was merely the passive owner of the intangible assets. In recital 431 of the contested decision, the Commission disputed the argument that LuxSCS had used the intangible assets by licensing them to LuxOpCo. In recital 432 of the contested decision, it reiterated its view that, in any event, LuxSCS did not have the capacity actually to use the intangible assets.

173    Lastly, in Section 9.2.1.1.3 of the contested decision (recitals 436 to 446 of that decision), entitled ‘Risks assumed by LuxSCS’, the Commission analysed the risks borne by LuxSCS in so far as those risks were relevant in the context of the Licence Agreement. In recital 446 of that decision, it concluded in that regard that LuxSCS could not be regarded as having effectively assumed the risks associated with the development, enhancement, management and exploitation of Amazon’s intangible assets and that it did not have the financial capacity to assume such risks.

174    In addition, in Section 9.2.1.4.1 of the contested decision, entitled ‘The tested party should be LuxSCS’, the Commission stated, in essence, that it was necessary to avoid confusion between the complexity of the assets held and the complexity of the functions performed by the parties to the intra-group transaction concerned (recital 546 of the contested decision). Next, it asserted that there was no basis for the assumption that an associated group company that licenses an intangible asset to another company in the group performs more complex functions than the latter company merely because it legally owns a complex asset (recital 546 of the contested decision). Consequently, in its view, the Luxembourg tax administration should not have accepted Amazon’s claim that the mere legal ownership of the intangible assets was in itself a ‘unique contribution’. Rather, it should have required a functional analysis demonstrating that LuxSCS performed ‘unique and valuable functions’ (recital 547 of the contested decision). Lastly, according to the Commission, although LuxSCS was the legal owner of the intangible assets during the relevant period, the functional analysis carried out in Section 9.2.1.1 of the contested decision demonstrates that LuxSCS performed no ‘active’ or critical function in relation to the development, enhancement, maintenance or exploitation of those assets.

175    The Commission’s assessments relating to LuxSCS’s functions overlap to a large extent with those relating to the assets used by LuxSCS. That is also true of the arguments raised by the Grand Duchy of Luxembourg and Amazon challenging those assessments. It is therefore appropriate to examine those arguments together, and then examine those relating to the risks assumed by LuxSCS, in order to examine whether the Commission correctly considered that LuxSCS should be regarded as the tested entity.

–       The functions and assets of LuxSCS

176    The Grand Duchy of Luxembourg and Amazon dispute the Commission’s assertions concerning LuxSCS’s functions. By contrast, as regards LuxSCS’s intangible assets, they agree that they were ‘unique and valuable’, without however taking care to define those terms.

177    First, the Grand Duchy of Luxembourg and Amazon criticise the Commission for failing to take account of the fact that the 1995 version of the OECD Guidelines stated that the party holding the intangible assets would typically not be the tested party for the application of the TNMM. In that regard, the Grand Duchy of Luxembourg and Amazon stress the fact that LuxSCS held unique and valuable intangible assets. They submit that the technology made available by LuxSCS played a central role in the development of the Amazon group’s business in Europe. Those intangible assets were essential for all of the Amazon group’s activities in Europe. In addition, the Grand Duchy of Luxembourg emphasises that, by licensing the intangible assets to LuxOpCo, LuxSCS enabled LuxOpCo to benefit from the development activities carried out by ATI and A9 in the United States and allowed it to make optimal use of those assets. Consequently, LuxOpCo had to remunerate LuxSCS not only for its contributions, but also indirectly the United States entities of the Amazon group for their contributions.

178    Secondly, the Grand Duchy of Luxembourg and Amazon dispute the Commission’s approach of distinguishing between ‘active’ and ‘passive’ functions and taking only the latter into account for the purposes of the functional analysis. In that context, they also criticise the Commission for not taking into account, in the analysis of functions, the fact that LuxSCS made the intangible assets available to LuxOpCo in the context of the controlled transaction. Amazon adds that making the intangible assets available by granting a licence to LuxOpCo constitutes an exploitation of those assets by LuxSCS, as envisaged by paragraph 6.32 of the 2017 version of the OECD Guidelines.

179    Thirdly, the Grand Duchy of Luxembourg and Amazon submit that LuxSCS performed unique and valuable functions, contrary to what is claimed by the Commission. In this context, they indicate in particular that, through its participation in the CSA, LuxSCS contributed to the continued development of the intangible assets, even though it had no employees. Again according to the Grand Duchy of Luxembourg and Amazon, the contributions made by the United States entities ATI and A9, namely the development and continued enhancement of the intellectual property, must be attributed to LuxSCS or regarded as part of LuxSCS’s contributions. In their view, LuxSCS thus performed ‘unique and valuable’ functions which justify treating it as the most complex part of the transaction. Amazon submits, moreover, that whether or not LuxSCS had the capacity to operate an electronic commerce business entirely on its own without licensing the intangible assets to another entity is irrelevant to the assessment of the uniqueness of LuxSCS’s functions.

180    The Commission disputes those arguments.

181    The Commission insists that LuxSCS merely held the intangible assets ‘passively’ and that it did not actually use them. It submits that the mere ownership of a unique and valuable intangible asset is not sufficient for that entity to be regarded as complex. Nor, in the present case, is it sufficient to justify the attribution to LuxSCS of almost all the profits generated by LuxOpCo, even if none of LuxOpCo’s activities could be carried out without access to the intangible assets. Following the conclusion of the Licence Agreement, LuxSCS was no longer entitled to use the assets, nor did it have the capacity to do so. Only LuxOpCo used the intangible assets in the course of its business. In that context, the Commission also submits that LuxSCS did not have any employees and did not have the capacity to carry out functions in relation to the development, enhancement and exploitation of the intangible assets.

182    Moreover, according to the Commission, the Grand Duchy of Luxembourg and Amazon wrongly refer to the contributions of Amazon group entities located in the United States (see paragraph 179 above), since those entities are not concerned by the Licence Agreement and act independently of LuxSCS. Any functions those entities may have performed in relation to the intangible assets, the circumstance that Amazon.com was directing LuxSCS or LuxOpCo or the characteristics of the Buy-In Agreement and the CSA are therefore irrelevant to the functional analysis of LuxSCS. The development functions performed by ATI and A9 cannot therefore be attributed to LuxSCS, since the various parties to the CSA act on their own behalf and at their own risk. The Commission argues that, in any event, the Buy-In Agreement and the CSA already fixed the arm’s length remuneration for the functions performed by ATI and A9 in relation to the intangible assets. Any other intra-group transaction between the US entities and LuxOpCo in relation to the intangible assets – the existence of which has not, in any event, been demonstrated by either the Grand Duchy of Luxembourg or Amazon – cannot justify the payment of LuxOpCo’s residual profits to LuxSCS.

183    In that regard, in the first place, it must be borne in mind that, as has already been noted in paragraph 166 above, according to paragraph 3.43 of the 1995 version of the OECD Guidelines, the ‘associated enterprise to which the transactional net margin method is applied should be the enterprise for which reliable data on the most closely comparable transactions can be identified’, and that ‘this will often entail selecting the associated enterprise that is the least complex of the enterprises involved in the controlled transaction and that does not own valuable intangible property or unique assets’. The concept of ‘valuable’ or ‘unique assets’ is not expressly explained in the 1995 version of the OECD Guidelines.

184    It follows from paragraph 3.43 of the 1995 version of the OECD Guidelines that those guidelines recommended that the party owning unique and valuable assets should not be used as the tested party for the purposes of applying the TNMM; rather another party to the controlled transaction should be used. The logic underlying paragraph 3.43 is that, in general, it is more complicated to find reliable comparables for the purpose of examining the party to the controlled transaction that owns unique and valuable intangible assets. That understanding was also apparent from paragraph 6.26 of the 1995 version of the OECD Guidelines. According to that paragraph, in cases involving highly valuable intangible property, it may be difficult to identify comparable transactions between independent undertakings. It is apparent from that same paragraph that the mere ownership of unique or valuable intangible assets makes the identification of comparables more difficult. It should be noted that paragraph 6.26 of the 1995 version of the OECD Guidelines is based on the premiss that an intangible asset may be deemed to be ‘unique’ when there is no comparable for that asset. An intangible asset is ‘valuable’ if it enables the generation of substantial revenue. Furthermore, it should be noted that that understanding corresponds to the definition of the concept set out in paragraph 6.17 of the 2017 version of the OECD Guidelines. It follows from that paragraph that ‘single and valuable intangible assets’ are those (i) that are not comparable to intangible assets used by parties to potentially comparable transactions, and (ii) the use of which in business operations is expected to yield greater future economic benefits than would be expected in the absence of those intangible assets.

185    In the present case, first, it is common ground that LuxSCS held the rights over the Amazon group’s intangible assets in Europe and that it made those assets available to LuxOpCo under the Licence Agreement.

186    In that respect, it should be noted, in addition to the elements set out in paragraphs 4 and 5 above that, under the assignment agreement concluded between ATI and LuxSCS on 1 January 2005, which forms part of the Buy-In Agreement, the ownership of some of those assets was assigned to LuxSCS (see Sections 3.1 and 3.2 of that agreement), in particular and essentially, the internet domain names in Europe, such as amazon.co.uk, amazon.fr and amazon.de.

187    Next, under the licence agreement concluded between ATI and LuxSCS on 1 January 2005, LuxSCS received the right to use, in Europe, most of the Amazon group’s intangible assets already existing in 2005, namely technology, inventions, patents, trade marks, customer-related rights, etc., without this being an exclusive licence right for LuxSCS.

188    In addition, under Sections 6.2(a) and 6.3(a) of the CSA, LuxSCS held a non-exclusive licence to the intellectual property of A9 and ATI developed after 2005, as well as the ownership of derivative works developed after 2005 from the intangible assets legally owned by LuxSCS.

189    Lastly, LuxSCS also entered into intellectual property assignment and licensing agreements with the EU local affiliates, under which it received the registered trade marks and the intellectual property rights held by those companies in respect of the EU websites.

190    Accordingly, LuxSCS held rights to intangible assets including the following three types of intellectual property: technology, marketing-related intangible assets and customer data. The technology included a complete range encompassing all aspects of the Amazon group’s business and, in particular, technologies for that group’s software platform, the appearance of the websites, catalogues, order processing, logistics, search and navigation functionalities, customer service and personalisation functions.

191    Secondly, it should be noted that, although the Commission maintains that LuxSCS did not perform ‘unique and valuable functions’ in relation to the intangible assets, it does not dispute the ‘unique and valuable nature of the intangibles’ held by LuxSCS and made available to LuxOpCo in the context of the controlled transaction.

192    In particular, the Commission has not disputed, in a substantiated manner, Amazon’s assertion that the technology was unique, since no comparables existed and that it played an essential role in the various aspects of the Amazon group’s commercial activities in Europe and therefore enabled the generation of significant revenue. Moreover, it has not been called into question that, as Amazon submits, the group’s commercial activities could not have acquired such a large scale and achieved such success in Europe – as indeed in other regions of the world – without the technology. The Grand Duchy of Luxembourg’s assertion that, during the relevant period, the Amazon group relied on its technology, which was ‘at the heart of its business model’, as a means of competitive differentiation, in that it was precisely that technology that constituted the unique and valuable contributor that enabled (and still enables) the Amazon group to continue to be competitive in a highly competitive environment characterised by narrow margins, is also convincing. In addition, it is apparent from recital 338 of the contested decision that even some of the Amazon group’s competitors admit that, because it was ‘very aggressive in investing in technology’, the Amazon group’s retail sales platform ‘today presents a hard-to-match competitive advantage’. The technology was therefore an asset for which no comparable existed.

193    In that regard and, moreover, it must be pointed out that it is not necessary to examine the Commission’s arguments that the technology alone was not sufficient to carry out the Amazon group’s commercial activities in Europe and that the human functions performed by LuxOpCo employees were also important. Those arguments, even if well founded, do not call into question the finding that the technology played an essential role in the Amazon group’s commercial activities in Europe and thus constituted a unique and valuable asset.

194    As regards the trade marks registered in Europe, it must be noted that, on the date on which LuxSCS was transferred those assets, which already benefited from the international reputation of the Amazon group, it is not apparent from the documents before the Court that there were comparable assets on the European market. It must therefore be held that the trade marks in question were unique. It is common ground that their use enabled the generation of substantial revenue in Europe. Those trade marks were therefore also ‘valuable’. The customer data also did not have comparables and enabled the generation of significant profits. It must therefore be found that those intangible assets were also unique and valuable.

195    In those circumstances, having regard to paragraph 3.43 of the 1995 version of the OECD Guidelines and in view of the fact that the intangible assets of the Amazon group and, in particular, the technology constituted unique and valuable assets implemented by LuxSCS in the context of the controlled transaction, the Luxembourg tax authorities could not be criticised for having considered, like the authors of the 2003 transfer pricing report, that it was correct, according to the 1995 version of the OECD Guidelines, to choose a company other than LuxSCS as the tested party. Furthermore, although, as the Commission suggested in footnote 681 of the contested decision, according to the 2017 version of the OECD Guidelines, a passive owner cannot be the most complex party and may therefore be the tested party in the context of the application of the TNMM, it must be pointed out that that was not the case in the relevant period, which must be examined in the present case solely in the light of the 1995 version of the OECD Guidelines.

196    In paragraph 83 of the defence in Case T‑318/18, the Commission appears to wish to insist on the fact that, according to paragraph 3.43 of the 1995 version of the OECD Guidelines, it is only ‘often’ that the choice of the tested party will mean choosing the associated enterprise which is the ‘least complex of the enterprises involved in the controlled transaction and that does not own valuable intangible property or unique assets’, without however that being an absolute rule in that respect. In so far as the Commission seeks to assert that the rule contained in paragraph 3.43 of the 1995 version of the OECD Guidelines is not an absolute rule, but a rule which may be disregarded if justified by particular circumstances relating to the controlled transaction concerned, it must be stated that the Commission did not explain, in the contested decision, why that recommendation should, in the present case, be rejected. The Commission has not shown that the Luxembourg tax authorities should have departed from the rule contained in paragraph 3.43 of the 1995 version of the OECD Guidelines on account of a specific feature of the controlled transaction concerned in the present case, namely the Licence Agreement.

197    In the second place, and in any event, it should be noted that the Commission erroneously considered that, apart from the functions of maintaining its intellectual property, LuxSCS did not perform any ‘active and critical’ functions in relation to the intangible assets (see recitals 420 and 548 of the contested decision) or even ‘any functions that add value to the [intangible assets]’ (see recital 526 of that decision).

198    First, as regards the distinction drawn by the Commission between ‘passive’ ownership (recitals 418 and 430 of the contested decision) and ‘active’ ownership of intangible assets, and between ‘active’ and ‘passive’ functions (recital 548 of the contested decision), it should be noted that, as the Grand Duchy of Luxembourg and Amazon submit, the relevant OECD Guidelines in the present case do not provide for such a distinction.

199    The 1995 version of the OECD Guidelines merely states, in paragraph 1.20, that, in general, in determining the arm’s length nature of remuneration established in the context of a controlled transaction, it is necessary to examine whether that remuneration reflects ‘the functions that each enterprise performs’ and to carry out a ‘comparison of the functions taken on by the parties’.

200    Admittedly, it cannot be ruled out that paragraph 1.20 of the 1995 version of the OECD Guidelines may be interpreted as meaning that the term ‘taken on’ refers to ‘active’ functions.

201    However, it is not clear from paragraph 1.20 of the 1995 version of the OECD Guidelines that only ‘active’ functions could be taken into account for the purposes of the functional analysis of the parties to the transaction. Nor does it follow from that paragraph that an entity cannot be regarded as ‘performing’ or ‘taking on’ functions when it holds certain assets and merely finances, for example, their development or improvement.

202    In addition, it should be noted that, according to paragraph 1.22 of the OECD Guidelines, it may be ‘relevant and useful in identifying and comparing the functions performed to consider the assets that are employed or to be employed’ and that ‘this analysis should consider the type of assets used, such as plant and equipment, the use of valuable intangibles, etc., and the nature of the assets used, such as the age, market value, location, property right protections available, etc.’. In other words, it is recommended that account be taken of the fact that a company makes assets available in the context of the controlled transaction for the purpose of examining the functions carried out. It therefore follows that, contrary to the Commission’s assertions, the making available of intangible assets had to be taken into account for the purpose of examining the functions performed or assumed by a party to an intra-group transaction, without any distinction between ‘active’ and ‘passive’ functions being relevant.

203    Secondly, even if the Commission could indeed draw a distinction between ‘passive’ and ‘active’ functions, it wrongly concluded, as is apparent from recital 420 of the contested decision, that LuxSCS was merely a passive holder of the intangible assets, that it had merely maintained the intangible assets and that no other active function could be attributed to it.

204    First of all, the Commission failed to take into account the fact that LuxSCS did indeed exploit those assets by making them available to LuxOpCo in exchange for the payment of a royalty under the Licence Agreement.

205    It is common ground that, under the Licence Agreement, LuxSCS licensed all Amazon’s intangible assets in the European territory to LuxOpCo. That agreement covered not only all the intangible assets referred to in the Buy-In Agreement and the CSA, but also the intangible assets, and in particular the brands, that it had received in 2006 from its European affiliates, as well as the resulting derivative works. Licensing the intangible assets to LuxOpCo in return for the payment of the royalty constitutes exploitation of those assets, which amounts to performing an active function.

206    That exploitation corresponds to a use of the intangible assets by LuxSCS, the alleged absence of which is criticised by the Commission in recitals 430 to 432 of the contested decision.

207    The use of the intangible assets by LuxSCS by making them available to LuxOpCo under the Licence Agreement also satisfies the criterion suggested by the Commission in paragraph 83 of its defence in Case T‑318/18. According to that criterion, the rule set out in paragraph 3.43 of the 1995 version of the OCDE Guidelines, as mentioned in paragraphs 183 and 184 above, was created by the authors of those guidelines ‘on the assumption that the party to a controlled transaction that owns valuable intangible property will … be the party using those assets, while performing active functions in relation to that transaction’. In that regard, without it being necessary to establish whether the Commission is justified in considering that that paragraph must be interpreted as requiring some use of the intangible assets, it must be pointed out that making LuxSCS’s intangible assets available to LuxOpCo under the Licence Agreement constitutes use as described by the Commission.

208    Furthermore, it should be noted that LuxSCS contributed to the development of intangible assets through its financial participation under the CSA. In that context, it should be borne in mind that, as has already been pointed out in the last sentence of the second indent of paragraph 4 above, LuxSCS was required to pay an annual share of the costs relating to the CSA development programme.

209    In that regard, it should be noted that it is not apparent from the 1995 version of the OECD Guidelines that financial participation in a cost-sharing agreement cannot be regarded as genuine participation in the development of assets covered by such an agreement. On the contrary, it follows from paragraph 8.15 of the 1995 version of the OECD Guidelines – which states, with regard to cost-sharing agreements, that ‘it is unlikely to be a straightforward matter to determine the relative value of each participant’s contribution except where all contributions are made wholly in cash’ – that a financial contribution to such a cost-sharing agreement may indeed be a valid and valuable contribution, regardless therefore of whether the other entity which made the financial contribution also makes contributions of another nature. In some cases, it cannot be ruled out that the financial contribution to an intra-group transaction may be the sole enabler of the (commercial) success of the transaction.

210    In addition, pursuant to Sections 6.3(b) and 6.4 of the CSA, in return for its share of the costs, LuxSCS became co-owner, with A9, of part of the intangible assets that were continually being developed and improved in the United States. Those developments and improvements were made available by LuxSCS to LuxOpCo continually under the Licence Agreement, in such a way that it can be considered that, from LuxOpCo’s point of view, they were attributable to LuxSCS and not to the United States entities. Under the Licence Agreement, the results of the developments and improvements to the intangible assets are allocated to LuxSCS.

211    It therefore follows from the foregoing that the Commission erred in finding, in recital 429 of the contested decision, that ‘the only functions that could actually have been said to have been performed by LuxSCS were functions related to the maintenance of its legal ownership of the Intangibles’. The criterion used by the Commission, which refers to the distinction between active and passive functions, is not relevant. Furthermore, even if that criterion were to be used, it should be noted that LuxSCS made the intangible assets available to LuxOpCo and contributed to their development through its financial contribution under the CSA. Those functions should have been taken into account by the Commission in its functional analysis of LuxSCS and for the purposes of choosing the tested party.

212    That conclusion is not called into question by the other arguments put forward by the Commission.

213    First, the assessment made by the Commission in recitals 420 and 421 of the contested decision and reiterated in the present proceedings (see paragraph 181 above) according to which LuxSCS ‘could not perform any active and critical functions in relation to [the] development, enhancement, management or exploitation [of the intangible assets]’ because LuxSCS ‘was no longer entitled to economically exploit the Intangibles in [the Amazon group’s] European operations’ cannot succeed.

214    The Commission based that finding on the assertion, repeated several times in the contested decision, that LuxOpCo had received an ‘irrevocable’ and ‘exclusive’ licence from LuxSCS (see, for example, recitals 116, 419, 431, 438, 442 and 450 of the contested decision), which deprived LuxSCS of any possibility of using the intangible assets.

215    In that regard, it is sufficient to recall that granting a licence already constitutes an exploitation.

216    Secondly, the conclusion in paragraph 211 above is not called into question by the assessment made by the Commission in recital 421 of the contested decision and repeated in the present proceedings (see paragraph 181 above) according to which LuxSCS did not have the capacity to perform functions, since it had no employees.

217    In that regard, it should be noted that, contrary to what is claimed by the Commission, whether or not LuxSCS had the capacity to operate an e-commerce business entirely on its own has no bearing on the assessment of LuxSCS’s functions in connection with the exploitation of the intangible assets. As stated in paragraph 204 above, LuxSCS indeed exploited the intangible assets by licensing them to LuxOpCo.

218    Furthermore, contrary to the Commission’s contention, it was not necessary for LuxSCS to have its own employees in order to contribute to the continuous development of the intangible assets. LuxSCS contributed to this through its financial participation in the CSA.

219    Thirdly, the conclusion in paragraph 211 above is not called into question by the Commission’s argument that LuxSCS’s financial contribution to the development of the intangible assets was purely artificial, since the funding for the development of the intangible assets came from LuxOpCo’s accounts, with the result that LuxOpCo performed all the functions attributed to LuxSCS by the CSA.

220    The origin of the capital used by LuxSCS to meet its financial obligations under the CSA, and therefore the fact that this capital came from the payment of the royalty by LuxOpCo, is not relevant. The 1995 version of the OECD Guidelines do not require that the capital invested come from a particular source. It is not excluded that that capital have its origin in a royalty, such as that at issue in the present case, or that it come from another source of revenue, such as, for example, a loan.

221    In any event, it is common ground that LuxSCS had its own capital, in addition to the income derived from the royalty. As the Grand Duchy of Luxembourg has pointed out, it was as a result of its own capital that LuxSCS was able to absorb the losses sustained during its first years of operation without LuxOpCo’s intervention. In 2006, the amount of the royalty paid by LuxOpCo to LuxSCS was, moreover, significantly lower than the payments made by LuxSCS under the Buy-In Agreement and the CSA.

222    In the third place, the Grand Duchy of Luxembourg and Amazon dispute the Commission’s assessment, set out inter alia in recitals 407 and 547 of the contested decision, that LuxSCS could not be regarded as performing ‘unique and valuable’ functions (see inter alia recitals 407 and 547 of the contested decision).

223    It should be noted that the 1995 version of the OECD Guidelines do not use the terms ‘unique and valuable functions’. Only the expression ‘unique and valuable assets’ is used, several times, inter alia in the sections relating to the TNMM and the profit split method, most often in reference to intangible assets (development or ownership) (see, for example, paragraphs 1.8, 3.19, 3.43 and 6.26 of those guidelines).

224    By contrast, it is only in the 2017 version of the OECD Guidelines, which are not relevant in the present case, that reference is clearly made to ‘unique and valuable’ functions or contributions and that a distinction is made between, on the one hand, ‘unique and valuable functions’ and, on the other hand, ‘routine functions’. As has already been pointed out in paragraph 184 above, paragraph 6.17 of the 2017 version of the OECD Guidelines contains a definition of the concept of ‘unique and valuable assets’. By contrast, although the authors of the 2017 version of the OECD Guidelines often refer to the concept of ‘unique and valuable functions’, they do not provide any definition in that regard.

225    The main parties have explained what they understood by the expression ‘routine functions’. At the hearing, the Grand Duchy of Luxembourg stated that an entity performs ‘routine functions’ when it performs normal functions, namely functions which other undertakings could also perform. They are therefore essentially functions for which comparables can easily be found. Amazon pointed out at the hearing that the concept of ‘routine functions’ did not mean that the functions in question were of not valuable, but that they could easily be benchmarked and remunerated. The Commission has not called that understanding into question. It is apparent from paragraph 14 (footnote 18) of the defence in Case T‑318/18 that, according to the Commission, the word ‘routine’ refers to functions which are not unique and for which uncontrolled comparables can be found. Similarly, in paragraph 17 (footnote 21) of the defence in Case T‑816/17, the Commission compares ‘routine’ functions to those which are ‘not unique and valuable’.

226    In the present case, there is no need to determine whether, on the basis of the 1995 version of the OECD Guidelines, it was already open to the Commission to verify the arm’s length nature of a price by using the concept of ‘unique and valuable functions’ because this concept was already applicable at the time the 1995 version of the OECD Guidelines applied, even if they did not expressly refer to that concept, or whether it was only after the adoption of the 2017 version of the OECD Guidelines that the criterion relating to ‘unique and valuable functions’ could be taken into account for that purpose.

227    In any event, the main parties have not called into question the relevance of that criterion; rather they unanimously put that criterion at the core of their arguments as a relevant parameter for assessing their situation. In that respect, it should be noted that, as is the case as regards the concept of ‘unique and valuable assets’ (see paragraph 176 above), the parties have not taken care to define the terms ‘unique and valuable functions’.

228    As regards the meaning of the terms ‘unique and valuable functions’, in line with the conclusion in paragraph 184 above, and in view of the meaning of those terms accepted by the parties (see paragraph 225 above), for the purposes of the present case the concept of a ‘unique function’ must be understood as referring to the situation in which no comparable exists for a certain function. The concept of a ‘valuable function’ relates in particular to the fact that the function in question enables the generation of significant revenue. In that respect, it must be noted that, although the designation of a certain function as ‘unique’ means that the same function cannot be categorised as ‘routine’, the same reasoning cannot apply to the concept of a ‘valuable function’. There are also ‘routine’ functions which make it possible to generate substantial revenue and which therefore merit classification as ‘valuable functions’.

229    In the present case, as stated in paragraph 191 above, it is not disputed that the intangible assets covered by the Licence Agreement were unique and valuable.

230    Moreover, LuxSCS not only exploited, but also contributed financially to the development of those unique and valuable intangible assets held by it. Consequently, it follows from what has just been indicated in paragraphs 203 to 211 above that all of LuxSCS’s functions in relation to the intangible assets should have been regarded by the Commission as unique and valuable. The assertion in recital 547 of the contested decision that the Luxembourg tax administration should have required a functional analysis showing that LuxSCS performed ‘unique and valuable functions’ is therefore unjustified and must, accordingly, be rejected. Consequently, in view of LuxSCS’s functions and assets, the Commission’s conclusion that LuxSCS should have been regarded as the tested party is unconvincing.

–       The risks assumed by LuxSCS

231    The Grand Duchy of Luxembourg and Amazon submit, in essence, that LuxSCS bore the risks associated with the intangible assets as such, while LuxOpCo bore only the risks associated with its retail business. LuxSCS also assumed financial risks in connection with the intangible assets, since it had to meet its obligation under the Buy-In Agreement and the CSA to pay the Buy-In payments and the CSA payments to ATI and A9.

232    The Commission disputes those arguments.

233    It maintains, in particular, that neither the resolutions of LuxSCS’s sole manager nor the minutes of LuxSCS’s general meetings mention any critical decisions relating to risk management. In reality, LuxSCS had neither the financial capacity nor the operational capacity to assume those risks. LuxSCS could only bear the costs related to the Buy-In Agreement and the CSA through the funds received annually by way of the royalties paid by LuxOpCo under the Licence Agreement, with the result that LuxSCS’s capital was never exposed to risk. Moreover, LuxSCS benefited from a substantial initial capitalisation from its parent company, which covered the Buy-In payment. In any event, under the Licence Agreement, LuxSCS transferred the financial risks to LuxOpCo. Accordingly, the risks borne by LuxSCS are theoretical since LuxSCS could have terminated the Licence Agreement and granted a licence to another related or independent party. The financial risks borne by LuxSCS were also theoretical, because its financial participation in the CSA was financed by the royalty paid by LuxOpCo and the amount of the CSA payments was correlated with LuxOpCo’s revenue.

234    In that regard, it should be noted at the outset that, since LuxSCS had acquired full ownership of part of the intangible assets under Section 3.1 of the assignment agreement concluded with ATI on 1 January 2005, it bore all the risks linked to the existence of the intangible assets as such. These were, for example, risks such as third-party challenges or the revocation of the intangible assets. That is the logical consequence of the fact that LuxSCS was the owner of those assets. With regard to the licensing agreement concluded with ATI on 1 January 2005, LuxSCS also assumed the risks related to the development of intangible assets by the American entities ATI and A9.

235    Since it had, during the relevant period, a licence on the other part of the intangible assets, referred to in Section 3.1 of the licence agreement entered into with ATI on 1 January 2005, as well as in Sections 6.1 and 6.2 of the CSA, LuxSCS bore financial risks in connection with those assets used as a result of its participation in the CSA. More specifically, provision was made for the allocation of costs between the parties to the CSA in accordance with Sections 4 and 5 of the CSA. Under those sections of the CSA, LuxSCS was obliged to bear the costs relating to the development of the intangible assets. Although the cost allocation depended on the share of the profits made in Europe by comparison with worldwide profits, the costs as such were wholly unrelated to the level of profits achieved in Europe. In that regard, it should be noted that, if the development costs were higher than the royalty paid by LuxOpCo, LuxSCS would have had to bear the financial consequences of that difference. Thus, if LuxOpCo had recorded losses or low profits, the royalty would not have been sufficient to cover the fixed costs borne by LuxSCS, namely essentially the payments under the Buy-In Agreement and the CSA. In other words, LuxSCS risked not having sufficient revenue to make the Buy-In and cost-sharing payments provided for in the Buy-In Agreement and the CSA.

236    With regard to these financial risks, it should be pointed out that, despite an unsubstantiated assertion made by it at the hearing, the Commission has not established that LuxSCS’s obligation to make the payments due under the CSA was in fact exactly correlated to the payment of the royalty by LuxOpCo. On the contrary, and as the Commission itself pointed out in recital 445 of the contested decision, the amounts received by LuxSCS by way of the royalty did not correspond directly to the amounts owed by LuxSCS in respect of the CSA. Thus, in 2006, the amount of the royalty paid by LuxOpCo to LuxSCS was significantly lower than the Buy-In and cost-sharing payments made by LuxSCS.

237    Still with regard to the financial risks borne by LuxSCS, the Commission has not been able to demonstrate that that company did not have substantial equity. As regards LuxSCS’s initial capital, which was dismissed as irrelevant by the Commission in recital 445 of the contested decision, it is common ground that, at least in respect of 2006, it was as a result of that capital that LuxSCS was able to absorb the losses sustained during its first years of operation, without LuxOpCo’s intervention.

238    Lastly, it is true that, according to Sections 2.3 and 9.2 of the Licence Agreement, LuxOpCo was obliged to safeguard the intangible assets. First, under the terms of Section 2.3 of that agreement, LuxOpCo was required to take all necessary actions to safeguard LuxSCS’s rights over the intangible assets and, secondly, under Section 9.2 of that agreement, LuxOpCo was obliged, at its sole expense, to prevent and prosecute any unauthorised use of the intangible assets. LuxOpCo therefore assumed the risks associated with the protection of the intangible assets.

239    Nevertheless, the other risks related to the intangible assets were borne by LuxSCS due to its financial participation in the CSA.

240    It is not apparent from the provisions of the Licence Agreement that LuxSCS transferred risks to LuxOpCo other than those stemming from Sections 2.3 and 9.2 of that agreement, namely those relating to the obligation to safeguard the intangible assets. Thus, contrary to what the Commission suggests, the Licence Agreement does not contain any clause on the transfer, as such, of all risks associated with LuxSCS’s intangible assets to LuxOpCo. In particular, the Licence Agreement does not contain any clause relating to the transfer of risks linked to the development of the intangible assets.

241    Since it is not supported by the provisions of the Licence Agreement, the conclusion expressed by the Commission, inter alia in recital 438 of the contested decision, that LuxSCS passed on to LuxOpCo the risks related to the development, management and exploitation of the intangible assets cannot therefore be accepted.

242    It is therefore apparent from the foregoing that the Grand Duchy of Luxembourg and Amazon are justified in asserting that LuxSCS bore the risks associated with the ownership and development of the intangible assets used to operate the European business, including the financial risks associated with the exploitation of those intangible assets, whereas LuxOpCo essentially bore only the risks associated with its own retail business and, in particular, the risks associated with sales and the marketplace services.

–       Conclusion on the functional analysis of LuxSCS and the impact of that conclusion on the choice of that company as the tested party

243    In the light of the considerations set out in paragraphs 162 to 242 above, it is appropriate to make two findings.

244    In the first place, the functional analysis of LuxSCS carried out by the Commission cannot be accepted. The Commission underestimated LuxSCS’s functions. As regards the intangible assets, the Commission failed, inter alia, to take into account the fact that, under both the contractual arrangements and in practice, LuxSCS made available intangible assets for which no comparables existed on the market and which were therefore unique and valuable. According to the relevant version of the OECD Guidelines in the present case, that factor was sufficient, in principle, to support the conclusion that LuxSCS could not be regarded as the least complex party and therefore the tested party.

245    In any event, if it were to be considered, as the Commission contends, that the Luxembourg tax authorities should have taken into account ‘unique and valuable functions’, it must be held that the Commission ignored the fact that LuxSCS did indeed exploit the intangible assets in the context of the examined controlled transaction. The making available of highly valuable intangible assets amounted to the performance of a unique and valuable function in the context of the Licence Agreement (the controlled transaction). As is apparent from paragraphs 203 to 242 above, LuxSCS performed a series of functions in the context of the controlled transaction other than making the intangible assets available to LuxOpCo. The Commission overlooked those functions, which could be regarded as unique and valuable.

246    The Commission also failed to take due account of the fact that, both under the contractual arrangements and in practice, LuxSCS assumed all of the risks related to these assets and their development under the Licence Agreement, regardless of whether LuxSCS was itself controlled by the US entities and whether it was LuxSCS that technically developed the intangible assets or whether, with LuxSCS contributing financially, the intellectual property developments were the result of the technical efforts of the US entities ATI and A9. In doing so, the Commission also minimised the description of the risks assumed by LuxSCS.

247    In those circumstances, the Luxembourg tax authorities cannot be criticised for having considered, like the authors of the 2003 transfer pricing report, that it was correct, according to the 1995 version of the OECD Guidelines, not to choose LuxSCS as the tested party.

248    In the second place, in any event, even if the Commission’s assertion that LuxSCS was merely a passive holder of the intangible assets and did not perform active functions in relation to those assets were accepted, it must be found that the Commission erred in considering that LuxSCS should have been chosen as the tested party.

249    It must be borne in mind that, in general, the tested party is the one to which the TNMM can be applied in the most reliable manner and for which the most reliable comparables can be found.

250    In the present case, it must be held that the Commission has not demonstrated that it was easier to find undertakings comparable to LuxSCS than undertakings comparable to LuxOpCo, or that choosing LuxSCS as the tested entity would have made it possible to obtain more reliable comparison data.

251    As is apparent from recital 557 of the contested decision, the Commission should have accepted, when seeking to determine the appropriate mark-up for the royalty, that there were no comparables for LuxSCS.

252    It follows that the argument put forward by the Grand Duchy of Luxembourg and Amazon that the Commission wrongly considered that LuxSCS should have been chosen as the tested entity for the purposes of applying the TNMM must be upheld. That being so, the foregoing considerations are sufficient to uphold the entire line of argument put forward by the Grand Duchy of Luxembourg and Amazon as regards the Commission’s primary finding of an advantage, without there being any need to carry out a functional analysis of LuxOpCo, or to examine whether the Commission was justified in rejecting the CUP.

253    However, for the sake of completeness, it should be noted that the Commission’s assessments concerning the existence of an advantage for the purpose of Article 107(1) TFEU must also be rejected for reasons other than those relating to the choice of the tested party and to the functional analysis of LuxSCS, as set out above. Thus, even if it were necessary to accept the Commission’s unjustified conclusion that LuxSCS should have been the tested party, the line of argument put forward by the Grand Duchy of Luxembourg and Amazon would still have to be upheld, for the following reasons.

(ii) The remuneration calculated by the Commission for LuxSCS on the basis that it was the tested party

254    In recitals 550 to 560 of the contested decision, the Commission sought to apply the TNMM itself, using LuxSCS as the tested party. At the end of its analysis, in recital 559 of the contested decision, the Commission concluded that the ‘arm’s length remuneration’ for LuxSCS, under the Licence Agreement, should be equal to the sum of two components, namely, on the one hand, the Buy-In costs and the CSA costs incurred by LuxSCS in relation to the intangible assets, without a mark-up, and, on the other hand, general operating costs incurred directly by LuxSCS to perform functions related to the maintenance of its legal ownership of the intangible assets (‘the maintenance costs’), increased by 5% (‘LuxSCS’s remuneration’). It should be noted in that regard that LuxSCS’s remuneration in fact corresponds to the royalty which, according to the Commission, should have been collected by LuxSCS from LuxOpCo.

255    In the second complaint in the second part of the first plea and the third part of the first plea in Case T‑816/17 and in the fourth plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon submit, in essence, that, even if it were accepted that LuxSCS could be considered the tested party for the application of the TNMM (quod non), the Commission committed other errors in applying the TNMM. They argue that the calculation carried out by the Commission in order to determine the ‘arm’s length remuneration’ for LuxSCS, namely the royalty payable by LuxOpCo to LuxSCS, is unconvincing.

256    It is appropriate to address that line of argument by considering the two components identified by the Commission (see paragraph 254 above).

–       The first component of the royalty due to LuxSCS (Buy-In costs and CSA costs)

257    As regards the first component of the royalty due to LuxSCS (see paragraph 254 above), the Grand Duchy of Luxembourg submits in its first complaint in the second part of the first plea that the royalty paid by LuxOpCo to LuxSCS should reflect not only the development costs, but also the value of the intangible assets. That value is uncorrelated with those costs and therefore with the payments made by LuxSCS under the CSA. In paragraph 73 of the application and paragraph 32 et seq. of the reply in Case T‑318/18, Amazon puts forward, in essence, the same complaint. In addition, according to the Grand Duchy of Luxembourg, the costs related to the CSA and the Buy-In Agreement, which constitute the consideration for the making available by LuxSCS of the intangible assets under the Licence Agreement, should have been included in the costs to which a mark-up is applied.

258    The Commission disputes those arguments.

259    It contends that the objective of a cost-sharing agreement such as the CSA is to share the costs associated with the development of the intangible assets, and not to obtain an operational profit on the European activities. Accordingly, ATI and A9 should not obtain any share of the profits from the European business activities beyond the reimbursement of the Buy-In and CSA costs. The contested decision was therefore correct in determining that LuxSCS’s remuneration should comprise a reimbursement of the Buy-In payments and the development costs under the CSA. The Commission submits in that context that, in its view, LuxSCS existed solely for tax purposes. The Licence Agreement was not concluded directly between the United States entities and LuxOpCo, but between LuxSCS and LuxOpCo, in order to avoid the royalty being subject to tax in the United States. If LuxSCS had not existed, ATI and A9 would have entered into a cost-sharing agreement with LuxOpCo (and not a licence agreement), with the result that LuxOpCo alone would have had to make the payments. Moreover, LuxSCS’s activity was limited merely to holding the intangible assets. LuxSCS did not itself directly perform any functions relating to the development of the intellectual property and should not therefore have received remuneration in that respect. It played no role in the use or development of the intangible assets, nor did it exercise any control over those development functions or the related risks. According to the Commission, no mark-up should therefore be applied to the Buy-In costs and the CSA costs, since these are merely costs passed on by LuxSCS to LuxOpCo and LuxSCS did not perform any function in relation to the intangible assets. On the contrary, LuxSCS’s remuneration should have reflected the fact that the functions and risks attributed to LuxSCS under the CSA were in fact borne by LuxOpCo. In any event, the Commission did not ignore, in its functional analysis, the fact that LuxSCS was the legal owner of the intangible assets.

260    As a preliminary point, it should be noted that the exercise consisting of examining whether a royalty such as that at issue in the present case corresponds to a market outcome presupposes, according to the 1995 version of the OECD Guidelines, relying on the value of the intangible assets and not on the costs of developing and maintaining them. It is apparent from paragraph 6.27 of those guidelines that, although the costs of developing the intangible assets may be taken into account as an aid to determining the comparability or relative value of the contribution of the various parties to a transaction, there is no necessary link between those costs and the value of the intangible assets. In particular, the actual fair market value of intangible property is frequently not measurable in relation to the costs involved in developing and maintaining the property. As is apparent from paragraph 6.2 of those guidelines, ‘intangible property’ may be of considerable value even if it has no book value in the company’s balance sheet. Lastly, as is apparent from paragraphs 1.22 and 6.27 respectively of those guidelines, the value in question is known as ‘market value’ or ‘fair market value’. Furthermore, it must be pointed out that that value may be subject to fluctuations over time.

261    In the present case, the question arises whether the first component of LuxSCS’s remuneration, as calculated by the Commission in the contested decision, namely the Buy-In payment without a mark-up and the CSA payments, also without a mark-up, does indeed reflect the value of the intangible assets licensed to LuxOpCo.

262    In the first place, it is true that the Buy-In payment made by LuxSCS to the US entities in consideration for the transfer of ownership of part of the pre-existing intangible assets and a licence to the remaining pre-existing intangible assets (see paragraph 4 above) may indeed be regarded as reflecting the value of the intangible assets at the time of the conclusion of the Buy-In Agreement, namely in 2005.

263    Although the amount of the Buy-In payment does not constitute a price which was freely negotiated on the market, it is, as Amazon states in paragraph 73 of the application in Case T‑318/18, the price paid in return for the acquisition of the intangible assets existing in 2005. Such a payment, unlike the development costs, is likely to reflect the value of the intangible assets which were the subject of the transfer of ownership, namely the intangible assets already existing in 2005.

264    Nevertheless, it must be pointed out that, as has been stated in particular by the Grand Duchy of Luxembourg, without being contradicted on that point by the Commission, during the relevant period, the intangible assets increased in value significantly as a result of the continuous innovations in respect of the technology developed inter alia by the US Amazon entities and as a result of the development of the Amazon brand’s reputation, and therefore of the intangible assets, linked to marketing in Europe and worldwide. The simple addition of the development costs without a mark-up (the CSA payments) to the price paid to obtain the pre-existing intangible assets (the Buy-In payment), carried by the Commission in recital 555 of the contested decision, does not take into account the fact that, in the present case, the value of the pre-existing intangible assets increased during the relevant period, since those assets were gradually developed and improved by the US entities and partly replaced. The mere passing on of the payment under the Buy-In Agreement, relied on by the Commission, which may be accepted as being the initial value of the intangible assets in 2005, does not therefore reflect the market value of those intangible assets during the entirety of the relevant period.

265    Moreover, the Commission incorrectly took the view that the payments made by LuxSCS under the Buy-In Agreement could be passed on to LuxOpCo without applying a mark-up. The absence of a mark-up does not reflect what independent parties would have accepted in the context of an open market transaction and therefore constitutes an error in the calculation of LuxSCS’s remuneration. It is reasonable to consider, and it is apparent, moreover, in particular from paragraph 6.14 of the 1995 version of the OECD Guidelines, that independent parties operating on the market seek to obtain profits from making available their assets. Accordingly, the application of a mark-up in the calculation of remuneration such as that at issue appears to be a commonplace practice on the market. As Amazon submits in paragraph 98 of the application in Case T‑318/18, if the Commission had examined the options available to LuxSCS, as suggested in paragraph 6.14 those guidelines, it could have found that there were numerous operators of e-commerce businesses in Europe, with the result that LuxSCS could have valued the intangible assets above their development costs alone.

266    Next, in the second place, as regards the payments under the CSA, it should be noted that, as has been stated above, it is apparent from paragraph 6.27 of the 1995 version of the OECD Guidelines that, although the costs of developing intangible assets may be taken into account as an aid to determining the comparability or relative value of the contribution of the various parties to a transaction, there is no necessary link between the development costs and the value of intangible assets. The mere passing on of the CSA payment suggested by the Commission corresponds only to the reimbursement of the costs to be borne by LuxSCS for the development of the intangible assets and does not reflect the value of the improved intangible assets. The mere reimbursement of development costs, without a mark-up being applied, is an approach which does not reflect a market outcome.

267    In that context, it should be noted that the object of the controlled transaction examined by the Commission is the licensing of intangible assets by LuxSCS to LuxOpCo, bearing in mind that LuxSCS was a party to the CSA. It is common ground that ATI and A9 carried out the functions of developing part of the intangible assets. However, the Commission’s argument that ATI and A9 were ‘remunerated’ by the payments under the CSA for those functions reflects a misunderstanding of the CSA by the Commission. It is clear from Section 4.3 of the CSA that the payments made by LuxSCS under the CSA were calculated only as a percentage of the development costs incurred by the parties to the CSA. It is true that LuxSCS’s contribution to the development costs is proportional to the profits made by the entities held by LuxSCS, and thus by LuxOpCo, by comparison with the profits made by ATI and A9. Nevertheless, the payments under the CSA thus correspond to a fraction of the development costs of the intangible assets developed in the context of the CSA and made available to LuxOpCo under the Licence Agreement and they therefore do not reflect the market value of those intangible assets. That is the value that an arm’s length royalty under the Licence Agreement should reflect.

268    In the light of the foregoing, the fact that LuxSCS did not itself directly perform the development functions does not call into question the finding that the amount of the royalty paid by LuxOpCo must reflect the value of the intangible assets.

269    Accordingly, it was inappropriate for the Commission to assert that LuxSCS’s remuneration could be calculated on the basis of the mere passing on of the development costs of the intangible assets.

270    The conclusion in paragraph 269 above is not called into question by the Commission’s other arguments.

271    First, the Commission has argued that LuxSCS was merely an intermediary and that it merely passed on to LuxOpCo the costs incurred in relation to the Buy-In Agreement and the CSA, and then transferred a portion of the royalty received from LuxOpCo under the Licence Agreement to A9 and ATI in the amount of those costs. The difference between the amounts received in respect of the royalty and the payments made under the CSA was attributed to LuxSCS and then possibly remitted by its partners, without LuxSCS having performed any function that would justify the attribution of these amounts to it.

272    However, even if it were to be considered that LuxSCS was merely an intermediary, that is to say that it was interposed between LuxOpCo and the US entities ATI and A9, which had not performed development functions, the fact remains that the amount of the royalty that LuxOpCo should have paid, and thus LuxSCS’s remuneration, should have reflected the market value of the intangible assets made available under the Licence Agreement. The mere passing on of the payment under the CSA, relied on by the Commission, corresponds only to the reimbursement of the costs that LuxSCS incurred for the purposes of developing the intangible assets and does not reflect the market value of those intangible assets.

273    If, through the arguments mentioned in paragraph 271 above, the Commission seeks to argue that LuxOpCo’s tax base was reduced as a result of the interposition of LuxSCS between LuxOpCo and the US entities ATI and A9, and the conclusion of the Licence Agreement with LuxSCS, as opposed to the conclusion of a licence agreement with those entities, it should be noted that, in the contested decision, the Commission did not rely on such reasoning to demonstrate the existence of an advantage in favour of LuxOpCo.

274    Moreover, it has not been established that, if the Licence Agreement had been concluded by LuxOpCo directly with the US entities, without LuxSCS being interposed between those companies, the amount of a royalty paid to those entities would have been different from the amount of the royalty payable to LuxSCS.

275    Secondly, the conclusion mentioned in paragraph 269 above is not called into question by the argument raised by the Commission at the hearing that the CSA could have been concluded directly with LuxOpCo.

276    In that regard, it should be noted that the reasoning according to which, if LuxSCS had not existed, a cost-sharing agreement would have been entered into with LuxOpCo, is purely hypothetical and is a matter of conjecture.

277    Moreover, in the contested decision, the Commission did not base its reasoning on the fact that LuxOpCo could or should have been directly involved in the CSA. Indeed, it must be noted that nowhere in the contested decision did the Commission question the existence of LuxSCS as such, or the validity, under Luxembourg law, of the arrangement resulting from the conclusion of the CSA and the Licence Agreement, on the ground that that arrangement made it possible to reduce LuxOpCo’s tax burden. The Commission merely challenged the amount of the royalty payable by LuxOpCo to LuxSCS.

278    Thirdly, the conclusion referred to in paragraph 269 above is not called into question by the argument raised by the Commission at the hearing that LuxSCS was created solely for tax purposes.

279    The mere fact that an entity belonging to a group of companies was created solely for the purposes of tax optimisation and that it receives a royalty for intangible assets developed within the group of companies in question is not sufficient, in itself, to support the conclusion that there was a tax advantage within the meaning of Article 107(1) TFEU for the entity liable for payment of the royalty and therefore does not necessarily demonstrate the existence of State aid in favour of the entity liable for payment of the royalty.

280    In the present case, it is true that the different tax treatment of LuxSCS in Luxembourg (LuxSCS was ‘fiscally transparent’ in Luxembourg) and the United States (LuxSCS was ‘fiscally non-transparent’ in the United States) arises from a ‘hybrid mismatch’, that is to say, a difference in the tax rules applicable in Luxembourg and the United States as regards the identification of the taxpayer.

281    However, as the Commission itself pointed out in footnote 16 to paragraph 13 of the defence in Case T‑816/17, the consequences of that mismatch (the non-taxation of profits) are not the subject of the contested decision. The relevant question in the context of the present action is therefore not whether LuxSCS was created purely for tax purposes, nor whether the income which it generated was actually taxed in the United States in the hands of its partners, but rather whether LuxOpCo paid a royalty which was overpriced and whether, as a result, LuxOpCo’s remuneration and, therefore, its taxable base were artificially reduced.

282    Fourthly, the conclusion referred to in paragraph 269 above is not called into question by the Commission’s assertion at the hearing, even if it were established, that LuxSCS was a ‘fictitious’ company.

283    In that regard, it should be noted that LuxSCS did indeed have legal existence, which the Commission does not call into question. LuxSCS was established in Luxembourg and was listed in the business register of the Grand Duchy of Luxembourg as a Luxembourg company.

284    In the light of the foregoing, it must be held that the conclusion in recital 555 of the contested decision that the first component of LuxSCS’s remuneration should have consisted of the ‘re-charge of the pass through costs it bore in relation to the Buy-In Agreement and the CSA (i.e. the Buy-In and CSA Costs)’ is erroneous, since such a royalty does not correspond to a market outcome. That error in the application of the TNMM is also sufficient for it to be held that the Commission’s primary finding of an advantage within the meaning of Article 107(1) TFEU cannot be upheld. However, it is appropriate to pursue the examination of the parties’ arguments as follows.

–       The second component of LuxSCS’s remuneration (maintenance costs)

285    As regards the second component of LuxSCS’s remuneration (see paragraph 254 above), the Grand Duchy of Luxembourg submits that the assessment in recital 555 of the contested decision, namely the argument that ‘LuxSCS should be remunerated with a mark-up on a cost-base consisting solely of the costs incurred for the external services acquired to maintain its legal ownership of the Intangibles’, is incorrect. In that regard, the Grand Duchy of Luxembourg submits that the Commission was wrong to set the ‘arm’s length’ mark-up at 5% of external costs on the basis of the JTPF report. More specifically, according to the Grand Duchy of Luxembourg, the 5% mark-up, deemed to be the ‘arm’s length’ mark-up, is arbitrary, as is the cursory analysis on which that mark-up is based. The JTPF report is based on an analysis of the practices observed by the tax administrations of the Member States and not on an analysis of the practice in Luxembourg in respect of Article 164(3) of the LIR. Leaving aside the fact that it has no value under Luxembourg law and that it was adopted after the tax ruling at issue and was therefore not available at the time that decision was adopted, the JTPF report referred to mark-ups observed in the context of intra-group transactions and could not therefore be used as a basis for determining an arm’s length mark-up, namely the mark-up corresponding to conditions on the open market.

286    The Commission disputes those arguments.

287    It submits that the second component of the royalty payable by LuxOpCo to LuxSCS represents a small fraction of that royalty, with the result that it does not really affect LuxSCS’s ‘remuneration’ as calculated by the Commission. In its view, it was not necessary, in the present case, to carry out a genuine transfer pricing analysis and to determine what the precise amount of LuxOpCo’s remuneration should have been. Rather, the JTPF report could be used as a ‘safe harbour’ and made it possible to determine the amount of low-value intra-group transactions, for which it would be too costly and time-consuming to carry out a genuine transfer pricing analysis. The Grand Duchy of Luxembourg is a member of the Joint Transfer Pricing Forum and the JTPF report is also based on the practice in Luxembourg. Although the mark-ups found in the JTPF report were observed for intra-group transactions, they are, according to the Commission, mark-ups generally accepted by the tax authorities in that they reflect the profitability of companies under market conditions. Lastly, the Commission states that, although the JTPF report dates from 2010, it is based on data relating to the period between 1999 and 2007, adding that those data may be used, since the tax ruling at issue was implemented only from 2006.

288    As a preliminary point, as noted in paragraph 254 above, the second component of LuxSCS’s ‘remuneration’, as calculated by the Commission, corresponds to costs which could be described as ‘maintenance costs’, plus a 5% mark-up That return of 5% was adopted by the Commission on the basis of the JTPF report, since it is the rate of return most often observed for transfer pricing, in relation to low value adding intra-group services.

289    As the Grand Duchy of Luxembourg and Amazon submit, the approach adopted by the Commission is problematic in several respects.

290    First of all, the Commission itself acknowledged in recital 557 of the contested decision that no comparables existed for the purpose of determining LuxSCS’s remuneration for its functions corresponding to the maintenance of its ownership of the intangible assets.

291    According to paragraph 3.26 of the 1995 version of the OECD Guidelines, where the TNMM is to be applied, the ‘net margin of the taxpayer from the controlled transaction … should ideally be established by reference to the net margin that the same taxpayer earns in comparable uncontrolled transactions’. The absence of comparables should have led the Commission not to apply the TNMM to LuxSCS.

292    Admittedly, the approach adopted by the Commission seeking to use the JTPF report, instead of carrying out its own comparability analysis and its own examination of the comparable net margins on the market, is not incompatible with the rules for the application of the TNMM, as they emerge from the 1995 version of the OECD Guidelines. First, as is apparent in particular from paragraphs 3.29 and 3.30 of those guidelines, it is well known that it is difficult to find sufficiently specific information concerning existing net margins on the open market and the parameters frequently used on the open market as profit indicators. Secondly, the form and nature of the sources of information used for that purpose are, as such, irrelevant. If there is a publication relating to profit indicators or net margins observed in a certain area of economic activity, that publication may, in principle, be used. Such a publication does not, however, necessarily have to be used or understood as a ‘safe harbour’, such as that referred to by the Commission in the context of its argument referred to in paragraph 287 above.

293    However, the use of such a report can be accepted only if the data contained therein are relevant and reliable. In particular, such a report should, at the very least, contain data which relate to transactions comparable to the controlled transaction and to functions comparable to those of the entity tested, so that the comparison is actually reliable.

294    In the present case, it must be stated that the mark-up chosen by the Commission on the basis of the JTPF report corresponds to the mark-up generally observed, according to the authors of that report, for certain ‘low value adding intra-group services’. However, LuxSCS did not provide such services. The functions associated with maintaining its ownership of intangible assets cannot be treated in the same way as an intra-group supply of a ‘low value adding’ service. It follows that, while in principle the use of the JTPF report does not raise any methodological difficulties, the fact remains that the information contained in that report had no connection with LuxSCS’s functions in the context of the controlled transaction at issue in this case, namely the Licence Agreement.

295    In the light of the considerations set out in paragraphs 257 to 292 above, it is necessary to uphold the arguments raised by the Grand Duchy of Luxembourg and Amazon claiming that the Commission erred in carrying out the functional analysis of LuxSCS, which had an impact on its conclusion that LuxSCS should be used as the tested party in applying the TNMM. The Commission also erred in determining the appropriate net margin applicable to the controlled transaction in the present case.

(3)    Conclusion on the primary finding

296    In the light of those various considerations, the arguments of the Grand Duchy of Luxembourg and Amazon challenging the primary finding of an advantage must be upheld. First, the Commission wrongly considered that LuxSCS should be used as the tested party. Secondly, the Commission’s calculation of LuxSCS’s ‘remuneration’, on the basis that LuxSCS had to be the tested entity, is vitiated by numerous errors and cannot be regarded as sufficiently reliable or capable of achieving an arm’s length outcome. Since the calculation method used by the Commission must be rejected, that method cannot serve as a basis for the Commission’s finding that the royalty paid by LuxOpCo to LuxSCS should have been lower than the royalty actually received, pursuant to the tax ruling at issue, during the relevant period. The elements set out in the primary finding of an advantage therefore do not establish that LuxOpCo’s tax burden was artificially reduced as a result of an over-pricing of the royalty.

297    Consequently, the first and second complaints in the second part and the third part of the first plea in Case T‑816/17 and the second and fourth pleas in Case T‑318/18, claiming that the Commission has not demonstrated the existence of an advantage in the context of its primary finding, must be upheld, without it being necessary to examine the other pleas and arguments challenging the primary finding.

3.      Pleas and arguments challenging the subsidiary line of reasoning concerning the advantage

298    In the third complaint in the second part of the first plea in Case T‑816/17 and in the fifth plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon contest the Commission’s subsidiary line of reasoning concerning the existence of a tax advantage in favour of LuxOpCo.

299    In order to examine those pleas in detail, it should be recalled that, as set out in paragraphs 65 to 68 above, under its subsidiary line of reasoning concerning the existence of an advantage, the Commission made three findings, according to which the transfer pricing method endorsed by the tax ruling at issue is based on three erroneous methodological choices.

300    First of all, in its first subsidiary finding (recitals 565 to 569 of the contested decision), the Commission identified an error in the choice of transfer pricing method endorsed by the tax ruling at issue. In that regard, it should be borne in mind that the parties are in agreement that the method applied in the 2003 transfer pricing report corresponded, in fact, to the TNMM. By contrast, contrary to what is apparent from the 2003 transfer pricing report itself, the authors of that report did not choose, or actually apply, the profit split method. Amazon confirmed, in its answers to the written questions, that the transfer pricing method endorsed in the tax ruling at issue consisted of, first, calculating LuxOpCo’s remuneration using the TNMM and, secondly, allocating all residual profit to LuxSCS in order to remunerate it for the intangible assets. In addition, the fact that the authors of the 2003 transfer pricing report in fact used the TNMM and not the profit split method on the basis of the residual analysis was mentioned by the Commission in recital 540 of the contested decision.

301    In support of its first subsidiary finding, the Commission stated that, even if LuxSCS did perform unique and valuable functions, which the Commission disputes, the Luxembourg tax authorities could not disregard the fact that LuxOpCo also performed unique and valuable functions in connection with the Amazon group’s intellectual property and business operations in Europe, and not routine management functions. Consequently, the Commission took the view that the pricing method used in the tax ruling at issue did not enable a reliable outcome to be reached and that the profit split method, with a contribution analysis, was more appropriate. According to the Commission, had the profit split method been used, LuxOpCo’s remuneration and, consequently, taxable income would have been greater.

302    Next, in its second subsidiary finding (recitals 570 to 574 of the contested decision), the Commission took the view that the choice of profit level indicator endorsed in the tax ruling at issue was incorrect. Specifically, it found that, even if the functional analysis contained in the 2003 transfer pricing report was correct, by endorsing a mark-up on operating expenses and not on total costs, the tax ruling at issue inappropriately reduced LuxOpCo’s taxable income, thereby conferring an economic advantage on LuxOpCo.

303    Lastly, in its third subsidiary finding (recitals 574 to 578 of the contested decision), the Commission concluded that, in any event, the inclusion of a ceiling in the pricing method for determining LuxOpCo’s taxable base, as endorsed in the contested decision, was neither appropriate nor economically justified. According to the Commission, in so far as it led to a reduction of LuxOpCo’s taxable income for the 2006, 2007, 2011, 2012 and 2013 tax years, the inclusion of such a ceiling conferred an economic advantage on LuxOpCo.

304    It should be noted that each of the subsidiary findings in Sections 9.2.2.1 to 9.2.2.3 of the contested decision is independent of the others. Each is therefore capable of establishing the existence of an advantage. The Commission confirmed, both in its answers to the written questions and at the hearing, that each of the subsidiary findings independently and autonomously supports the finding of the existence of an advantage.

(a)    Preliminary observations on the three subsidiary findings

305    In recital 564 of the contested decision, the Commission stated that the purpose of the assessment it had undertaken in Section 9.2.2 relating to the subsidiary findings of an advantage was not to determine a ‘precise’ arm’s length remuneration for LuxOpCo, but to demonstrate that the tax ruling at issue conferred an economic advantage on LuxOpCo by endorsing the three erroneous methodological choices which led to a reduction of its taxable income.

306    In that regard, in addition to what has been set out in paragraphs 123 to 126 above, it is important to clarify, in the light of the content of the judgment of 24 September 2019, Netherlands and Others v Commission (T‑760/15 and T‑636/16, EU:T:2019:669), the level of evidence which the Commission must take into account when examining the existence of State aid in the context of a tax measure such as the tax ruling at issue.

307    First of all, in paragraph 152 of the judgment of 24 September 2019, Netherlands and Others v Commission (T‑760/15 and T‑636/16, EU:T:2019:669), the Court stated that, when the Commission applies the arm’s length principle to check whether the taxable profit of an integrated undertaking pursuant to a tax measure (first comparable) corresponds to a reliable approximation of a taxable profit generated under market conditions (second comparable), the Commission can identify an advantage within the meaning of Article 107(1) TFEU only if the variation between the two comparables goes beyond the inaccuracies in the methodology used to obtain that approximation.

308    It follows that, to demonstrate that an advance tax ruling used to calculate an undertaking’s remuneration confers an economic advantage, the Commission must prove that that remuneration deviates from an arm’s length outcome to such an extent that it cannot be regarded as remuneration that would have been received on the market under competitive conditions.

309    Next, in paragraphs 201 and 211 of the judgment of 24 September 2019, Netherlands and Others v Commission (T‑760/15 and T‑636/16, EU:T:2019:669), the Court stated that mere non-compliance with methodological requirements does not necessarily lead to a reduction in the tax burden. It is further necessary for the Commission to demonstrate that the methodological errors in the advance tax ruling that it identified do not allow a reliable approximation of an arm’s length outcome to be reached and that they led to a reduction in taxable profit. The Court thus concluded that the mere finding of errors in the choice or application of the transfer pricing method does not, in itself, suffice to demonstrate the existence of an advantage and, therefore, to establish that there is State aid within the meaning of Article 107 TFEU.

310    It should be noted in that regard that although it is for the Commission to demonstrate specifically that the methodological error resulted in a reduction in the tax burden of the recipient of the advance tax ruling, the Court did not rule out that, in some cases, a methodological error may be such as to make it impossible to arrive at an approximation of an arm’s length outcome and may necessarily lead to an undervaluation of the remuneration that would have been received under market conditions.

311    Such a reading of the judgment of 24 September 2019, Netherlands and Others v Commission (T‑760/15 and T‑636/16, EU:T:2019:669), follows from the use of the expression ‘in principle’ in paragraphs 201 and 211 and from paragraph 212 of that judgment, in which it is stated that, in that case, the Commission had not put forward any evidence from which it could conclude, without a comparison being carried out with the result that would have been obtained using the method advocated by it, that the choice of the method endorsed in the advance tax ruling at issue necessarily led to a result that was too low.

312    In the light of the foregoing, and in the absence of a comparison in the contested decision between, on the one hand, the result that would have been obtained using the transfer pricing method advocated by the Commission and, on the other, the result obtained pursuant to the tax ruling at issue, the Commission’s approach, set out in recital 564 of the contested decision, at the end of which the Commission merely identifies errors in the transfer pricing analysis, is, in principle, insufficient to establish that there was in fact a reduction in LuxOpCo’s tax burden.

313    Nevertheless, it is necessary to ascertain whether, despite the assertion in recital 564 of the contested decision, the Commission’s subsidiary line of reasoning relating to the advantage contains specific evidence establishing that the errors in the transfer pricing analysis identified by the Commission led to a genuine reduction in LuxOpCo’s tax burden.

(b)    The first subsidiary finding concerning the advantage

314    In the third complaint in the second part of the first plea in Case T‑816/17 and in the fifth plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon contest the Commission’s first subsidiary finding concerning the existence of a tax advantage in favour of LuxOpCo (Section 9.2.2.1 of the contested decision). In essence, the Grand Duchy of Luxembourg and Amazon dispute the assertion that the transfer pricing method advocated by the Commission, namely the profit split method with the contribution analysis, was appropriate. They submit that the Commission was wrong to conclude that LuxOpCo performed unique and valuable functions. The Grand Duchy of Luxembourg notes that the Commission did not itself seek to apply the profit split method.

315    The Commission disputes those arguments.

316    According to the Commission, the contested decision correctly identified inappropriate methodological choices as regards the transfer pricing arrangement endorsed in the tax ruling at issue. The Commission maintains that, even if LuxSCS were considered to perform unique and valuable functions in relation to the intangible assets, so did LuxOpCo, with the result that the transfer pricing arrangement based on the profit split method would have been more appropriate and would have led to a higher remuneration for LuxOpCo than that confirmed by the tax ruling at issue.

317    In the present case, it should be noted that, in recitals 565 to 568 of the contested decision, the Commission stated, in essence, that, even if it were to be accepted that LuxSCS performed unique and valuable functions in relation to the intangible assets, the fact that LuxOpCo also performed such functions means that, in the present case, the profit split method, on the basis of the contribution analysis, should have been preferred over the TNMM.

318    In that regard, two different points must be clarified.

319    First, the Commission stated in recital 565 of the contested decision that, far from performing ‘routine’ management functions during the relevant period, LuxOpCo performed a range of unique and valuable functions in relation to the intangible assets and the Amazon group’s European business operations.

320    In that context, it should also be pointed out that the Commission did not find that some of LuxOpCo’s functions, as identified in the context of its own functional analysis, could have been described as ordinary or routine, or that such functions should, despite that routine nature, have been subject to additional remuneration.

321    Secondly, in recital 568 of the contested decision, the Commission concluded that application of the contribution analysis in the present case would have led to a remuneration for LuxOpCo corresponding to all of the functions it performs, its assets and its risks, as set out in Section 9.2.1.2 of the contested decision, and therefore to a remuneration greater than that endorsed in the tax ruling at issue. In so doing, the Commission took the view that the endorsement of the TNMM in the tax ruling at issue resulted in a lowering of LuxOpCo’s taxable income as compared to companies whose taxable profit reflects prices negotiated on the market. In particular, as is apparent from recital 566 of the contested decision, in the Commission’s view, it was inappropriate to endorse a transfer pricing arrangement according to which the entire residual profit generated by LuxOpCo in excess of [confidential] of its operating expenses was attributed to LuxSCS. 

322    Furthermore, it is apparent from paragraph 45 of the Commission’s answers to the Court’s written questions that, in its view, LuxOpCo’s remuneration was ‘necessarily’ higher with the application of the profit split method on the basis of the contribution analysis, since that method also remunerated LuxOpCo’s unique and valuable functions.

323    It is on the basis of the considerations in paragraphs 316 to 322 above that the complaints of the Grand Duchy of Luxembourg and Amazon challenging the first subsidiary finding are to be examined.

324    As is apparent from paragraph 314 above, the Grand Duchy of Luxembourg and Amazon raise three complaints contesting, first, the assertion that LuxOpCo performed unique and valuable functions, secondly, the finding that the tax ruling at issue erroneously endorsed the use of the TNMM and that the profit split method on the basis of the contribution analysis should have been used in the present case and, thirdly, the conclusion that use of the profit split method on the basis of the contribution analysis ‘necessarily’ resulted in higher remuneration.

(1)    LuxOpCo’s exercise of ‘unique and valuable’ functions

325    First and foremost, it should be noted that the first subsidiary finding expressly refers to Sections 9.2.1.2.1, 9.2.1.2.2, 9.2.1.2.3 and 9.2.1.2.4 of the contested decision, in which the Commission carried out its own functional analysis of LuxOpCo, and is directly based on the findings contained in those sections.

326    The findings made in Sections 9.2.1.2.1, 9.2.1.2.2, 9.2.1.2.3 and 9.2.1.2.4 of the contested decision and the finding that LuxOpCo performed unique and valuable functions are covered by the third complaint in the second part of the first plea in Case T‑816/17 and by the fifth plea referring to the second and third pleas in Case T‑318/18, which challenge the Commission’s functional analysis of LuxOpCo.

327    It is appropriate to examine together all of the arguments of the Grand Duchy of Luxembourg and Amazon challenging the merits of the Commission’s functional analysis of LuxOpCo and the finding that LuxOpCo performed unique and valuable functions.

328    First, according to the Grand Duchy of Luxembourg and Amazon, LuxOpCo did not have significant functions relating to the development, enhancement, management and exploitation of the intangible assets in Europe, but was responsible only for operating the business. The bulk of the development, management and enhancement of the intangible assets took place in the United States.

329    Next, the Grand Duchy of Luxembourg and Amazon submit, in essence, that LuxOpCo’s functions in connection with the Amazon group’s European business operations were routine contributions rather than unique and valuable contributions, since they were largely reliant on the intangible assets made available by LuxSCS. LuxOpCo’s functions in connection with the Amazon group’s European business operations were thus limited to management functions.

330    Lastly, as regards the assets and risks assumed by LuxOpCo, Amazon submits that the risks associated with LuxOpCo’s business operations were managed and mitigated by technology.

331    The Commission disputes those arguments.

332    First, it claims, in essence, that it is LuxOpCo which, with the support of the EU local affiliates, performed all the relevant unique and valuable functions relating to the three components of the intangible assets, namely technology, customer data and marketing.

333    Next, it submits that ‘human’ functions were not replaced by technology in the areas of pricing, the Amazon group’s relations with sellers and customers, stock management or inventory decisions. The Commission states that the fact that LuxOpCo used the intangible assets in the performance of its functions does not mean that those functions could not be considered unique and valuable.

334    Lastly, as regards the assets used and the risks assumed by LuxOpCo, the Commission states that the Grand Duchy of Luxembourg has not directly criticised the recitals of the contested decision dealing with those two issues and, moreover, the Commission disputes Amazon’s argument that the technology allowed the management of LuxOpCo’s risks without the need for any human intervention.

335    As a preliminary point, it is important to note that the examination of whether LuxOpCo did perform ‘unique and valuable functions’, as the Commission maintains, or only ‘routine functions’, as the Grand Duchy of Luxembourg and Amazon submit, is to be carried out in the light of the concepts referred to in paragraph 227 above. Although the concept of ‘unique and valuable functions’ is not expressly defined in the 1995 version of the OECD Guidelines, it is the opposite of the concept of ‘routine functions’, which are functions that can be easily benchmarked. As pointed out in paragraph 228 above, the concept of a ‘unique function’ refers to the situation where there is no comparable for a particular function. The concept of ‘valuable function’ relates in particular to the fact that the function in question generates significant revenue.

336    In addition, it should also be noted that, in so far as the Commission essentially based its functional analysis of LuxOpCo on the statements made by LuxOpCo’s employees in proceedings before the United States Tax Court (‘the Amazon employees’ testimonies’), the Grand Duchy of Luxembourg submits, in the application in Case T‑816/17, that those testimonies date from 2014 and relate to the Amazon group’s activities between 2005 and 2014, with the result that the Luxembourg authorities could not have been aware of that information at the time of issuing the tax ruling at issue.

337    It should be noted in that regard, first of all, that that argument put forward by the Grand Duchy of Luxembourg is contrary to the position adopted in its answers to the Court’s written questions. As regards the possibility of taking into account the opinion of the United States Tax Court and the 2017 transfer pricing report, the Grand Duchy of Luxembourg stated that, to determine whether LuxOpCo had received an advantage, it was necessary to examine the tax it would have had to bear in the absence of the tax ruling at issue, which necessarily entails taking into account information subsequent to the issuing of the tax ruling at issue.

338    It is true that, in the judgment of 24 September 2019, Netherlands and Others v Commission (T‑760/15 and T‑636/16, EU:T:2019:669, paragraphs 247 and 250), the Court held that the examination of the existence of an advantage conferred by an advance pricing arrangement that is the subject of the contested decision should be carried out in view of the context at the time that arrangement was concluded. Nevertheless, the Court based that finding on the fact that, in that case, the only measure contested by the Commission was the advance pricing arrangement.

339    In the present case, the measure of the Luxembourg authorities which is the subject of the contested decision is not only the tax ruling at issue, which was adopted in 2003, then extended in 2004 and 2010, but also the subsequent acceptance of LuxOpCo’s annual declaration, based on that decision, so that information relating to LuxOpCo’s actual situation during the relevant period was necessarily information that was available to the tax authorities when they adopted the measures forming the subject of the advance tax ruling.

340    It follows that, in the circumstances of the present case, the Commission cannot be criticised for having based its analysis on the Amazon employees’ testimonies. That evidence should therefore be taken into account for the purpose of assessing the complaints made by the Grand Duchy of Luxembourg and Amazon as regards the functional analysis of LuxOpCo.

(i)    The functions performed by LuxOpCo in relation to the intangible assets (Section 9.2.1.2.1 and recitals 449 to 472 of the contested decision)

341    In general, the parties disagree as to whether LuxOpCo performed significant ‘unique and valuable’ functions in relation to the intangible assets. According to the Commission, that was the case because LuxOpCo was responsible for adapting technology to the specific characteristics of the European market, for developing customer data and for activities concerning the marketing-related assets.

342    In Section 9.2.1.2.1 of the contested decision, the Commission found that, under the Licence Agreement, LuxOpCo was entrusted with performing ‘unique and valuable’ functions in relation to the intangible assets. Those functions included, according to the Commission, the development, enhancement and management of intellectual property generally and also as regards each of the three components of the intangible assets, namely technology, customer data and the registered trade mark through independent European technological and business innovations, the creation and management of customer data and the development and maintenance of the registered trade mark. Thus, in essence, according to the Commission, LuxOpCo did not merely use the technology to manage the EU websites, but actively contributed to its development, enhancement and management during the relevant period (recitals 449, 450 and 465 of the contested decision).

343    First, the Commission noted that LuxOpCo was granted an exclusive and irrevocable licence over the intangible assets and thus the right to continue to develop, enhance, maintain and protect them, although LuxSCS retained ownership of the derivative works created by LuxOpCo (recital 450 of the contested decision).

344    Secondly, the Commission found that, in general, LuxOpCo contributed to the development, maintenance and management of the intangible assets through the ‘EU IP Steering Committee’ (recitals 452 to 455 of the contested decision). According to the Commission, the EU IP Steering Committee was a forum where business and technology leaders employed by LuxOpCo and ASE met to discuss and recommend actions concerning the intangible assets in Europe, as presented to them by Amazon’s lawyers. The actual decisions on the development, enhancement, management and exploitation of the intangible assets were then taken by LuxOpCo’s and ASE’s members of that committee, in their capacity as decision-taking managers responsible for the Amazon group’s European retail and service business (recitals 452 to 455 of the contested decision).

345    Thirdly, the Commission stated that LuxOpCo contributed to the development of technology (recitals 466 to 472 of the contested decision). It noted that while the technology made available to LuxOpCo by LuxSCS was ‘Amazon US’s existing technology’, as ‘continually developed in the United States’ (recitals 456 and 461 of the contested decision), in its view, it was necessary for several functions in Amazon’s software, used in the United States, to be adapted so that it could be deployed in Europe. In particular, the Commission stated that, for the Amazon group’s European business operations to succeed, LuxOpCo developed, enhanced and managed that US technology with the support of its subsidiaries during the relevant period (recitals 456 to 460 of the contested decision). In addition, LuxOpCo and its EU local affiliates specifically developed significant technology for use in the European retail and service business. An example of such technology was the ‘European Fulfilment Network’ (EFN), which made it possible to pool the Amazon group’s inventory situated in various Member States and to combine the European fulfilment centres, so that, thanks to that tool, customers from any EU country could purchase items from any national website of the Amazon group in Europe (recitals 462 and 463 of the contested decision).

346    Fourthly, as regards customer data, the Commission found that, while legal ownership of customer data for the EU websites lay with LuxSCS, LuxOpCo performed active and critical functions in relation to the development, enhancement and management of customer data during the relevant period (recitals 466 to 468 of the contested decision). It noted, in that regard, that LuxOpCo actively accumulated that data as a service to LuxSCS and was responsible for its maintenance and ensuring compliance with the applicable data protection laws.

347    Fifthly, as regards the ‘registered trade mark’ (recitals 469 to 470), namely Amazon’s trade marks, in so far as they were registered in the European Union, the Commission noted that, although Amazon’s brand was well recognised and that strong global brand recognition was a major asset in attracting customers, the value of that brand name was of subordinate importance to the proper execution of the three pillars in the operation of the Amazon group’s European business, namely selection, price and convenience (‘the three pillars’). In view of the fact that the Amazon group’s brand and reputation were strongly reliant on the consistent delivery by LuxOpCo and the EU local affiliates of a highly satisfactory customer service, the value generation for the Amazon brand in Europe must be said, according to the Commission, to have taken place at the level of LuxOpCo and the EU local affiliates and not at the level of LuxSCS (recitals 469 and 470 of the contested decision). In addition, marketing activities were carried out by LuxOpCo and the EU affiliates, on the basis of local know-how (recital 472 of the contested decision).

348    Since the Grand Duchy of Luxembourg and Amazon dispute each of the five points of the Commission’s analysis, the arguments relating to each of those issues should be examined separately.

349    Before carrying out that examination, it should be noted, as a preliminary observation, that the Grand Duchy of Luxembourg and Amazon do not dispute that LuxOpCo performed certain functions associated with the development of the intangible assets, and in particular the technology; they dispute only that LuxOpCo played a significant part in the development of the intangible assets and thus performed unique and valuable functions in connection with them.

350    In their pleadings, the Grand Duchy of Luxembourg and Amazon concede that LuxOpCo performed development functions which they describe as ‘de minimis’, and that it played a role, albeit ‘secondary’, in creating value in respect of the intangible assets.

351    It therefore follows that, even according to the Grand Duchy of Luxembourg and Amazon, LuxOpCo did perform functions, even if only secondary, in connection with the development of the intangible assets.

–       The nature of the licence granted to LuxOpCo (recital 450 of the contested decision)

352    As already pointed out in paragraph 343 above, to demonstrate the importance of LuxOpCo’s functions in connection with the development of the intangible assets, the Commission highlighted, in recital 450 of the contested decision, the fact that LuxOpCo had ‘a right to further develop, enhance, [and] maintain … the intangibles for their entire lifetime’. In that context, the Commission stated in paragraph 100 of the defence submitted in Case T‑816/17 that LuxOpCo had an ‘exclusive right to continue to develop, improve [and] maintain Amazon’s intangibles’.

353    It is true that the term ‘exclusive’ was used in Section 2.1 of the Licence Agreement to describe the licence granted to LuxOpCo, which, in any event, covered only the European territory. Section 2.1(a) of that agreement reads as follows:

‘Exclusive intellectual property licence grant’

[LuxSCS] irrevocably grants [LuxOpCo], under all [LuxSCS’s] intellectual property rights in or comprising [LuxSCS’s] intellectual property, existing now or in the future, the following sole and exclusive right and licence to [LuxSCS’s] intellectual property during the term [of the Licence Agreement], …’

354    Nevertheless, in view of the contractual arrangements between LuxSCS and the US entities, in practice, LuxOpCo was not the only entity with the right to enhance and develop those intangible assets.

355    The rights LuxOpCo enjoyed with regard to the development of the intangible assets under the Licence Agreement were necessarily non-exclusive, in so far as the other parties to the CSA, namely ATI and A9, retained the right to develop, enhance and exploit the technology. The fact that the other parties to the CSA, namely ATI and A9, retained the right to develop and enhance the technology is, moreover, not disputed by the Commission.

356    In that regard, it should be noted that the rights LuxOpCo received under the Licence Agreement were not limited to intangible assets existing at the time that agreement was concluded, but also covered any future intangible assets created as a result of the continuous development, maintenance and enhancement efforts led by the US entities ATI and A9. This clearly shows that LuxOpCo was not the sole entity authorised to develop and enhance the intangible assets covered by the Licence Agreement.

357    Moreover, the Commission accepts that the assets, as transferred to LuxSCS on 1 January 2005 under the Buy-In Agreement, were subject to a ‘progressive replacement’ with intangible assets developed and enhanced subsequently under the CSA during the relevant period. It also acknowledges that the technology owned by LuxSCS and licensed to LuxOpCo was developed by the US entities, in particular ATI and A9.

358    It therefore follows from the foregoing that the Commission erred in finding that LuxOpCo enjoyed an exclusive right to continue to develop the intangible assets. That finding is not, however, sufficient to invalidate the Commission’s reasoning that LuxOpCo performed significant functions, or even unique and valuable functions in connection with the development of the intangible assets. The fact that LuxOpCo did not have an exclusive right of use over the intangible assets neither supports nor counters the Commission’s assertion that LuxOpCo performed unique and valuable functions in relation to the development of the intangible assets.

–       The EU IP Steering Committee (recitals 452 to 455 of the contested decision)

359    In essence, the Grand Duchy of Luxembourg and Amazon dispute that the EU IP Steering Committee had the role attributed to it by the Commission (paragraph 344 above). In their view, that committee did not take decisions concerning the development or enhancement of the intangible assets. In addition, not only were the majority of the participants in that committee US personnel, but EU IP Steering Committee decisions were in fact taken by employees of the Amazon group in the United States and, specifically, by the Vice-President, Intellectual Property.

360    The Commission disputes those arguments.

361    In recitals 452 and 453 of the contested decision, the Commission stated that the EU IP Steering Committee was created to provide technical and business guidance with regard to the development and deployment of intellectual property in Europe. The Commission stated that it is apparent from the ‘EU Policies and Procedures Manual’, first, that that committee met, inter alia, to examine Amazon’s intellectual property portfolio and business strategy as it related to the development and deployment of intellectual property. Secondly, that committee consisted of, inter alia, the Vice-President of EU Services, the EU Legal Director (employed by LuxOpCo), the Amazon group’s IP Counsel and the Vice-President, European Operations.

362    Next, the Commission stated that the fact that it was merely an advisory body, as Amazon explained during the administrative procedure, does not mean that its recommendations had no impact on the development, maintenance and management of the intangible assets. It noted that, in practice, the EU IP Steering Committee met, first, to make recommendations on filings to protect the intangible assets (and thereby LuxOpCo’s exclusive rights under the Licence Agreement between LuxSCS and LuxOpCo), secondly, to review the status of legal proceedings in Europe relating to the intangible assets and, thirdly, to provide training to European employees regarding the use of the technology and other intangible assets (recital 454 of the contested decision).

363    Lastly, on the basis of the testimony of the Vice-President, Intellectual Property, employed in the United States, the Commission concluded that the EU IP Steering Committee was a forum in which LuxOpCo’s business leaders met to discuss actions concerning the intangible assets, as presented to them by the Amazon group’s lawyers, and that the actual decisions on the development, enhancement, management and exploitation of the intangible assets were then taken by LuxOpCo’s and ASE’s members of that committee, in their capacity as decision-taking managers responsible for the Amazon group’s European retail and service business (recital 455 of the contested decision).

364    It is clear, on reading recitals 452 to 455 of the contested decision, that the Commission has failed to establish, in the contested decision, that the EU IP Steering Committee took significant decisions concerning the development or enhancement of the intangible assets.

365    First, the Commission acknowledges in recitals 452 and 453 of the contested decision that the role of the EU IP Steering Committee was limited, in so far as it merely provided ‘technical and business guidance’ and ‘assistance’ in strategic decision-making concerning the development of the intellectual property owned by LuxSCS, or the conclusion of several licensing agreements with third parties.

366    Moreover, it is apparent from the Amazon group’s EU Policies and Procedures Manual that the EU IP Steering Committee had no decision-making powers as such, but was merely a body intended to assist the development and deployment of intellectual property in Europe. That is also implicitly admitted by the Commission in recital 454 of the contested decision where the Commission refers to the ‘impact’ of that committee’s ‘recommendations’ on the development, maintenance and management of the intangible assets.

367    Next, it is apparent from recital 454 of the contested decision (see paragraph 362 above) and from the testimony of the Vice-President, Intellectual Property employed in the United States, referred to in recital 455 of that decision, that, in practice, the EU IP Steering Committee merely reviewed issues relating to the protection and maintenance of rights over the intangible assets and that the issue of the development or enhancement of the intangible assets as such was not discussed.

368    Lastly, to the extent that the EU IP Steering Committee may have been a forum for discussing the enhancement and development of the intangible assets, it is clear that decisions relating to the development of the intangible assets were not adopted by that committee but, in principle, were taken by employees of the Amazon group in the United States and specifically by the Vice-President, Intellectual Property. The Grand Duchy of Luxembourg’s assertion to that effect has not been challenged by the Commission.

369    In addition, as regards the composition of that committee, contrary to what is claimed by the Commission in its pleadings, employees with managerial functions within the US entities, and in particular the Vice-President, Intellectual Property employed by Amazon US, attended and even chaired meetings of the EU IP Steering Committee.

370    It follows from the foregoing that the Commission’s findings relating to the EU IP Steering Committee do not support its conclusion, in paragraph 455 of the contested decision, that decisions on the development and enhancement of the intangible assets were taken by LuxOpCo’s and ASE’s members of that committee, in their capacity as decision-taking managers responsible for the Amazon group’s European retail and service business.

371    At most, the Commission has succeeded in demonstrating that LuxOpCo performed functions in connection with the management and protection of the intangible assets and that LuxOpCo’s employees decided on appropriate measures, such as, for example, the filing of a patent, on the basis of recommendations discussed by the EU IP Steering Committee.

372    It follows from the foregoing that the Commission’s contentions relating to the EU IP Steering Committee are insufficient to establish that LuxOpCo performed development functions in respect of the intangible assets that could be described as ‘unique and valuable’.

–       LuxOpCo’s functions concerning the development of the technology

373    In recital 449 of the contested decision, the Commission noted that LuxOpCo’s functions included the development, enhancement and management of the technology.

374    In support of that assertion, it stated first of all, in recital 456 of the contested decision, in essence, that the technology developed in the United States could not, as it stood, be used in Europe and that adaptations were necessary in order to meet EU specifications. It stated that the development of the Amazon group’s European business operations required specific technology (different software, local adaptations).

375    Next, the Commission stated that LuxOpCo had the necessary technological resources to conduct research and development. It stated, inter alia, that approximately 60 people performed technology-related tasks, a Localisation and Translation team performed functions in relation to the customisation of the EU websites, adapting them to local preferences, and some 10 additional people were employed as ‘Technical Program Manager’, whose role was to translate into a technical description the technological needs identified by local retail business teams. According to the Commission, it was that process which allowed the technology to be continuously developed and adapted to the local market.

376    In that regard, it is true that the Commission acknowledged, in recital 461 of the contested decision, that the technical resources based in LuxOpCo were limited. Nevertheless, it stated that, in reality, the unique value of the technology would result from local know-how, identification of new business needs and their translation into the software project, not from the coding itself.

377    Lastly, the Commission stated that LuxOpCo also contributed to catalogue development, translation technology and local adaptations. Those tasks were carried out by teams previously placed in the EU local affiliates or by newly recruited personnel. In addition, it stated that the EU local affiliates also played an important role in the development of new technology specific to European national markets.

378    The Commission added that, in particular, LuxOpCo and the EU local affiliates contributed to the development of the EFN. That technology responded to a specific European business need by allowing European customers to buy items from any Amazon group EU website.

379    The Grand Duchy of Luxembourg and Amazon submit that LuxOpCo did not perform any significant development, enhancement or maintenance functions in relation to the technology intangible assets. They dispute in particular the Commission’s assertions that LuxOpCo played a major role in the development of the EFN and claim that that technology, although specific to Europe, was developed in the United States and that LuxOpCo did not participate in its design or creation.

380    The Commission contests those arguments and maintains its position expressed in recitals 456 to 465 of the contested decision, as set out in paragraphs 373 to 378 above.

381    As a preliminary point, it should be noted that, contrary to what is suggested by the Grand Duchy of Luxembourg and Amazon, the Commission did not state, in the contested decision, that LuxOpCo was the primary developer of the technology, either globally or only at European level, but stated that, as set out in recitals 449 and 465 of the contested decision, LuxOpCo actively contributed to the development, enhancement and management of the technology during the relevant period. Nor does the Commission dispute that the technology was continually developed in the United States.

382    In that regard, it is true that, as the Grand Duchy of Luxembourg and Amazon submit, technological platforms were developed principally by the US entities and provided to the European businesses in their final form. The software used by the various EU websites was also common software.

383    First, it is apparent from the material before the Court that the majority of decisions on the development of the intangible assets and prioritising the projects to be developed, including the technology specific to Europe, were taken in the United States.

384    Secondly, it is common ground that most of the technicians and engineers contributing to the development of the technology were based in the United States. As many as [confidential] Amazon group employees contributed to the development of the intangible assets, of which more than [confidential] were employed in the United States in technology-related jobs during the relevant period. The Commission does not dispute those figures. Furthermore, it is apparent from the documents in the file, and in particular from the various testimonies of the Amazon group employees, that the central services and US technicians were responsible for developing the tools specific to the European market.

385    Next, as regards LuxOpCo’s contribution to the development of technology, the following points should be noted.

386    In the first place, the Commission correctly ascertained that adaptations were sometimes necessary in order to implement the technology in Europe.

387    Although, as the Grand Duchy of Luxembourg and Amazon submit, the business model and underlying technology are the same in the United States as in Europe, it is apparent from the various testimonies of the Amazon group employees provided by the parties, and in particular from the testimony of the Vice-President International Retail, employed in the United States, that, because of the specific nature of the European market as compared with the US market, technology developed in the United States could not always be used as it was for the EU websites. Therefore, besides the EFN, which was technology that was specifically developed for the Amazon group’s European operations, adaptations or ‘localisations’ were necessary. It is apparent from the documents in the file that those adaptations included, inter alia, translation efforts [confidential].

388    In the second place, it is apparent from the Amazon group employees’ testimonies that, although adapting the technology to the European market was largely carried out in the United States, particularly in respect of software, LuxOpCo did, to some extent, contribute to those adaptations.

389    As Amazon also confirms in the application in Case T‑318/18, at the end of the relevant period, no more than approximately 60 people worked in technology-related positions in Luxembourg.

390    In that regard, it is apparent from the documents in the file and in particular from the testimonies of the Vice-President International Retail, employed in the United States, and the former Head of European Third Party Business (such as MarketPlace), that LuxOpCo began to use its own technicians no later than the year following the 2006 restructuring. Specifically, during the relevant period, LuxOpCo employed software developers who contributed to the development of specific programmes for the European business and worked on local adaptations.

391    Similarly, it is apparent from the testimony of the former Head of European Third Party Business (such as MarketPlace) that [confidential].

392    In that context, as regards the Localisation and Translation team, identified by the Commission in recital 459 of the contested decision, it is apparent from the material in the file that that team was responsible for adaptations to the EU websites, particularly in respect of translation, and that it contributed to the development of software. It is apparent from the Amazon group employees’ testimonies that [confidential]. However, even if those activities and that technology played an important role in LuxOpCo’s business operations, the fact remains that they played a minor role as compared with the rest of the technology developed in the United States.

393    Next, as regards the catalogue development activities identified by the Commission in recital 456 of the contested decision, it should be noted that, as Amazon and the Grand Duchy of Luxembourg submit, those activities did not include the design of the software underlying the catalogue, which was carried out in the United States. The catalogue activities developed in Luxembourg [confidential]. Local work on the development of the catalogue [confidential].

394    It is apparent from the testimony of the Vice-President International Retail, employed in the United States, [confidential]. Therefore, although LuxOpCo may have carried out work on the software relating to the catalogue, that work remained very limited as compared with developments carried out by the Amazon group’s central services.

395    It is apparent from the foregoing that LuxOpCo contributed to the development of the technology by making certain adaptations, principally relating to translation, and, to a lesser extent, the development of certain software and functionalities. In addition, the bulk of the work of adapting the technology to the European operations was largely up to the Amazon group’s central services. Accordingly, LuxOpCo’s contributions to the development of the technology could play only a minor role in the value creation of that technology. It follows that, although the Commission correctly stated, in recital 461 of the contested decision, that the adaptations needed in Europe were prepared in proximity to the local markets, it was wrong to attach such importance to LuxOpCo’s contributions to those adaptations and to draw the conclusion that those contributions were unique and valuable.

396    In the third place, as the Commission noted in recitals 460 to 461 of the contested decision, in addition to the adaptations carried out locally by the Luxembourg teams, LuxOpCo also contributed to the development of the technology through its involvement in the process of identifying new technological business needs and their translation into the software project.

397    In that regard, LuxOpCo had Technical Program Managers whose role was to translate business needs into technical descriptions so that a software developer could code them (see recital 460 of the contested decision). The Grand Duchy of Luxembourg does not dispute that the functional and technical specifications of the platforms and adaptations needed in Europe were prepared in proximity to the local markets.

398    It is true that, as the Commission stated, in essence, in recital 461 of the contested decision, the identification of business needs and the preparation of specifications played an important role in the development of the technology. It is apparent from the documents before the Court that the value of the Amazon group’s technology is based on its ability to serve the Amazon group’s three pillars, namely low prices, selection and convenience (see paragraph 347 above), and therefore to meet customers’ needs. The value of Amazon’s technology therefore also lies, to a certain extent, in the adaptation to local needs and, in particular, in the ability of the local teams to prepare specifications so that the technology can be adapted to consumer needs.

399    Nevertheless, it should be noted that, as Amazon maintains, only a dozen program managers were employed in Luxembourg, as against [confidential] in the United States, and [confidential]  of those employed in Luxembourg were not employed until the end of the relevant period [confidential].

400    Moreover, although LuxOpCo identified the technological needs of the business and the specifications relating to those needs, the design and creation of the technology were developed in the United States. In that context, contrary to what the Commission appears to suggest in paragraph 103 of the defence in Case T‑318/18, the activities of the US entities were not limited to mere coding activities, but were genuine development activities.

401    Lastly, as Amazon submits, the overwhelming majority of the strategic decisions regarding the development of the technology, particularly in relation to Europe, were taken by the US entities and not by LuxOpCo.

402    It follows that those developments and enhancements of the technology, for the purposes of improving customer experience, were carried out principally on the basis of the model developed in the United States, which is the same in Europe and the United States, and only to a lesser extent on the basis of the technical specifications that may have been prepared by the local teams.

403    It is apparent from the foregoing that, while the Commission was right to find that LuxOpCo contributed to the development of the intangible assets by preparing technical specifications in respect of the technology covered by the Licence Agreement, those contributions remained limited. Furthermore, in so far as, prior to 2006, those functions were previously performed by the local affiliates, it should be held, as the Grand Duchy of Luxembourg submits in paragraph 109 of the application in Case T‑816/17, that those functions cannot be regarded as unique, but are routine functions.

404    In the fourth place, the Commission found that LuxOpCo contributed to the development of the EFN, the only technology that was truly specific to the Amazon group’s European business.

405    As the Commission set out in recital 463 of the contested decision, the EFN is a combination of technological advances, such as the introduction of new functionalities, and logistical optimisations.

406    It is not disputed that the EFN played an essential role in the European retail and service business. As the Commission stated in recital 462 of the contested decision, and as Amazon itself claimed in its post-trial brief filed before the United States Tax Court, the creation of the EFN sought to address the problem of having multiple websites with country-specific fulfilment centres. That technology made it possible to combine the European fulfilment centres and pool inventory, thereby enabling customers throughout the European Union to purchase items from any Amazon website in Europe.

407    It is also apparent from the Amazon employees’ testimonies, as well as from Amazon’s post-trial brief filed before the United States Tax Court, that the EFN was crucial in order to be able to launch operations in two new European countries, namely Spain and Italy.

408    It is apparent from the material before the Court that the Commission was indeed entitled to find that LuxOpCo was involved in the development of the EFN (see paragraph 404 above).

409    In the post-trial brief filed in the proceedings before the United States Tax Court, Amazon itself [confidential]. According to that brief, ‘AEHT’ (Amazon Europe Holding Technology, a term corresponding to the official designation of LuxSCS) [confidential]. It should be noted, in that regard, that no distinction was made in the US proceedings referred to in paragraph 14 above between the various Luxembourg group entities and that, in those proceedings, the term ‘AEHT’ was used indiscriminately to refer to LuxOpCo and LuxSCS. As regards participating in the development of the EFN, however, it is clear that Amazon was referring to LuxOpCo and not LuxSCS.

410    That finding is, moreover, supported by the Amazon group employees’ testimonies, and in particular that of the Vice-President International Retail, employed in the United States, which confirms that, at that time, the Head of European Retail Business, employed by LuxOpCo, actively contributed to the development and conceptualisation of the EFN.

411    Nevertheless, it would be wrong to consider that LuxOpCo assumed responsibility for the entirety of the EFN development process.

412    First, it is apparent from the documents before the Court that the US headquarters played a decision-making role in the launch of the EFN project.

413    Secondly, as the Grand Duchy of Luxembourg submits and as is apparent from the Amazon group employees’ testimonies and the opinion of the United States Tax Court, the development of the EFN was carried out with the help of the US entities. Specifically, work relating to the development of the software underlying the EFN was carried out by technicians from the central teams and, inter alia, on the basis of specifications prepared by the LuxOpCo teams. Furthermore, the Commission does not dispute that, from an operations standpoint, the warehouse definitions and requirements on which that platform was based were also drawn up in the United States.

414    Although, in recital 462 of the contested decision, the Commission states that ‘the EFN was developed in Europe’ without providing any other details, the Commission did not entirely disregard the US entities’ contribution. The Commission cites, inter alia, in footnote 481 of the contested decision, one of the testimonies according to which the technology was developed in Europe with the help of central technology teams.

415    It therefore follows from the foregoing that, although the EFN was based, to a large extent, on the technology developed in the United States, LuxOpCo also actively contributed to the development of that technology. In the light of the importance of that technology for expanding the Amazon group’s European business, the Commission did not err in finding those contributions to be unique and valuable. Even if LuxSCS was the ultimate owner of that technology, the fact remains that its development was also the result of LuxOpCo’s efforts.

416    It is therefore apparent from the findings made in paragraphs 386 to 415 above that, aside from the EFN, to the development of which LuxOpCo actively contributed, the most important adaptations to the technology were carried out in the United States, in dialogue with the European teams which formulated their requirements, and that, in addition, certain minor adaptations could be made directly by the local teams.

417    It follows from all the foregoing that although the Commission was correct to find that, because of its contribution to the development of the EFN, LuxOpCo performed unique and valuable functions in relation to the technology, as for the remainder, it exaggerated the importance of LuxOpCo’s functions in connection with the development of the technology. Apart from the development of the EFN, LuxOpCo’s functions were limited principally to adaptations and the preparation of technical specifications. Accordingly, the conclusion reached in recital 465 of the contested decision, and in particular the assertion that LuxOpCo undertook significant enhancements to the technology, cannot be upheld in its entirety.

418    In so far as that finding is based on the Amazon employees’ testimonies, it does not appear necessary to examine in detail the arguments of the Grand Duchy of Luxembourg and Amazon that the Commission erroneously used the testimonies, inter alia to state that the technology (in particular the EFN) was developed in Europe, whereas it was developed in the United States. Even if the Commission’s analysis cannot be upheld in its entirety, the Commission’s errors are not such as to affect the finding that LuxOpCo did in fact contribute to the development of the intangible assets, in particular the EFN. 

–       Customer data (recitals 466 to 468 of the contested decision)

419    The Grand Duchy of Luxembourg and Amazon dispute the merits of the Commission’s finding that LuxOpCo performed active and critical functions in relation to the development, enhancement and management of customer data. In essence, they claim that customer data were collected automatically, using the technology developed in the United States, and without the involvement of LuxOpCo’s employees.

420    The Commission disputes those arguments.

421    As a preliminary point, it should be noted that the parties disagree as to whether LuxOpCo actively contributed to the development of the database containing customer information, such as, for example, sales records. It is therefore necessary to determine whether the accumulation of customer data during the relevant period and the protection of that data is attributable to LuxOpCo.

422    First, it should be noted that, as the Commission illustrated in Table 19 in the contested decision, during the relevant period, unique customer count by year increased significantly, from 17 million customers in 2006 to over 60 million in 2014.

423    Secondly, it must be noted, as the Commission has done, that customer data are a key asset for online retailers such as the Amazon group, in particular as regards marketing. Certain tools, in particular the use of recommendations and similarities technology, depend on customer data. Customer data are therefore a unique and valuable intangible.

424    Thirdly, it is common ground that LuxOpCo is the entity which collected the customer data and is also responsible for ensuring compliance with applicable data protection laws. The Grand Duchy of Luxembourg and Amazon do not dispute the assertion in recital 468 of the contested decision that LuxOpCo accumulated customer data from the EU websites as a service to LuxSCS.

425    Admittedly, it must be stated, as Amazon has done, that the collection of customer data was automated and that it was by means of the technology developed in the United States and supplied by LuxSCS to LuxOpCo that LuxOpCo was able to collect customer data.

426    Nevertheless, as the Commission states in paragraph 107 of its defence in Case T‑318/18, it is LuxOpCo that actively contributed to the accumulation of customer data by executing the three pillars of the Amazon group strategy (see paragraph 347 above), thereby attracting customers to the EU websites and collecting more customer data. The collection of customer data is necessarily dependent on the appeal of the Amazon group’s websites to customers. The increase in the numbers visiting the EU websites, and thus in the customer data collected, was itself linked to the execution by LuxOpCo of the three pillars referred to above, namely price, selection and convenience. While it cannot be disputed that the technology developed in the United States played a key role in the proper execution of the three pillars, the fact remains that LuxOpCo played an active and critical role in relation to the accumulation of new customer data and thus contributed to the development of those unique and valuable intangible assets.

427    Furthermore, the Commission correctly found, in recital 468 of the contested decision, that LuxOpCo had to ensure the maintenance of customer data and compliance with applicable data protection laws. Although protecting the customer database is an important activity for a business model that involves retail and services provided, inter alia, to end consumers, and for the reasons set out by the Commission in recital 466 of the contested decision, it is nevertheless a normal activity for any licensee that works with that type of database.

428    In the light of the foregoing, the Commission’s findings in recitals 466 to 468 of the contested decision should be confirmed, at least the finding that LuxOpCo performed active and critical functions in relation to the enhancement of customer data during the relevant period. In that regard, it should be held that, by accumulating customer data, which increased threefold between 2006 and 2014, as is apparent from paragraph 422 above, LuxOpCo contributed to the value of that intangible, which constitutes a unique and valuable intangible. Accordingly, LuxOpCo performed unique and valuable functions.

–       The ‘Amazon brand’ (recitals 469 to 472 of the contested decision)

429    The Grand Duchy of Luxembourg and Amazon challenge the Commission’s assertion that the value of the Amazon ‘brand’ is generated at the level of LuxOpCo and the EU local affiliates.

430    The Commission disputes those arguments.

431    First, it should be pointed out that, as the Commission implicitly acknowledged in recitals 469 and 471 of the contested decision, the Amazon brand is well recognised and enjoys a strong global brand recognition, which is a major asset in attracting customers. It should be noted that that reputation existed prior to the creation of LuxOpCo. Nevertheless, the Commission’s finding, set out in recitals 469 and 470 of the contested decision, that the brand name is not the focus of the Amazon group’s business model since the group’s business strategy focuses on the three pillars (price, convenience and product catalogue), should be confirmed. The value of the trade mark registered in Europe also depends on the ability to provide a good selection, attractive prices and optimised ease of use. Amazon itself also states that the value of the Amazon group’s marketing-related intangible assets in Europe depends on customer satisfaction.

432    Secondly, as the Grand Duchy of Luxembourg and Amazon submit, the technology plays an important, if not decisive, role in the development of Amazon’s brand name. First, the technology is central to the execution of the three pillars. Customer satisfaction is thus to a large extent dependent on the technology. Secondly, the technology plays a crucial role in marketing and maximising the possibility that the ‘Amazon’ name appears in potential customers’ searches. It is common ground that that technology is developed in the United States.

433    Nevertheless, the technology alone is not sufficient for the execution of the three pillars. The pillars were also executed by LuxOpCo through the adoption of strategic decisions necessary for the operation of the Amazon group’s European business.

434    In that regard, it is apparent from the post-trial brief in the proceedings before the United States Tax Court (see paragraph 14 above), that Amazon, in essence, claimed that [confidential].

435    In particular, Amazon stated in the post-trial brief filed before the United States Tax Court that [confidential] and, in a section dedicated to the marketing-related intangible assets, that [confidential].

436    In its opinion, the United States Tax Court concluded that ‘AEHT’ assumed sole responsibility for maintaining and developing the marketing-related intangible assets and paid, through cost sharing, for the technological improvements essential to maintaining the value of those intangible assets. It is true that, in the post-trial brief and in the opinion of the United States Tax Court, Amazon and that court refer to ‘AEHT’. Nevertheless, since no distinction is made, in the US proceedings referred to in paragraph 14 above, between the various Luxembourg group entities and the term ‘AEHT’ is used indiscriminately to refer to LuxOpCo and LuxSCS, ‘AEHT’ should be understood to refer, in this particular context, to LuxOpCo and not LuxSCS. Indeed, it is apparent from the file that LuxSCS performed no such functions.

437    In addition, it should be noted that neither the Grand Duchy of Luxembourg nor Amazon dispute the assertion in recital 472 of the contested decision that in Europe it was LuxOpCo and the local affiliates that handled the Amazon group’s online marketing based on their local know-how.

438    In those circumstances, even if the Amazon brand had an established reputation in Europe prior to the creation of LuxOpCo and enjoyed the international reputation of the Amazon group, it should be concluded that the Commission was entitled to determine that, during the relevant period, the maintenance and development of the brand value in part took place, at the very least, at the level of LuxOpCo and the European entities.

439    It should also be noted in that regard, as the Commission observes, that Amazon was known as a seller of books and media when it entered the European market and that the local teams had to make considerable efforts to build the brand’s reputation as regards other product categories.

440    Consequently, the Commission’s assessments in recitals 469 to 472 of the contested decision should be confirmed. In that regard, it should be held, in the light of the findings in paragraphs 433 to 438 above, that LuxOpCo contributed to the development of the value of the brand and the marketing-related assets and therefore performed valuable functions. However, there is no basis for finding that those functions were unique. In the light of the foregoing, it should be held that the Commission correctly concluded that LuxOpCo contributed to the development of the intangible assets in respect of each of their three components. As regards the development of the technology, LuxOpCo’s contribution mainly consisted of adaptations and preparing technical specifications. By contrast, LuxOpCo’s involvement in the development of the EFN may be equated with a unique and valuable function. In addition, LuxOpCo played an important role in the accumulation of customer data, inter alia through the execution of the three pillars, and therefore actively contributed to the development of that unique and valuable asset. As regards the marketing-related assets, LuxOpCo played an important role in increasing the Amazon brand’s reputation in Europe and therefore performed valuable functions. However, it is not apparent from the documents before the Court that those functions can be classified as unique.

441    It should therefore be concluded that the Commission correctly asserted, in recitals 414 and 415 of the contested decision, that the Luxembourg administration should have taken into account the fact that LuxOpCo performed unique and valuable functions in relation to the intangible assets. However, not all of LuxOpCo’s functions in connection with the intangible assets were unique and valuable.

442    It must also be noted in that regard that those functions should have been taken into account, if not when the tax ruling at issue was adopted, then at least at the time of its annual implementation. Any change in circumstances, which includes the performance of additional functions, should have been taken into account.

(ii) The functions performed by LuxOpCo in the operation of the Amazon group’s European retail and service business (Section 9.2.1.2.2 and recitals 473 to 499 of the contested decision)

443    As regards the functions performed by LuxOpCo in the operation of the Amazon group’s European retail and service business, in the contested decision, the Commission, in essence, highlighted that, according to the 2003 transfer pricing report and in practice, LuxOpCo acted as the EU headquarters and principal operator of that business. It was therefore LuxOpCo that was responsible for taking, and did take, strategic decisions in relation to the Amazon group’s European business operations (recitals 473 to 478 of the contested decision).

444    Specifically, the Commission found that LuxOpCo performed all strategic functions for the Amazon group’s online retail and service business during the relevant period and that it took all strategic decisions concerning merchandise and pricing, recorded sales and acted as the counterparty to customers. It therefore absorbed the corresponding costs and assumed the risks of sales and inventories (recital 475 of the contested decision).

445    The Commission therefore took the view that LuxOpCo independently took all relevant decisions concerning each of the three pillars of the Amazon group’s strategy in Europe (recital 478 of the contested decision).

446    The parties agree that LuxOpCo was operator of the online retail and service business. It is also not disputed that LuxOpCo served as the headquarters of the Amazon group’s European business.

447    Nevertheless, the Grand Duchy of Luxembourg and Amazon dispute that LuxOpCo performed significant exploitation functions. They take the view that LuxOpCo’s activities were largely reliant on the technology developed in the United States and amounted to European business management functions or routine business support functions, which generated little added value.

448    Specifically, the Grand Duchy of Luxembourg and Amazon maintain that human functions were replaced by the technology in the Amazon group’s business operations and that prices were fixed automatically. They add that it was impossible for individuals actively to set the price of the millions of items available on the Amazon group’s websites, that relations with sellers and customers were almost entirely automated, that the location of inventory and purchase collection instructions in the fulfilment centres were functions defined by technology and the fulfilment centre employees merely had to follow instructions given by the technology, and that inventory decisions (purchasing decisions, place of storage, etc.) were automated, so that employees only had to carry out the instructions given on the basis of the technology.

449    It is therefore necessary to examine whether the Commission was right to find that LuxOpCo performed significant exploitation functions and took strategic decisions in relation to each of the three pillars of the Amazon group’s strategy, with the result that it could not be treated as an undertaking that carried out routine management functions.

–       Selection (recitals 479 to 489 of the contested decision)

450    The Commission found, in recital 479 of the contested decision, that expanding and maintaining the largest possible selection was crucial to the Amazon group’s success in Europe. It added that the decision as to which product categories to sell was taken on the basis of market knowledge. Human intervention was therefore required; technology alone was insufficient.

451    In particular, the Commission stated, first, that LuxOpCo was able to rely on a large number of employees employed by the EU local affiliates, who assisted in creating the selection in Europe and expanding available new product lines on the basis of their knowledge of the local market, products and consumers (their ‘local know-how’) (recitals 470 to 482 of the contested decision). Secondly, LuxOpCo played a decisive role in the acquisition of other retailers active in the local market, the establishment of partnerships with suppliers and the establishment of third-party programmes for the development of MarketPlace (recitals 483 to 489 of the contested decision).

452    The Grand Duchy of Luxembourg and Amazon do not call in question the general finding in recital 483 of the contested decision that Amazon creates its selection through the acquisition of other retailers active in the market, partnerships with suppliers and third-party programmes, such as MarketPlace.

453    It is also apparent from the testimony of the former Head of European Third Party Business (such as MarketPlace) that the work of local recruiters was crucial in launching new products on the websites.

454    The Grand Duchy of Luxembourg and Amazon also do not dispute the fact that it was LuxOpCo, with the support of its EU local affiliates, that took on those ‘general’ activities of the Amazon group, acquired certain European retailers, entered into partnerships with European suppliers, defining policies and best practices for selecting and launching new categories, and determined standard contract terms for suppliers (see recital 485 of the contested decision).

455    It is also apparent from the testimony of the Senior Vice-President, International Retail, employed in the United States, [confidential]. Therefore, even if LuxOpCo did not decide on product categories entirely independently, it had a significant role in the launch of a new product line. In that regard, it is apparent from the testimony of the former Head of European Retail Business, employed by LuxOpCo, that, when launching a new product category, LuxOpCo was [confidential].

456    The 2006 restructuring therefore made it possible for the European operations to be managed more autonomously.

457    It should be concluded that the Commission was correct to find that LuxOpCo took important selection decisions and that the technology alone was insufficient for the execution of that pillar of the Amazon group’s strategy in Europe.

–       Price (recitals 490 to 493 of the contested decision)

458    As regards price, the Commission stated that, although pricing was automated and based on the use of an algorithm, this was no more than a tool to execute a certain pricing policy, which was determined by LuxOpCo in Europe.

459    In particular, the Commission stated that, without individual input based on local market know-how from the EU local affiliates, the pricing algorithm would not function effectively. In Europe, this was done by LuxOpCo with support from its EU local affiliates.

460    The Grand Duchy of Luxembourg and Amazon dispute those assertions. They submit, inter alia, that LuxOpCo’s business is significantly reliant on automation and that the involvement of LuxOpCo’s employees was minimal, particularly as regards prices.

461    It is not debatable that, as the Grand Duchy of Luxembourg claims, without the technology, LuxOpCo’s activities would have been less important.

462    Nevertheless, it should be noted that, as is apparent from recital 168 of the contested decision, until 2009 the Amazon group predominantly used manual pricing. It is only since 2009 that prices have been set by an algorithm. In that regard, the Commission was therefore correct to find that the algorithm was insufficient in itself as regards pricing and that it executed the pricing policy determined by LuxOpCo in Europe (see recital 490 of the contested decision).

463    First, as the Commission stated in recital 491 of the contested decision, without LuxOpCo’s individual input based on local market know-how from the EU local affiliates, the pricing algorithm could not function effectively.

464    Next, as the Commission identified in recital 492 of the contested decision, it is apparent from the Amazon group’s EU Policies and Procedures Manual that prices were established in Luxembourg by an EU Retail Pricing Committee composed of LuxOpCo’s senior employees. That committee was solely responsible for setting pricing guidelines for products offered by the Amazon group through its EU websites. LuxOpCo’s role in setting pricing guidelines is also confirmed by the testimonies of LuxOpCo’s employees.

465    In addition, it is not disputed that LuxOpCo employed a European pricing manager who had to agree prices, in particular when deviating from prices set by the algorithm, and a team that monitored [confidential]  globally. That team, which was set up in Luxembourg within LuxOpCo, monitored prices [confidential] and measured prices proposed globally, including in the United States (see recital 492 of the contested decision).

466    Lastly, the Grand Duchy of Luxembourg and Amazon do not call in question, in a substantiated manner, the fact that LuxOpCo had to develop a particular strategy to generate revenue and stand out from its competitors. In that regard, the influence of LuxOpCo and its EU local affiliates over pricing decisions was reflected in the promotions that applied to certain items sold on the EU websites (see recital 493 of the contested decision). It is not disputed that, during its first years of operation in Germany, the ‘Amazon.de’ website invented the ‘low price guarantee’ to incentivise customers to feedback price information in exchange for a rebate on their purchases and that, in the United Kingdom, certain types of price promotions common on the market, such as [confidential], made competing on price [confidential] more difficult [confidential]. As regards the sale of books in France and Germany, LuxOpCo set up a free shipping programme. These are examples of pricing decisions linked to market competitors and, therefore, strategic decisions specific to a retailer.

467    In the light of the foregoing, it should be concluded that LuxOpCo adopted strategic decisions on pricing and therefore performed significant functions. Even if pricing was dependent on the Amazon group’s technology, the fact remains that the involvement of LuxOpCo’s employees was also crucial.

–       ‘Convenience’ (recitals 494 to 499 of the contested decision)

468    As regards ‘convenience’, the Commission found that LuxOpCo was responsible for ensuring the convenience of the Amazon group’s retail and marketplace offer in Europe. Specifically, LuxOpCo had a team called the ‘Localisation and Translation’ team to check and adapt machine translation and, in particular, enable the merging of the different European catalogues to create and manage the EFN (recital 495 of the contested decision). Local logistical know-how, in particular as regards the delivery of items, was centred and developed in LuxOpCo and the EU local affiliates.

469    According to the Commission, the technology is a prerequisite for ‘convenience’, especially for translating the European product catalogues, delivering products and providing customer support services (see recitals 494 to 499 of the contested decision). LuxOpCo’s employees were involved with the product catalogues as well as with the delivery of products and customer support services. In that regard, the know-how belonged to LuxOpCo and the EU local affiliates (see recital 496 of the contested decision).

470    Those details are not, moreover, disputed by the Grand Duchy of Luxembourg and Amazon in a substantiated manner and should be confirmed.

471    It therefore follows from the foregoing that LuxOpCo took strategic decisions in relation to the Amazon group’s European business operations and thus had primary responsibility for executing the three pillars of the Amazon group’s strategy for that geographic area. The Commission was therefore right to find that, despite the importance of the technology, LuxOpCo also played a key role in the operation and expansion of the retail and marketplace business and, accordingly, performed valuable functions. However, it is not apparent from the documents before the Court that those functions were unique.

(iii) Assets used by LuxOpCo (recitals 500 to 505 of the contested decision)

472    As regards the assets used by LuxOpCo, the Commission noted, in recital 500 of the contested decision, that LuxOpCo used ‘significant assets’ to perform the functions described in Sections 9.2.1.2.1 (functions performed in relation to the intangible assets) and 9.2.1.2.2 (functions performed in the operation of the retail and service business) of its decision.

473    First, the Commission found, in recital 501 of the contested decision, that LuxOpCo owned and managed the entire inventory (the value of which amounted to EUR [confidential] billion during the relevant period) and that it held all the shares of ASE, AMEU and the EU local affiliates, which it provided with financing.

474    Secondly, the Commission stated that LuxOpCo’s cost structure demonstrated that significant assets were used to absorb the costs incurred in relation to the development and enhancement of the intangible assets in the performance of its functions (recital 502 of the contested decision). In particular, as regards the registered trade mark, LuxOpCo incurred significant direct marketing costs (for example, promotion costs) (recitals 503 to 504 of the contested decision). Consequently, the Commission found that, in the absence of any identifiable reimbursement of LuxOpCo, those costs should be considered as having been absorbed by LuxOpCo.

475    The Commission therefore concluded that LuxOpCo incurred the relevant costs associated with the economic exploitation of the intangible assets as well as their development, enhancement and management. According to the Commission, none of those costs can be regarded as costs incurred on behalf of LuxSCS (recital 505 of the contested decision).

476    The Grand Duchy of Luxembourg and Amazon, moreover, do not dispute those findings in a substantiated manner, except as regards the finding that LuxOpCo incurred the costs associated with the development, enhancement and management of the intangible assets. In that regard, it is sufficient to refer to paragraph 235 above. With the exception of the assertion that LuxOpCo incurred the costs associated with the CSA and the Buy-In Agreement, the Commission’s assessments in recitals 500 to 505 of the contested decision should be upheld.

(iv) Risks assumed by LuxOpCo (recitals 506 to 517 of the contested decision)

477    As regards the risks assumed by LuxOpCo, the Commission found, in recitals 506 to 517 of the contested decision, that LuxOpCo assumed, both under the terms of the agreements concluded with LuxSCS and in practice, the risks associated with the development, enhancement and management of the intangible assets. LuxOpCo also controlled and managed all the relevant business and entrepreneurial risks in relation to the Amazon group’s European retail and service business, including, but not limited to, credit risk, collections risk, inventory risk, market risk, risk of loss, and risks relating to maintaining a workforce capable of selling goods and providing services in an efficient and timely manner.

478    The Commission then dismissed Amazon’s assertion, made during the administrative procedure, that LuxOpCo heavily relied on the technology to manage or assume risks (recitals 506 and 508 of the contested decision).

479    First, according to the Commission, even if the technology minimised the risks, LuxOpCo assumed those risks because of its role as the European headquarters and the operator of the European retail and service business (recitals 509 and 510 of the contested decision). It observes that, even if LuxOpCo did rely on the technology to manage its business risks, it would only be due to a strategic decision taken by LuxOpCo (recital 511 of the contested decision).

480    In addition, it has not been established, either in the 2003 transfer pricing report or in any other document, that the strategic, financial and operational risks faced by LuxOpCo in its day-to-day operations were managed by risk management group policies (recital 512 of the contested decision). On the contrary, risks such as the loss of business were managed at the local level, with LuxOpCo as the responsible principal operator in Europe (recitals 513 to 515 of the contested decision).

481    The parties disagree as to which entity actually assumed the risks associated with the intangible assets and, specifically, whether LuxOpCo used assets to absorb the costs associated with the development of the intellectual property. In addition, according to the Grand Duchy of Luxembourg, part of the risks and functions identified by the Commission were assumed by the EU local affiliates and not by LuxOpCo.

482    As regards the risks associated with the development, enhancement and management of the intangible assets, to assert, as the Commission did in recital 507 of the contested decision, that LuxOpCo assumed those risks, both under the terms of the agreements concluded with LuxSCS and in practice, is unconvincing.

483    As the Grand Duchy of Luxembourg correctly stated in paragraph 104 of the application in Case T‑816/17, even if, pursuant to the Licence Agreement, LuxSCS transferred to LuxOpCo a certain number of risks associated with the operation of the business, the fact remains that LuxSCS, which was the legal owner of the right to use the intangible assets during the relevant period, still assumed the risks associated with the intangible assets, as it had to fulfil its obligation under the Buy-In Agreement and the CSA to make the Buy-In payments and the payments owing under the CSA to ATI and A9.

484    That finding is not disputed in a substantiated manner by the Commission. As regards LuxSCS’s financial capacity to absorb risks if they materialised, the Commission has not been able to substantiate its assertion that LuxSCS did not have substantial capital. As regards LuxSCS’s initial capital, which the Commission dismissed as irrelevant in recital 445 of the contested decision, it is common ground that, at least for 2006, it was as a result of its capital that LuxSCS was able to absorb the losses sustained during its first years of operation without LuxOpCo’s intervention. Accordingly, in 2006, the amount of the royalty paid by LuxOpCo to LuxSCS was much lower than the Buy-In and cost-sharing payments made by LuxSCS.

485    In those circumstances, it should be held that LuxOpCo assumed, at most, part of the risks associated with the existence, development, enhancement and management of the intangible assets.

486    By contrast, as regards the other risks referred to in recitals 507 to 517 of the contested decision, namely the risks inherent in an online retail and service business, the Grand Duchy of Luxembourg and Amazon do not challenge in a substantiated manner the assertion that LuxOpCo assumed those risks, such as the risk associated with the loss of business itself (see recital 514 of the contested decision) and risks associated with holding title of unsold inventory in Europe, housing servers and maintaining call centres, bad debts and the failure to fulfil contracts concluded with customers. In particular, while the Grand Duchy of Luxembourg and Amazon argue that the technology minimised LuxOpCo’s risks in connection with the operation of the business, and in particular inventory risk, it is clear that, as the Commission stated in recital 510 of the contested decision, the technology did not fully eliminate those risks. The Commission’s assertions should therefore be confirmed.

(v)    Conclusions on the functional analysis of LuxOpCo

487    In the light of the foregoing, the Commission’s functional analysis of LuxOpCo is not entirely convincing.

488    First, having regard to the information gathered by the Commission for the purpose of adopting the contested decision, as set out in the contested decision, part of which has been confirmed above, it cannot be ruled out that LuxOpCo did in fact perform certain unique and valuable functions in relation to the intangible assets. That is true in respect of the development of the EFN and the accumulation of customer data. For the remainder, as regards the marketing assets, although LuxOpCo’s functions are valuable, it has not been established that they can be regarded as unique.

489    Secondly, the analysis of LuxOpCo’s functions as an online retailer and service provider can be confirmed in all material respects. In that regard, contrary to what is claimed by the Grand Duchy of Luxembourg and Amazon, LuxOpCo did not merely carry out simple ‘management’ functions, but operated as an online retailer and assumed the risks inherent in that business. Such functions were ‘valuable’ because they were capable of contributing significantly to LuxOpCo’s turnover and therefore to the Amazon group’s business model. Nevertheless, such functions cannot be described as unique. To the extent that the Grand Duchy of Luxembourg and Amazon have stated that an undertaking may be regarded as an entity that performs routine functions (as opposed to an entity that performs unique and valuable functions) when those functions can easily be ‘benchmarked’ (see paragraph 225 above), it is sufficient to note that the [confidential] report, referred to by the main parties, concerns online retail businesses and addresses the issue of market remuneration for such retailers.

490    In the light of the foregoing, although the Commission was entitled to find that some of LuxOpCo’s functions in connection with the intangible assets were unique and valuable, its argument that LuxOpCo’s functions in connection with its business operations were unique and valuable is not completely convincing. Although the Commission was correct to find that LuxOpCo performed more functions than those identified for the purposes of adopting the tax ruling at issue, namely routine ‘management’ functions, it erred in finding that LuxOpCo’s functions relating to its retail business were unique and valuable.

491    That conclusion is not called in question by any of the other arguments put forward by the Grand Duchy of Luxembourg or Amazon.

492    First, the Grand Duchy of Luxembourg and Amazon submit that the fact that prior to the 2006 restructuring LuxOpCo’s functions were performed by the EU local affiliates and those companies were remunerated at cost plus demonstrates that LuxOpCo’s functions are merely routine. It is sufficient to note in that regard that the very purpose of the 2006 restructuring was to create a headquarters for the Amazon group’s European business, assigning to LuxOpCo much more significant functions than those of the EU local affiliates.

493    Secondly, the Grand Duchy of Luxembourg takes issue with the Commission for attributing to LuxOpCo the functions performed by the EU local affiliates. It is sufficient to note in that regard that the EU local affiliates operated as service providers within LuxOpCo and were remunerated as such. The functions they performed were therefore performed on behalf of LuxOpCo and at LuxOpCo’s risk. The Commission was therefore entitled to attribute those functions to LuxOpCo.

(2)    Choice of method

494    As noted in paragraph 317 above, the Commission found, in essence, that it was incorrect to use the TNMM to determine the amount of the royalty and the remuneration for LuxOpCo and that the profit split method on the basis of the contribution analysis should have been used.

495    Specifically, in recital 567 of the contested decision, the Commission stated that, where both parties to the intra-group transaction make unique and valuable contributions to that transaction, the profit split method is usually considered to be a more appropriate transfer pricing method because in such a case independent parties would be expected to share the profits of the transaction in proportion to their respective contributions.

496    In that context, the Commission stated, referring to recital 256 of the contested decision, that the OECD Guidelines distinguish between two approaches to dividing profits, namely the contribution analysis and the residual analysis. As regards the residual analysis, the Commission stated that it applied where a party is remunerated for its routine functions in addition to the remuneration it receives for its unique and valuable contributions to the transaction. As regards the contribution analysis, the Commission stated that the combined profits are split on the basis of the relative value of the functions performed (taking into account assets used and risks assumed) by each of the parties to the intra-group transactions.

497    The Commission added that, where both parties to the controlled transaction make unique and valuable contributions and there are no less complex transactions that need to be priced separately, it was more appropriate to apply the contribution analysis for the attribution of combined profits. The residual analysis is appropriate if some less complex transactions exist.

498    On the basis of those observations, the Commission concluded that, because both LuxSCS and LuxOpCo were considered as performing unique and valuable functions in relation to the intangible assets, the contribution analysis should be preferred over the residual analysis.

499    The parties to the proceedings disagree as to whether the Commission was right to find that the profit split method on the basis of the contribution analysis was appropriate in the present case and, consequently, that the decision to use the TNMM should not have been endorsed in the tax ruling at issue.

500    First and foremost, it should be borne in mind, as regards the choice of transfer pricing method itself, that, as set out in paragraph 202 of the judgment of 24 September 2019, Netherlands and Others v Commission (T‑760/15 and T‑636/16, EU:T:2019:669), the various methods for setting transfer prices endeavour to attain profit levels reflecting arm’s length transfer prices and that it cannot be concluded, as a rule, that one method does not allow a reliable approximation of an arm’s length outcome to be reached.

501    Furthermore, when examining a measure in the light of the arm’s length principle, the Commission is free to use a method other than that endorsed by the national tax authorities. It is apparent from paragraph 154 of the judgment of 24 September 2019, Netherlands and Others v Commission (T‑760/15 and T‑636/16, EU:T:2019:669), that, although the Commission cannot be criticised for having used a transfer pricing method that it considers appropriate in a given instance in order to examine the level of transfer pricing for a transaction or for several closely related transactions that is part of the contested measure, the Commission is nevertheless required to justify its choice of methodology.

502    In the present case, owing to the fact that both LuxSCS and LuxOpCo performed some unique and valuable functions, the Commission cannot be criticised for finding that, overall, the profit split method may have been appropriate for examining the controlled transaction.

503    Nevertheless, the Commission’s conclusion that the contribution analysis should have been applied is not convincing. In that regard, as is implicitly but necessarily apparent from recitals 256, 567 and 568 of the contested decision, the Commission proceeded on the basis that the fact that the parties to the controlled transaction performed unique and valuable functions as well as routine functions called for application of the profit split method on the basis of the residual analysis, whereas the profit split method on the basis of the contribution analysis was appropriate only when the undertakings concerned performed unique and valuable functions exclusively. As is apparent from paragraphs 488 and 489 above, although the Commission was right to find that some of LuxOpCo’s functions in relation to the intangible assets were unique and valuable, it erred in finding that LuxOpCo’s functions in connection with the business operations were unique and valuable. In addition, the Commission has not established that there was no comparable for LuxOpCo’s business operations, or that the remuneration for those functions could not have been set separately.

504    In addition, it follows implicitly but necessarily from paragraphs 3.6 and 3.8 of the 1995 version of the OECD Guidelines that the choice of profit split method, including on the basis of the contribution analysis, depends decisively on having identified external data from independent undertakings in order to determine the value of the contribution made by each undertaking associated with the transactions. However, the Commission did not seek to identify whether any such reliable data were available in order to conclude that the profit split method on the basis of the contribution analysis could be chosen in the present case.

505    In addition, the Commission has not explained how LuxOpCo’s functions in connection with the development of the intangible assets were such that it was the contribution analysis, and not the residual analysis, that was the appropriate method in the present case (see recital 568 of the contested decision).

506    It follows that, when it disregarded the profit split method on the basis of the residual analysis and opted for the contribution analysis, the Commission did not duly justify its choice of methodology and therefore failed to satisfy the requirements set out in paragraph 501 above.

507    In so far as the Commission based its first subsidiary finding on the fact that the contribution analysis was the only appropriate method in the present case, the error found in paragraph 506 above vitiates the Commission’s reasoning relating to proof of the existence of an advantage.

508    The Court nevertheless considers it necessary to continue the assessment of the applicants’ arguments that the Commission has failed to establish that LuxOpCo’s remuneration would have been higher had the profit split method on the basis of the contribution analysis been used.

(3)    The demonstration that LuxOpCo’s remuneration would have been higher had the profit split method been used (application of the contribution analysis)

509    As stated above, the Commission found in recital 568 of the contested decision that application of the profit split method on the basis of the contribution analysis would necessarily have led to LuxOpCo’s remuneration being higher. The Commission thus considered that, by endorsing the royalty calculation method proposed by Amazon which resulted in an undervaluation of LuxOpCo’s remuneration, the tax ruling at issue conferred an advantage on LuxOpCo. The Commission nevertheless stated, in recital 564 of the contested decision, that it had not sought to determine a precise arm’s length remuneration for LuxOpCo.

510    In that regard, it should be borne in mind that, as noted in paragraphs 317 to 320 above, the error identified by the Commission is not based on the finding that some of LuxOpCo’s functions, including routine or everyday functions, were not remunerated. By contrast, the Commission proceeded on the assumption that LuxOpCo assumed a series of unique and valuable functions and that, accordingly, the profit split method on the basis of the contribution analysis should have been applied.

511    The parties disagree as to whether the Commission succeeded, on the basis of the finding referred to in paragraph 509 above, in proving the existence of an advantage in favour of LuxOpCo.

512    To prove the existence of an advantage, the Commission had to show that LuxOpCo would have received greater remuneration had the profit split method been applied as compared with what it actually received pursuant to the tax ruling at issue.

513    As stated in paragraph 310 above, it is for the Commission to demonstrate specifically that the methodological errors it identified in the advance tax ruling did not allow a reliable approximation of an arm’s length outcome and that those errors led to a reduction in taxable profit, without it being ruled out that, in certain cases, a methodological error may be such that an approximation of an arm’s length outcome cannot be reached and it necessarily leads to an undervaluation of the remuneration that would have been received under market conditions.

514    The same considerations must apply where the Commission takes the view that it has identified an error in the functional analysis. It cannot be ruled out that, notwithstanding an error in the functional analysis, the remuneration calculated may be no different from the arm’s length outcome that would have been determined had the functions been correctly taken into account. In that context, it should be noted that, although certain functions were not correctly identified and were not taken into account in calculating the remuneration, it is likely that those functions would have been remunerated and that, because those additional functions were not taken into account, the remuneration of the undertaking in question should have been greater. Nevertheless, the Commission cannot reach such conclusions without specifically examining whether, in the present case and in the light of the specific characteristics of the transaction at issue, the error identified in the functional analysis may have led to additional remuneration.

515    Lastly, when examining, pursuant to the power conferred on it by Article 107(1) TFEU, a tax measure granted to an integrated undertaking, the Commission must compare the tax burden of that integrated undertaking resulting from the application of that tax measure with the tax burden resulting from the application of the normal rules of taxation under national law of an undertaking, placed in a comparable factual situation, carrying on its activities under market conditions (see, to that effect, judgment of 24 September 2019, Netherlands and Others v Commission, T‑760/15 and T‑636/16, EU:T:2019:669, paragraph 149). When identifying errors in the functional analysis on which a tax measure such as the tax ruling at issue is based, the Commission must compare the tax burden of the integrated undertaking resulting from the application of the tax measure in question with the tax burden of an undertaking active on the market performing functions comparable to those of the integrated undertaking, as identified by the Commission. It should be stated in that regard that such a comparison does not mean that the Commission must necessarily carry out a new analysis with the same level of detail as that carried out by the Member State for the purpose of adopting the tax measure in question. However, even if it does not have to carry out such an analysis, the Commission is at least required to identify a number of specific factors from which it can be concluded with certainty that the arm’s length remuneration for the company’s functions, as identified by the Commission, was necessarily greater than the remuneration received pursuant to the tax measure in question.

516    As a result of what is set out in paragraph 515 above, the Commission should have compared LuxOpCo’s remuneration obtained using the method endorsed in the tax ruling at issue with the arm’s length remuneration that should have been received in the light of LuxOpCo’s functions identified by the Commission itself in its own functional analysis. Even if it did not actually apply the profit split method on the basis of the contribution analysis, the Commission should at least have identified a number of factors from which it could be concluded that LuxOpCo’s remuneration calculated pursuant to the tax ruling at issue was necessarily lower than the remuneration that a company operating on the free market would have received had that company performed functions comparable to those identified by the Commission in its functional analysis. Specifically, if the Commission considered that the contribution analysis was the appropriate calculation method, instead of making mere unverified assumptions of the result that would have been obtained using the profit split method on the basis of the contribution analysis, it should have examined whether, on the free market, taking account of functions and risks comparable to those assumed by LuxOpCo under the controlled transaction (in particular its functions and risks as an online retailer) did in fact result in a share of profits (for an online retailer comparable to LuxOpCo) that would have been greater than that to which LuxOpCo was entitled under the calculation method referred to in the tax ruling at issue. Although the Commission did not have to give precise figures, it was required, at the very least, to provide verifiable evidence in that regard.

517    In the present case, the Commission merely found in the contested decision that, had the profit split method on the basis of the contribution analysis been used, LuxOpCo would have received greater remuneration, without seeking to apply that method. However, the Commission cannot make assumptions about what the result of applying a particular method would be or what remuneration would have been awarded to a particular function. By contrast, as already pointed out in paragraphs 515 and 516 above, the Commission must establish that the remuneration endorsed in the tax ruling at issue was lower than the reliable approximation of arm’s length remuneration that would have been received using the profit split method with the contribution analysis.

518    In that regard, it should be noted that the Commission did not seek to ascertain the correct allocation key for the combined profits of LuxSCS and LuxOpCo that would have been appropriate had those parties been independent undertakings, or even to identify specific factors from which it could be determined that LuxOpCo’s functions in connection with the development of the intangible assets or with its role as headquarters would have conferred entitlement to a greater share of profits as compared with the share of profits actually obtained pursuant to the tax ruling at issue.

519    In addition, it should be borne in mind that, even though LuxOpCo’s contributions in connection with the intangible assets (contribution to the development of a small part of the technology, contribution to the expansion of the customer database and to brand value) and with the operation of the business are very real, none of the considerations in the contested decision make it possible to measure the contribution of those functions as compared with LuxSCS’s functions (the provision of technology, which plays a crucial role in the operation of the Amazon group’s business and in generating profits). Accordingly, without a detailed analysis it is not possible to prejudge to what extent LuxOpCo’s contributions would give it a greater share of the profits generated in Europe than that obtained pursuant to the tax ruling at issue.

520    In those circumstances, it must be held that the Commission has not demonstrated the existence of an advantage, but has, at most, demonstrated the probability of the existence of an advantage.

521    It is true that, as the Commission mentioned at the hearing, the fact that certain assets are transferred free of charge without those functions being taken into account in LuxOpCo’s remuneration, endorsed by the tax ruling at issue, demonstrates that LuxOpCo’s remuneration is lower than what an independent undertaking would have received under market conditions. Nevertheless, it is clear that the contested decision contains no such reasoning.

522    Admittedly, it cannot be ruled out that, if LuxOpCo performed more functions than those taken into account for the purposes of calculating its remuneration as endorsed in the tax ruling at issue, LuxOpCo would have been entitled to additional remuneration. Nevertheless, the Commission’s reasoning, as set out in the contested decision, remains theoretical and is not sufficient to establish that LuxOpCo actually received an advantage as a result of applying the method for calculating its remuneration endorsed by the tax ruling at issue.

523    It is true that the Commission stated, in its replies to the Court’s questions and at the hearing, that, pursuant to the tax ruling at issue, LuxOpCo received only approximately 20% of the combined profits of LuxOpCo and LuxSCS and that, if the profit split method on the basis of the contribution analysis had been applied, taking into account its unique and valuable functions, it would necessarily have received a greater share of those combined profits.

524    Nevertheless, first, it should be noted that the assertion that LuxOpCo received only approximately 20% of the combined profits does not appear in the part of the contested decision relating to the first subsidiary finding, or in any part of the contested decision. At most, that figure can be calculated on the basis of the various figures contained in the contested decision. In that regard, that figure of approximately 20% of the combined profits of LuxOpCo and LuxSCS corresponds to LuxOpCo’s operating profits after deduction of the royalty paid to LuxSCS pursuant to the tax ruling at issue.

525    Secondly, the Commission adduces no specific evidence in the contested decision to prove that the fact that approximately 80% of the combined profits of LuxOpCo and LuxSCS were allocated to LuxSCS as remuneration for its contribution, namely the provision of the intangible assets, is not at arm’s length, or that the allocation of approximately 20% of the combined profits does not constitute sufficient remuneration in the light of the contributions made by LuxOpCo.

526    Thirdly, as Amazon stated at the hearing, the Commission did not state in the contested decision that additional remuneration should have been awarded to LuxOpCo over and above the remuneration calculated under the TNMM. The finding that the profit split method with the contribution analysis had to be applied means disregarding the remuneration initially calculated pursuant to the TNMM and making a new calculation. It cannot be ruled out that LuxOpCo’s remuneration calculated using the profit split method with the contribution analysis may be lower than the remuneration calculated under the TNMM, as endorsed by the tax ruling at issue.

527    It should be noted in that respect that, in the administrative procedure, Amazon submitted to the Commission a report on the mark-up for undertakings in the online retail business, namely the [confidential] report. The average mark-up for online retailing was 0.5% of online retailers’ total costs. Without it being necessary to ascertain whether, in the light of that evidence, Amazon was entitled to assert, as it did at the hearing, that that rate showed that LuxOpCo’s remuneration during the relevant period was ‘comfortable’, it is clear that, in the light of that report, the Commission should have examined in greater detail whether LuxOpCo’s remuneration corresponded to an arm’s length outcome for its functions as an online retailer. Without such an examination, it cannot be claimed with certainty that LuxOpCo could have received greater remuneration for its functions related to the business operations.

528    Furthermore, as regards the development activities carried out by the Amazon group, LuxOpCo’s functions in connection with the intangible assets, in particular as regards technology, remained limited. It is therefore not clear that those functions were such that the share of profits attributable to LuxOpCo should have been greater than 20% of the combined profits of LuxSCS and LuxOpCo.

529    Consequently, in the absence of information on the allocation key that should have been used, it is not possible to identify the order of magnitude of the remuneration that LuxOpCo would have received under arm’s length conditions, nor, a fortiori, to determine whether that remuneration is lower or higher than that obtained pursuant to the tax ruling at issue.

530    It follows that the Commission has not succeeded in establishing that, had the profit split method on the basis of the contribution analysis been applied, LuxOpCo’s remuneration would have been greater. Accordingly, the first subsidiary finding does not support the conclusion that the tax ruling at issue conferred an economic advantage on LuxOpCo. Apart from the fact that the Commission did not seek to determine what LuxOpCo’s arm’s length remuneration would have been in the light of the functions identified by the Commission in its own functional analysis, the first subsidiary finding contains no specific evidence to establish to the requisite legal standard that the errors in the functional analysis and the methodological error identified by the Commission, relating to the choice of method itself, actually led to a reduction in LuxOpCo’s tax burden.

531    None of the other arguments put forward by the Commission can call those findings into question.

532    First, it is true that, as the Commission stated at the hearing, there is a difference between proving the advantage and quantifying it. Therefore, it cannot be ruled out that it may be possible to prove that a methodological error necessarily leads to lower remuneration without that reduction in remuneration being quantified. Nevertheless, as stated in paragraph 529 above, in the present case, the contested decision contains no evidence capable of proving that use of the profit split method with the contribution analysis, instead of the TNMM, would necessarily have led to higher remuneration.

533    Secondly, it is true that, as the Commission stated at the hearing, the Grand Duchy of Luxembourg and Amazon do not deny that, if LuxOpCo had substantially developed the intangible assets, that would have led to additional remuneration.

534    Nevertheless, as Amazon stated at the hearing and as set out in paragraph 526 above, the Commission’s argument does not amount to finding that remuneration additional to that calculated pursuant to the tax ruling at issue should have been calculated. The Commission took the view that the contribution method with the residual analysis was not appropriate in the present case. It follows that the Commission’s argument is limited to contending that, had LuxOpCo’s remuneration been calculated using the profit split method according to the contribution analysis, that would have resulted in higher remuneration than that obtained under the TNMM. In so far as the contribution analysis would mean not taking into account the remuneration initially calculated on the basis of the TNMM and making a completely new calculation, it cannot be ruled out that LuxOpCo’s remuneration, calculated using the profit split method with the contribution analysis, may be lower than the remuneration calculated under the TNMM, as endorsed by the tax ruling at issue.

535    Thirdly, the Commission’s argument, set out in its pleadings, that applying the profit split method would necessarily have led to higher remuneration for LuxOpCo in so far as it would have led to a distribution of the residual profit between LuxOpCo and LuxSCS, rather than the allocation of the entire residual profit to LuxSCS, and the argument that it was not necessary to apply the profit split method in order to prove the existence of an advantage since that method consists of sharing the residual profit, must be rejected.

536    Such a residual profit split is relevant only when using the residual analysis. It is clear from the contested decision, and in particular from recitals 567 and 568 thereof, that the Commission took the view that only the contribution analysis, and not the residual analysis, could effectively be used in the present case for the purpose of applying the profit split method. As stated in paragraph 534 above, the contribution analysis consists of directly allocating the combined profits between the various parties to the transaction and does not take account of the remuneration initially calculated for LuxOpCo. Accordingly, it cannot be assumed that applying the profit split method would necessarily have led to higher remuneration for LuxOpCo.

537    It therefore follows from all of the above that the Commission has not succeeded in proving, in its first subsidiary finding, that the tax ruling at issue conferred an advantage on LuxOpCo.

538    The pleas and arguments challenging the first subsidiary finding should therefore be upheld.

(c)    The second subsidiary finding concerning the advantage

539    In the second subsidiary finding of an advantage, and in particular in recital 569 of the contested decision, the Commission stated that, even if the Luxembourg tax authorities were right to accept the assumption that LuxSCS performed unique and valuable functions in relation to the intangible assets, and even if they were subsequently right to find that LuxOpCo performed solely ‘routine’ management functions, the choice of a profit level indicator based on operating expenses in the transfer pricing arrangement endorsed by the tax ruling at issue was inappropriate. It is apparent from recitals 569 to 574 of the contested decision that, according to the Commission, if the Luxembourg tax authorities had taken into account, as a profit level indicator in the application of the TNMM, LuxOpCo’s total costs, as the 2003 transfer pricing report had done, LuxOpCo’s remuneration would have been higher than that found in the tax ruling at issue. Consequently, LuxOpCo’s taxable base would also have been higher.

540    In support of its second subsidiary finding of an advantage, the Commission observed that the method endorsed by the tax ruling at issue used operating expenses as the profit level indicator, whereas the 2003 transfer pricing report, on which the request for the tax ruling at issue was based, used total costs as the profit level indicator. Next, the Commission stated that, during the administrative procedure, Amazon acknowledged that there was an inconsistency between the method endorsed in the tax ruling at issue and the method proposed by the 2003 transfer pricing report. In that regard, Amazon merely asserted that that inconsistency had no impact on the result, since operating costs represented the largest proportion of the total costs of the comparable undertakings examined in the 2003 transfer pricing report (recital 571 of the contested decision). In response to those arguments, the Commission stated, first, that selecting comparable undertakings which had low costs of goods sold, even though those costs represented the largest component of LuxOpCo’s costs, ‘indicated’ an inappropriate choice of comparable undertakings. Secondly, the Commission noted that several of the comparable undertakings selected for the comparables analysis had significant costs of goods sold, raw materials and consumables (recital 572 of the contested decision).

541    The Commission concluded that, since total costs is a broader base than operating expenses, LuxOpCo’s taxable income would have been higher had total costs been used – as the authors of the 2003 transfer pricing report had done – as the profit level indicator (recital 574 of the contested decision). To illustrate this finding, in Table 20 of the contested decision, the Commission compared LuxOpCo’s taxable profit under the tax ruling at issue with LuxOpCo’s profit at [confidential]% of total costs without a ceiling. According to that table, for 2006 to 2013, the former was EUR [confidential] million, while the latter amounted to EUR [confidential] million.

542    First, it should be noted that, at the hearing, the Commission stated that, in the second subsidiary finding, as set out in recitals 569 to 574 of the contested decision, it ‘never said’ that total costs were the most appropriate. By contrast, it merely stated that operating expenses were not an appropriate profit indicator for the purpose of determining LuxOpCo’s remuneration. Moreover, the Commission merely applied the logic displayed by the authors of the 2003 transfer pricing report. In other words, it used total costs to determine the arm’s length remuneration for LuxOpCo solely because the authors of the 2003 transfer pricing report had done so.

543    In that regard, it should be borne in mind that, as stated in paragraph 125 above, the burden of proof is on the Commission to show that the conditions of Article 107(1) TFEU are met. In particular, the Commission must demonstrate the existence of an advantage in favour of the undertaking which it considers to be the recipient of State aid. That advantage must be a genuine advantage.

544    It should be remembered that, in the present case, the issue of whether there is an advantage involves an examination of whether or not the royalty payable by LuxOpCo to LuxSCS, as referred to in the calculation method endorsed by the tax ruling at issue, was at arm’s length. In that regard, the Commission identified errors in the application of the TNMM proposed by Amazon and endorsed in the tax ruling at issue. Specifically, the Commission identified an error in the choice of profit level indicator endorsed by the tax ruling at issue.

545    As has already been pointed out, the mere finding of a methodological error is not, in principle, sufficient in itself to demonstrate that a tax ruling conferred an advantage on a particular company and, therefore, to establish the existence of State aid within the meaning of Article 107 TFEU (see paragraph 123 above). Moreover, as has already been pointed out, the burden of proof is on the Commission to show that the conditions for the existence of State aid have been met (see paragraph 125 above).

546    It follows, in the present case, that the Commission was required to show that the error it identified in the choice of indicator led not only to a different outcome but also to a reduction in the tax burden of the recipient of the tax ruling at issue. That meant answering the question of which profit level indicator was appropriate.

547    Having regard to the interpretation given by the Commission at the hearing to recitals 569 to 574 of the contested decision, that is that it used total costs not on the ground that they were an appropriate profit level indicator, but only for the purpose of transposing the underlying ‘logic’ of the 2003 transfer pricing report (see paragraph 542 above), it should be held that the Commission did not seek to ascertain that it was arm’s length remuneration or, a fortiori, whether LuxOpCo’s remuneration, endorsed by the tax ruling at issue, was lower than the remuneration that LuxOpCo would have received under arm’s length conditions.

548    It follows that, by the second subsidiary finding, the Commission has not succeeded in proving the existence of an advantage.

549    Secondly, and for the sake of completeness, it should be noted that, even if it were appropriate to find that, in recitals 569 to 574 of the contested decision, the Commission did in fact find that total costs were the correct profit level indicator (as opposed to a mere transposition – that serves no useful purpose – of the logic underlying the 2003 transfer pricing report), the considerations supporting the second subsidiary finding of an advantage and the conclusion set out by the Commission in recital 547 of the contested decision cannot succeed.

550    In that regard, it should be recalled and highlighted that, as the Commission stated in recital 569 of the contested decision, it based its second subsidiary finding concerning the advantage – and therefore identified that the choice of operating costs was a methodological error – on the argument that LuxSCS performed unique and valuable functions, whereas LuxOpCo performed solely ‘routine management functions’. The premiss of the Luxembourg tax authorities, as accepted by the Commission in recital 569 of the contested decision, was therefore that LuxOpCo performed only the limited functions of a management company.

551    It is not apparent from the 1995 version of the OECD Guidelines, however, that total costs were the appropriate profit level indicator for remunerating a management company. Since the activity of a management company is similar to the activity of a company that provides services and whose value is unconnected with the volume of actual sales and of purchases of raw material, such an undertaking’s operating expenses, and not total costs, could, according to the 1995 version of the OECD Guidelines, be used to determine the appropriate profit indicator.

552    In the present case, first, when it referred, in recitals 572 and 573 of the contested decision, to total costs – as opposed to operating costs – to calculate LuxOpCo’s remuneration, the Commission in fact departed from the premiss that it itself had set in recital 569 of the contested decision.

553    Contrary to the approach mentioned in recital 569 of the contested decision, according to which LuxOpCo would be regarded as a company ‘with management functions’, the findings in recital 572 of the contested decision are based on the view that LuxOpCo is a ‘retailer’. On closer inspection, the Commission preferred the option of total costs since LuxOpCo was a retailer and not because it was a ‘company with management functions’. In the light of the premiss mentioned in recital 569 of the contested decision, the Commission should have sought to establish the profit level indicator for a management company and not for a company operating as a retailer.

554    Secondly, as is apparent from paragraph 551 above, according to the 1995 version of the OECD Guidelines, in general, as regards management companies, it is not clear that total costs are an appropriate profit level indicator.

555    Admittedly, it cannot automatically be excluded that, in the case of a particular management company, for reasons specific to that company, the appropriate profit level indicator for that company may in fact be total costs. However, the Commission did not explain the reason that might have justified the choice of total costs as an appropriate profit level indicator in the particular case of LuxOpCo as a management company.

556    Thirdly, even if it were accepted that the choice of total costs is the most appropriate profit level indicator for LuxOpCo as a retailer (see recital 572 of the contested decision), it is clear that the Commission did not analyse the issue of the choice of the appropriate profit level indicator for LuxOpCo as administrator of a marketplace for third-party sellers. Moreover, as regards its own sales, the Commission did not examine the extent to which total costs were an appropriate profit level indicator for LuxOpCo’s activity as an online retailer.

557    In that regard, it should be borne in mind that, in its reply to a written question put to it by the Court, the Commission stated that the use of a mark-up on operating costs – and not total costs – as profit level indicator for distribution activities is appropriate where the tested party acts as an intermediary and does not risk its own funds through, inter alia, purchasing goods that are resold. In that regard, the Commission relied on paragraphs 2.101 and 2.102 of the 2010 version of the OECD Guidelines, which are not, however, relevant in the present case.

558    Even if it were accepted that such a recommendation was itself apparent from the 1995 version of the OECD guidelines and was relevant in the present case, as administrator of a marketplace, from the point of view of third-party sellers, LuxOpCo was simply an intermediary between third-party sellers and consumers and did not risk its own funds in relation to sales made by third-party sellers.

559    Lastly, even if it were accepted that the Commission was justified in finding that total costs were the correct indicator for LuxOpCo as an online retailer, given that LuxOpCo used the technology for retail activities, and, in particular, the technology relating to automatic pricing, the choice of total costs as profit level indicator for LuxOpCo would have necessitated downward adjustments to the rate to be applied, in order to take account of material differences between LuxOpCo’s cost structure and that of traditional retailers.

560    The Commission did not envisage, let alone make, any such adjustments.

561    Secondly, according to Table 20 in the contested decision, LuxOpCo’s remuneration, calculated applying the mark-up of [confidential]% of its total costs, amounted to between EUR 2 billion and EUR 3 billion.

562    At the hearing, Amazon stated, without being contradicted on this point by the Commission, that LuxOpCo’s remuneration calculated by the Commission was higher than the ‘total profits’ generated in the European Union, which, according to the Commission, amounted to between EUR 1 billion and EUR 1.5 billion. The remuneration calculated by the Commission under the second subsidiary finding is equivalent to twice the value of all the Amazon group’s profits obtained in Europe and is therefore unrealistic. However, it must be noted that, as is apparent from Annex C1 in Case T‑318/18 to which Amazon referred at the hearing, the figure of between EUR 1 billion and EUR 1.5 billion does not correspond solely to LuxOpCo’s accounting profit over the relevant period, but to the consolidated profit of LuxSCS and LuxOpCo, and therefore a deduction is made in respect of the amounts paid by LuxSCS under the CSA and the Buy-In Agreement.

563    In any event, it is apparent from a comparison between LuxOpCo’s remuneration calculated by applying a mark-up of [confidential]% of the total costs for each year of the relevant period as set out in the last row of Table 20 in the contested decision and LuxOpCo’s resulting operating profit for the same years, as identified by the Commission in the eighth row of Table 2 in the contested decision, that LuxOpCo’s remuneration, calculated under the second subsidiary finding of an advantage, was higher than its operating profit for 2012 and 2013. Such a result clearly departs from an arm’s length outcome.

564    It follows from the foregoing that applying the mark-up of [confidential]% of LuxOpCo’s total costs, on which the second subsidiary finding is based, does not produce reliable results for calculating LuxOpCo’s remuneration for the entire relevant period. It is therefore not an outcome that corresponds to an arm’s length remuneration, so that it must be concluded that the calculation made by the Commission under its second subsidiary finding does not demonstrate that LuxOpCo obtained an advantage as a result of the choice of operating costs as the profit level indicator endorsed by the tax ruling at issue.

565    Consequently, the pleas and arguments of the Grand Duchy of Luxembourg and Amazon challenging the merits of the second subsidiary finding of an advantage should be upheld.

(d)    The third subsidiary finding concerning the advantage

566    In the third complaint in the second part of the first plea in Case T‑816/17 and in the fifth plea in Case T‑318/18, the Grand Duchy of Luxembourg and Amazon contest the Commission’s third subsidiary finding concerning the existence of a tax advantage in favour of LuxOpCo (Section 9.2.2.3 of the contested decision).

567    It will be recalled that, as stated in paragraph 68 above, in the third subsidiary finding, the Commission found, in essence, that the inclusion of a ceiling in the transfer pricing arrangement, under which LuxOpCo’s remuneration could not exceed 0.55% of its annual sales, was inappropriate and conferred an advantage on LuxOpCo in that it led to a reduction in its taxable income.

568    Specifically, the Commission found that, during the 2006, 2007, 2011, 2012 and 2013 tax years, the tax administration accepted tax declarations in which LuxOpCo’s taxable income was determined by the ceiling of 0.55% of its annual sales (recital 575 of the contested decision).

569    The Commission stated that neither the 2003 transfer pricing report, the ex post studies, nor the arguments put forward by the Grand Duchy of Luxembourg and Amazon during the administrative procedure justify the inclusion of that ceiling (recitals 576 and 577 of the contested decision). It added, in recital 577 of the contested decision, that, in addition to the erroneous application of the mark-up to the operating costs, application of the ceiling led to a further reduction of LuxOpCo’s taxable basis, which could not, therefore, lie within the range of arm’s length results.

570    According to the Grand Duchy of Luxembourg and Amazon, the Commission was wrong to find that the inclusion of the ceiling conferred an advantage on LuxOpCo.

571    First, they claim that the purpose of including the ceiling was to force LuxOpCo to operate efficiently and to reduce its costs. Secondly, they claim that, in any event, the application of the ceiling never pushed LuxOpCo’s remuneration outside an arm’s length range, as demonstrated by the 2017 transfer pricing report.

572    The Commission disputes those arguments.

573    It submits, in that regard, that the floor and ceiling mechanism is not provided for in the OECD Guidelines and there is no justification for it from a transfer pricing perspective. Furthermore, it adds that, contrary to what is claimed by the Grand Duchy of Luxembourg and Amazon, the application of such a mechanism is not justified by the objective of ensuring a low but stable remuneration for LuxOpCo, since the very objective of the TNMM is precisely to guarantee that the party to whom it is applied receives a low but stable remuneration.

574    First and foremost, it should be noted that, as the Commission confirmed in its answers to the Court’s written questions, the third subsidiary finding is autonomous and independent of the second subsidiary finding. As is apparent from the last sentence of recital 575 of the contested decision, under the third finding, the Commission proceeds from the premiss that operating expenses may be used as the profit level indicator.

575    As the Commission was correct to state in the contested decision and in its pleadings, there is no economic justification or rationale for the floor and ceiling mechanism. It is hardly conceivable that, under market conditions, an undertaking would accept a cap on its remuneration at a percentage of its annual sales. In addition, the TNMM ensures a low but stable remuneration, without the need for a floor and ceiling mechanism. Nor is such a mechanism provided for in the 1995 version of the OECD Guidelines. The TNMM merely involves identifying a profit indicator and a mark-up.

576    The Commission was therefore correct to find that the inclusion of such a ceiling was a methodological error.

577    Nevertheless, that finding alone is not sufficient to establish the existence of an advantage.

578    It is clear that, for each year the tax ruling at issue applied, even after applying the ceiling mechanism, LuxOpCo’s remuneration remained within an arm’s length range, calculated on the basis of the 2003 transfer pricing report, that is between [confidential]% and [confidential]% of operating expenses. Moreover, the Commission does not contest that finding.

579    As the Commission itself explained at the hearing, since the remuneration is within the interquartile range, it must, in principle, be regarded as being at arm’s length.

580    Admittedly, the Commission stated that such a conclusion could not be reached where the comparable undertakings on the basis of which that range was calculated had not been properly selected.

581    Nevertheless, in the third subsidiary finding, the Commission challenged neither the arm’s length range nor the choice of comparable undertakings on the basis of which that range was calculated.

582    In recital 575 of the contested decision, the Commission took issue with the Luxembourg authorities for having accepted that LuxOpCo’s taxable income was determined by applying the ceiling, ‘instead of being determined as [confidential]% of its operating expenses’. It is therefore clear that, in the third subsidiary finding, the Commission does not call in question the rate of return on the basis of which it is applied, but only the ceiling itself.

583    Moreover, first, it is not apparent from recitals 575 to 578 of the contested decision that the Commission disputed the arm’s length range, calculated in the 2003 transfer pricing report, which is between [confidential]% and [confidential]% of operating expenses. Although the Commission noted, in recital 577 of the contested decision, that the Grand Duchy of Luxembourg and Amazon argued, during the administrative procedure, that LuxOpCo’s taxable income was never outside the arm’s length range, the Commission did not dispute the range itself, but merely asserted that application of the ceiling led to a further reduction to that identified in the second subsidiary finding. As stated in paragraph 574 above, the second and third subsidiary findings are autonomous and independent.

584    Secondly, and in addition, the Commission did not call in question, in recitals 575 to 578 of the contested decision, the choice of comparable undertakings used in calculating the arm’s length range. Recital 571 of the contested decision, in which the Commission referred to an error in the choice of comparable undertakings, is part of the second subsidiary finding. As stated in paragraph 574 above, the third subsidiary finding was autonomous and independent of the other findings.

585    In the light of the foregoing, it should be held that, as inappropriate as the ceiling mechanism may be, and although it is not provided for in the 1995 version of the OECD Guidelines, the Commission has not demonstrated that that mechanism had an impact on the arm’s length nature of the royalty paid by LuxOpCo to LuxSCS.

586    Consequently, the sole finding that the ceiling was applied for 2006, 2007, 2011, 2012 and 2013 is not sufficient to establish that LuxOpCo’s remuneration received in respect of those years did not correspond to an approximation of an arm’s length outcome.

587    In fact, the Commission found, at most, a methodological error in the calculation of LuxOpCo’s remuneration, without succeeding in showing that that error had the effect of artificially reducing LuxOpCo’s remuneration to such an extent that that level of remuneration could not have occurred under market conditions.

588    In those circumstances, it should be held that, by the third subsidiary finding, the Commission has not demonstrated the existence of an advantage for LuxOpCo.

589    Accordingly, the pleas and arguments of the Grand Duchy of Luxembourg and Amazon challenging the third subsidiary finding of an advantage should be upheld.

590    It therefore follows from all the foregoing that the Commission has not succeeded, by any of the findings set out in the contested decision, in demonstrating to the requisite legal standard the existence of an advantage within the meaning of Article 107(1) TFEU. Accordingly, the contested decision should be annulled in its entirety, without the need to examine the other pleas and arguments raised by the Grand Duchy of Luxembourg and Amazon.

 Costs

591    Under Article 134(1) 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. Since the Commission has been unsuccessful, it must be ordered to bear its own costs and to pay those of the Grand Duchy of Luxembourg and Amazon, in accordance with the forms of order sought by the applicants.

592    In accordance with Article 138(1) of the Rules of Procedure, Ireland shall bear its own costs.

On those grounds,

THE GENERAL COURT (Seventh Chamber, Extended Composition)

hereby:

1.      Joins Cases T816/17 and T318/18 for the purposes of the present judgment;

2.      Annuls Commission Decision (EU) 2018/859 of 4 October 2017 on State aid SA.38944 (2014/C) (ex 2014/NN) implemented by Luxembourg in favour of Amazon;

3.      Orders the European Commission to bear its own costs and to pay those incurred by the Grand Duchy of Luxembourg, Amazon.com, Inc. and Amazon EU Sàrl;

4.      Orders Ireland to bear its own costs.

Van der Woude

Tomljenović

Marcoulli

Delivered in open court in Luxembourg on 12 May 2021.

E. Coulon

 

M. van der Woude

Registrar

 

      President


Table of contents


I. Background to the dispute

A. The tax ruling at issue

B. The administrative procedure before the Commission

C. The contested decision

1. Presentation of the factual and legal context

(a) Presentation of the Amazon group

(b) Presentation of the tax ruling at issue

(c) Description of the relevant national legal framework

(d) Presentation of the OECD transfer pricing framework

2. Assessment of the tax ruling at issue

(a) Analysis of the existence of an advantage

(1) The primary finding of an advantage

(2) The subsidiary findings of an advantage

(b) Selectivity of the measure

(c) Identification of the beneficiary of the aid

II. Procedure and forms of order sought

A. The procedure in Case T816/17

1. The composition of the formation of the Court and priority treatment

2. The intervention

3. Applications for confidential treatment

4. Forms of order sought

B. The procedure in Case T318/18

1. The composition of the formation of the Court and priority treatment

2. Applications for confidential treatment

3. Forms of order sought

C. Joinder of the cases and the oral part of the procedure

III. Law

A. Joinder of Cases T816/17 and T318/18 for the purposes of the decision closing the proceedings

B. Pleas in law and arguments relied on

1. Preliminary remarks

(a) Determination of the conditions for the application of Article 107(1) TFEU in the context of national tax measures

(b) The burden of proof

(c) The intensity of the review to be conducted by the Court

2. Pleas and arguments challenging the primary finding of an advantage

(a) Admissibility of certain arguments put forward by Ireland concerning the existence of an advantage

(b) The merits of the pleas in law and arguments of the Grand Duchy of Luxembourg and Amazon concerning the primary finding of an advantage

(1) Temporal relevance of certain OECD Guidelines used by the Commission for the purposes of applying the TNMM

(2) Alleged errors committed by the Commission in the application of the TNMM in the contested decision

(i) The functional analysis and the Commission’s choice of LuxSCS as the tested party

– The functions and assets of LuxSCS

– The risks assumed by LuxSCS

– Conclusion on the functional analysis of LuxSCS and the impact of that conclusion on the choice of that company as the tested party

(ii) The remuneration calculated by the Commission for LuxSCS on the basis that it was the tested party

– The first component of the royalty due to LuxSCS (Buy-In costs and CSA costs)

– The second component of LuxSCS’s remuneration (maintenance costs)

(3) Conclusion on the primary finding

3. Pleas and arguments challenging the subsidiary line of reasoning concerning the advantage

(a) Preliminary observations on the three subsidiary findings

(b) The first subsidiary finding concerning the advantage

(1) LuxOpCo’s exercise of ‘unique and valuable’ functions

(i) The functions performed by LuxOpCo in relation to the intangible assets (Section 9.2.1.2.1 and recitals 449 to 472 of the contested decision)

– The nature of the licence granted to LuxOpCo (recital 450 of the contested decision)

– The EU IP Steering Committee (recitals 452 to 455 of the contested decision)

– LuxOpCo’s functions concerning the development of the technology

– Customer data (recitals 466 to 468 of the contested decision)

– The ‘Amazon brand’ (recitals 469 to 472 of the contested decision)

(ii) The functions performed by LuxOpCo in the operation of the Amazon group’s European retail and service business (Section 9.2.1.2.2 and recitals 473 to 499 of the contested decision)

– Selection (recitals 479 to 489 of the contested decision)

– Price (recitals 490 to 493 of the contested decision)

– ‘Convenience’ (recitals 494 to 499 of the contested decision)

(iii) Assets used by LuxOpCo (recitals 500 to 505 of the contested decision)

(iv) Risks assumed by LuxOpCo (recitals 506 to 517 of the contested decision)

(v) Conclusions on the functional analysis of LuxOpCo

(2) Choice of method

(3) The demonstration that LuxOpCo’s remuneration would have been higher had the profit split method been used (application of the contribution analysis)

(c) The second subsidiary finding concerning the advantage

(d) The third subsidiary finding concerning the advantage

Costs


*      Languages of the case: English and French.


1      Confidential data omitted.