The Amazon Luxembourg State Aid Decision

In 2003, Amazon requested and obtained an advance tax ruling (ATR) from the Luxembourg tax authorities. The ruling confirmed how LuxOpCo's taxable profit would be calculated: the royalty to LuxSCS would be set equal to LuxOpCo's actual operating profit minus a benchmarked routine return, defined as approximately [4-6]% of LuxOpCo's total operating costs (OpEx), subject to a floor of 0.45% and a ceiling of 0.55% of European revenues.

In practice, as LuxOpCo grew rapidly, so did the royalty. The formula ensured that the royalty absorbed virtually all the upside from Amazon's European expansion, leaving LuxOpCo with a thin, fixed-percentage margin. The ruling was renewed in 2011 on substantially similar terms. Amazon itself withdrew the ATR in 2014 after the Commission opened its formal investigation.

The arrangement produced an effective non-taxation outcome at group level. LuxSCS was transparent for Luxembourg tax purposes: it was not taxed in Luxembourg. Under a US "check-the-box" election, however, LuxSCS's US partners treated it as a corporation (not a transparent partnership) for US tax purposes, meaning its Luxembourg income was classified as controlled-foreign-corporation income for the US entities. Under the US rules then in force, that income was not taxed in the United States until it was repatriated.

The practical result: LuxOpCo paid Luxembourg corporate tax only on its thin margin; the royalty income at LuxSCS was neither taxed in Luxembourg (transparent entity) nor currently taxed in the United States (CFC deferral). This hybrid mismatch allowed Amazon to accumulate large untaxed reserves in Luxembourg.

Under the Transactional Net Margin Method (TNMM), the tested party should be the entity performing routine, non-unique functions for which reliable comparable data exist from independent companies. The ATR treated LuxOpCo as the routine entity and LuxSCS as the "principal" entitled to the residual profit.

The Commission found this reversal had no economic basis. A functional analysis showed that LuxOpCo performed all the value-creating functions: technology development and enhancement, marketing, logistics, customer relations. LuxSCS had no employees, performed no active functions, and did not genuinely manage or bear the risks allocated to it by contract. Under a correct analysis, LuxSCS should have been benchmarked at a low routine return (it performed no complex functions); the residual profit would then have remained with LuxOpCo, which would have been taxed on it in Luxembourg.

The Commission concluded this error alone was sufficient to establish the selective advantage. The subsequent errors were identified as supporting findings.

Even if the TNMM applied to LuxOpCo as the tested party had been the correct methodology, the choice of operating costs (OpEx) as the profit-level indicator was flawed. LuxOpCo's OpEx did not include the royalty itself. This created a circularity: as the royalty rose, the OpEx figure used to calculate the benchmarked routine return fell, which in turn allowed the royalty formula to produce an even higher royalty in the following year. The PLI thus self-reinforced the advantage rather than providing an independent check on it.

The ATR contained a downward collar: in any year, the royalty could not fall below 0.45% of LuxOpCo's revenues, regardless of what the TNMM calculation produced. In years of high revenue growth, this floor meant the royalty could be set at a level that left LuxOpCo with less than even its benchmarked routine return, further compressing taxable income beyond what even the flawed TNMM would have produced.

The Commission did not set a fixed recovery amount in the decision itself. Instead, it set out a methodology for Luxembourg to apply for each tax year from 2006 to 2014:

• Step 1: Apply the correct TNMM by treating LuxSCS as the tested party and benchmarking it against comparable companies performing simple, low-risk holding or intermediary functions. This produces a low routine return for LuxSCS.
• Step 2: Deduct that routine return from LuxOpCo's actual profit; the remainder is the arm's-length profit attributable to LuxOpCo.
• Step 3: Calculate the royalty that would have given LuxOpCo that arm's-length profit. This is the arm's-length royalty.
• Step 4: The difference between the royalty actually paid (higher) and the arm's-length royalty (lower) is the excess royalty that incorrectly reduced LuxOpCo's taxable base.
• Step 5: The tax advantage for each year equals the excess royalty multiplied by the applicable Luxembourg corporate income tax rate.
• Step 6: Sum all annual tax advantages over the 2006-2014 period. The Commission indicated the total was in the order of EUR 250 million.

Luxembourg was required to complete this calculation, present it to the Commission for approval, and effect actual recovery within four months of notification of the decision.

Amazon EU S.à r.l. and Amazon.com Inc. brought the first challenge (T-816/17) in December 2017, immediately after notification of the Commission decision. Luxembourg filed its own action (T-318/18) in March 2018. The two cases were joined and heard together. The General Court delivered its judgment on 12 May 2021, annulling the Commission decision in full.

The Court's principal findings:

• The Commission bore the burden of proving, to a sufficient legal standard, that the ATR conferred a selective advantage. It had to show not merely that the methodology was different from the correct methodology but that the correct methodology would have produced a materially higher tax result.
• The Commission's own alternative calculation (applying the TNMM to LuxSCS as the tested party) involved methodological choices for which it had not provided sufficient justification. The Court found the Commission could not simply assert that a different benchmark for LuxSCS would have produced a substantially higher royalty for LuxOpCo without demonstrating this with reliable comparables.
• The three errors identified were not individually or collectively shown to produce a tax reduction relative to what would have been due under Luxembourg law applying a correct arm's-length analysis.
• The Court did not find that the structure was acceptable; it found only that the Commission had not met the evidentiary burden it bore in the particular procedural context of an individual State aid investigation.

The Commission appealed the General Court's annulment to the Court of Justice of the EU (CJEU), arguing that the lower court had applied an incorrect evidentiary standard and had failed to give proper weight to the Commission's functional analysis of LuxSCS. The appeal was heard by a chamber of five judges. The CJEU delivered its judgment on 14 December 2023, dismissing the Commission's appeal in its entirety.

The CJEU confirmed that:

• The General Court had correctly identified that the Commission must establish a selective advantage by reference to the rules of the relevant national tax system, not by reference to an idealised arm's-length standard.
• The Commission's analysis of LuxSCS's functions was insufficiently supported to satisfy the required standard of proof.
• The Commission had failed to demonstrate that the specific TNMM variant it considered correct would have produced a materially lower royalty and thus a higher taxable profit for LuxOpCo.

With the CJEU's judgment, the case was definitively closed. No recovery was ordered, and Amazon retained the tax advantage it had enjoyed between 2006 and 2014.

The Commission found it was illegal State aid and ordered recovery in 2017. The General Court and the Court of Justice both disagreed, finding that the Commission had not proved its case to the required standard. The ruling was therefore not definitively found to be illegal; the Commission's decision was annulled. Whether the arrangement was problematic from a tax policy perspective is a separate question from whether it met the specific legal test for State aid.

Amazon paid Luxembourg corporate income tax on LuxOpCo's benchmarked routine margin. The royalty income that flowed to LuxSCS was not taxed in Luxembourg (transparent partnership) and was deferred in the United States under controlled foreign corporation rules. The effective tax rate on the overall profit routed through the LuxSCS/LuxOpCo structure was substantially lower than the Luxembourg statutory rate, though the exact figure was not published. After the 2014 withdrawal of the ATR, Amazon restructured its European tax arrangements and began booking revenues through national subsidiaries in each Member State.

The arm's-length principle requires that transactions between related companies (within the same corporate group) be priced as if the companies were independent and acting in their own commercial interests. If an intra-group royalty, loan, or service fee departs from what unrelated parties would have agreed, the taxing authorities can disregard the agreed price and substitute an arm's-length price for tax purposes. The principle is enshrined in Article 9 of the OECD Model Tax Convention and in the domestic law of most countries, including Luxembourg's Income Tax Act.

The courts found that the Commission had not assembled sufficient evidence to demonstrate that LuxOpCo's tax burden was lower than it would have been under a correct arm's-length analysis applying Luxembourg law. The Commission identified methodological errors in the ATR but did not, in the courts' view, prove with adequate rigour that correcting those errors would have produced a materially higher tax result. The burden of proof in State aid cases lies with the Commission, and the courts held it had not discharged that burden.

The Apple case (SA.38373, Commission Decision 2017/1283) concerned advance opinions by the Irish Revenue Commissioners that the Commission said allowed Apple to attribute most of its European profits to stateless "head offices" of two Irish companies, resulting in a very low effective tax rate. The Commission ordered Ireland to recover approximately EUR 13 billion. Like Amazon, Apple and Ireland challenged; the General Court annulled in 2020, but in a striking reversal the Court of Justice overturned the General Court's annulment in September 2024 and upheld the Commission's original finding. Apple was therefore required to pay back taxes. The two cases illustrate how outcomes at the CJEU can vary significantly depending on the specific evidentiary record and legal arguments in each case.

Largely no. Several legal changes since 2014 make the original Amazon structure either impossible or significantly more expensive from a tax perspective. ATAD 2 neutralises hybrid mismatches of the LuxSCS type. BEPS Actions 8-10, now integrated into the OECD Transfer Pricing Guidelines, require profit attribution to align with value creation, not contractual arrangements. Pillar Two (global minimum tax, effective from 2024 for large groups) imposes a top-up tax to ensure a minimum 15% effective rate in every jurisdiction. DAC6 requires disclosure of arrangements with features similar to the Amazon structure. And CbCR means that high-profit, low-employee jurisdictions in any large group's structure are routinely flagged for audit.