Tax Rulings as State Aid: Apple, Fiat, Amazon and What Survived

The tax-rulings state aid cases reshaped international tax planning more than any legislative reform of the same decade. They established that a binding ruling from a national tax authority offers no protection against EU recovery, that transfer-pricing methodology is reviewable as state aid, and — after the annulments — that the reference system must be found in national law rather than constructed by the Commission. This guide walks the saga and its practical consequences, as part of the state-aid pillar of our Competition & Antitrust hub.

How did tax rulings become a state aid problem?

Through selectivity. A tax ruling is simply advance confirmation of how the law applies to a taxpayer’s facts — unobjectionable in itself, and used by every developed tax system to provide certainty. The Commission’s insight was that a ruling endorsing an allocation of profit no independent party would accept effectively grants that taxpayer a lower tax burden than the general system imposes, which is an advantage, financed by the state, available only to that undertaking.

The context made the intervention politically irresistible: LuxLeaks and the wider disclosure of intra-group financing and IP structures had made effective rates in the low single digits publicly visible, and no tax-harmonisation route was available given unanimity requirements. State aid law offered the only unilateral EU instrument capable of reaching national tax decisions — and, uniquely, one whose remedy (recovery) delivered the money to the very member states that had granted the rulings, several of which resisted it. That awkward politics runs through the whole saga, including Ireland’s decade-long opposition to receiving €14 billion.

What did each major case decide?

Fiat (Luxembourg) and Starbucks (Netherlands), both 2015 decisions of roughly €20–30 million each, tested intra-group financing and coffee-roasting royalties. The General Court upheld Fiat and annulled Starbucks in 2019; the Court of Justice then annulled Fiat in November 2022 on the reference-system point, which fatally undermined the methodology behind several pending cases. Amazon (Luxembourg), a €250 million decision on royalty payments to a limited-partnership IP holder, was annulled by the General Court in 2021 and the annulment upheld by the Court of Justice in December 2023 on the same reasoning.

Apple (Ireland) was the outlier in scale and outcome: the Commission’s 2016 decision found that Ireland’s 1991 and 2007 rulings allowed Apple Sales International and Apple Operations Europe to attribute nearly all profits to “head offices” existing only on paper, leaving effective rates on European profits of a fraction of one percent. The General Court annulled in 2020 for failure to prove the advantage; the Court of Justice set that judgment aside in September 2024 and gave final judgment restoring the recovery — because, it held, the branch-attribution analysis was grounded in Irish law’s own rules, not an invented standard. Engie (Luxembourg) followed a different route (abuse-of-law and consistency of the national system) and was likewise annulled in 2023, while cases against Nike and McDonald’s ended without recovery.

What is the doctrine now — and what remains at risk?

Narrower but real. The Commission can still attack a ruling where it can demonstrate, on the member state’s own legal rules, that the taxpayer was taxed more favourably than the system requires — misapplied branch-attribution rules, rulings inconsistent with the state’s own transfer-pricing legislation, arrangements departing from published national practice. What it can no longer do is substitute its preferred allocation methodology for the national one and call the difference an advantage.

Residual exposure therefore concentrates on: rulings that are inconsistent with the granting state’s own law or practice; schemes rather than individual rulings, where selectivity is easier to demonstrate against a general system (the UK’s controlled-foreign-company group financing exemption case, decided against the UK, shows this route working); and new-generation measures — special regimes, patent boxes, and incentives designed for particular investors. Meanwhile the centre of gravity has shifted to legislation: the OECD Pillar Two minimum tax and EU implementing directives address the same economics prospectively, which reduces the appetite for retrospective aid theories without eliminating them.

What should multinationals do about ruling risk?

Treat rulings as evidence, not immunity. Every advance pricing agreement and tax ruling should be supported by an economic analysis capable of standing on its own under the granting state’s law — because the ruling’s protective value evaporates precisely when it is most needed. That means contemporaneous transfer-pricing documentation, functional analyses that match operational reality (the Apple decision turned on head offices with no staff), and periodic re-testing as the business changes.

Structurally: avoid arrangements whose tax outcome depends on entities without substance; align legal structures with where value is genuinely created, which is also where Pillar Two and the anti-avoidance directives now push; and model recovery exposure in the same way as any contingent liability, with a ten-year horizon. In transactions, tax-ruling exposure belongs in diligence alongside other inherited competition risks, since recovery obligations attach to the beneficiary undertaking and can follow a business through a sale.

What did the cases change in practice, beyond the money?

Three things. Member states’ ruling practice tightened materially: several jurisdictions narrowed the scope of advance agreements, published more, and became noticeably more conservative on structures involving low-substance entities. Multinationals restructured — the specific arrangements at issue in Ireland, Luxembourg and the Netherlands were largely unwound during the litigation decade, often before any judgment. And transparency rules followed: mandatory exchange of cross-border rulings between EU tax authorities became standard, removing the confidentiality that made the arrangements viable.

The deeper change is behavioural. Tax structures are now designed with an EU state aid overlay in mind alongside domestic law and treaty analysis, and boards ask a question they did not ask in 2013: if this arrangement became public and was tested against our own jurisdiction’s law by a hostile regulator a decade from now, does it hold? That is a durable governance shift, and it survived the annulments intact.

Could the doctrine be revived at scheme level?

It already operates there. Where the Commission attacks a general regime rather than an individual ruling, the reference-system problem is easier: the scheme itself is the derogation, and comparability is assessed within the national system’s own logic. The UK’s group financing exemption within its controlled-foreign-company rules was condemned on exactly this basis, and patent boxes, tonnage-tax variants and sector-specific exemptions have all been examined.

Expect future activity to concentrate there and in new industrial-policy measures — green-transition incentives, semiconductor support, targeted investment regimes — where selectivity is often explicit by design. The compatibility route (Article 107(3)) rather than the existence of aid becomes the battleground, which is a different and more political conversation than the rulings decade’s technical one.

What does the Apple judgment mean for other member states?

That branch- and permanent-establishment attribution done inconsistently with national law remains genuinely exposed. The Court’s reasoning was not Apple-specific: where a state’s own rules require profits to be attributed to activities actually performed in its territory, a ruling attributing them elsewhere departs from the reference system, and the resulting tax saving is aid. Jurisdictions with substantial non-resident branch structures should read the judgment closely.

For taxpayers, the practical test is substance-based and answerable in advance: do the entities to which profit is attributed actually perform the functions, employ the people and bear the risks that justify it under local law? Where the answer is no, the arrangement carries state aid risk irrespective of any ruling — and, increasingly, Pillar Two exposure as well.

How should boards think about the residual exposure?

As a ten-year contingent liability attached to structures, not to rulings. The governance questions are simple to ask and uncomfortable to answer: which of our arrangements produce an effective rate materially below the statutory rate in the jurisdiction concerned; does that outcome follow from that jurisdiction’s own law applied to our actual functions and people; and could we demonstrate it, with contemporaneous evidence, to a hostile reviewer in 2035? Where any answer is uncertain, the exposure belongs in the risk register with a quantified worst case including compound interest.

Frequently Asked Questions