Brussels Fined Google €890 Million Today. The AI Part Nobody Is Reading.

Three weeks ago, when the EU's highest court made Google's €4.125 billion Android fine final, gafam.ai argued that the ruling mattered less as a verdict on 2018 than as a template for 2026 — because the default-entrenchment mechanism the court condemned for Search is the same mechanism now distributing artificial intelligence. Today Brussels validated that reading twice over. Once with a fine, and once with something more important than the fine.

What the Commission Decided Today

The European Commission today took two decisions finding non-compliance by Google with the Digital Markets Act: one for self-preferencing its own services on Google Search, and one for restrictions preventing businesses from directing consumers to alternative, often cheaper, purchase channels on Google Play.

The Commission issued fines of €460 million and €430 million respectively. Together: €890 million.

On the Search side, the Commission found that Google gives preferential treatment to its own services — including shopping, hotels, transport and sports results — over those of third parties, in breach of its DMA obligations, which require gatekeepers to apply transparent, fair and non-discriminatory ranking conditions.

Alongside the fines, the Commission ordered Google to bring the non-compliance to an end. The Commission said the decisions followed a thorough investigation including market feedback and extensive dialogue with Google, that the fine levels were set by assessing gravity, duration and recurrence, and that Google may appeal. Reporting ahead of the decision indicated the enforcement would carry the threat of daily penalty payments if Google fails to comply within a 60-day window.

Why €890 Million Is Deliberately Small

This is the largest DMA penalty issued so far — Apple received €500 million last year and Meta €200 million, the previous benchmarks under the law. And yet, measured against the instrument Brussels was holding, it is strikingly restrained.

The DMA's theoretical ceiling for a first violation is 10% of a gatekeeper's global annual turnover — on Alphabet's most recent reported revenues, above $40 billion. A penalty calibrated in the hundreds of millions represents a deliberate choice to prioritise behavioural compliance over maximum punishment.

Commission spokesperson Thomas Regnier had put the logic plainly in May: Brussels is more interested in finding future compliance solutions with Google than in issuing a fine.

That framing invites criticism from both directions, and both deserve airing. Google's consistent position across EU cases has been that the Commission misreads competitive reality — after the July 2 Android ruling, a Google spokesperson said the judgment failed to recognise the company's investment in keeping Android open, interoperable and free. The company has not yet issued a statement on today's decisions in the reporting available at publication. From the opposite side, more than 30 civil society organisations led by Open Markets Institute Europe wrote to Commission President von der Leyen in May expressing grave concern that this very fine, originally scheduled for March 2026, had reportedly been delayed — a charge the Commission has not accepted. A penalty that arrives late and lands at roughly two percent of the legal maximum will not settle that argument.

The Document That Matters More

Here is the part almost no coverage today will foreground, and it is where gafam.ai's lens earns its keep. The fine punishes what Google did to Search.

A decision adopted seven days earlier governs what Google may do with AI.

On 16 July 2026, the Commission adopted two DMA specification decisions telling Google exactly how to comply with two obligations: opening eleven defined Android features to rival AI assistants by 1 August 2027 at the latest, and sharing anonymised Google Search ranking, query, click and view data with competing search engines and AI chatbots on FRAND terms from January 2027.

The Commission had opened those specification proceedings on 27 January 2026 under Article 8(2) DMA, and adopted the final decisions ahead of the statutory deadline of 27 July.

Read the two tracks together and the strategy becomes legible. The €890 million is retrospective — it prices conduct in Search and Play that has already happened. The specification decisions are prospective — they define, in advance and in technical detail, how the same platform power may be used in the AI era. The non-compliance investigation predates Gemini; it was designed around a concrete, legible form of self-preferencing. The specification decisions are what Brussels built once it understood that the next round would be about assistants, not shopping boxes.

This is the lesson the Android judgment taught, applied. In 2018 the Commission acted, and by the time the courts finished in 2026 the abuse had migrated to a new layer. This time, the remedy is being written before the defaults harden.

What It Actually Changes for European AI

Two provisions deserve close attention from anyone building or buying AI in Europe.

The Android access obligation attacks the distribution chokepoint directly. If eleven defined system features must be available to rival AI assistants on equal terms, then Gemini's position as the assumed default intelligent layer on Android becomes contestable by law rather than by market power. That is the closest thing to the AI choice-screen remedy this publication has argued for since the Android judgment.

The search data provision is the more radical one, and it is aimed squarely at the moat beneath the moat. Ranking, query, click and view data is the accumulated behavioural record that makes Google's results good — and it is precisely what a challenger, including a European one, cannot build from scratch. Requiring that it be shared on fair, reasonable and non-discriminatory terms with competing search engines and AI chatbots from January 2027 is a structural intervention, not a fine. It does not ask Google to behave better. It redistributes the input.

Whether it works is genuinely open. FRAND terms are notoriously litigable, the timelines run into 2027, and Google has consistently argued that compelled data sharing would hand competitors its proprietary work product. But if European AI providers are ever to reach European users at scale on the devices Europeans actually hold, distribution access and data access are the two locks. Brussels just addressed both.

The Transatlantic Frame

None of this happens in a vacuum, and gafam.ai has spent this month documenting why. Enforcement against American AI companies is no longer a technical matter of competition law; it now runs directly into an American administration that treats these firms as strategic assets — an alignment we examined when OpenAI proposed handing Washington a 5% equity stake. Previous EU penalties against Google have drawn direct presidential objection. A €890 million fine plus binding orders to open Android and share search data will not be read in Washington as routine market regulation.

That is the environment in which the EU AI Act's transparency and GPAI enforcement powers activated on 2 August-eve. Europe is enforcing more, in more areas, against companies with more political protection than at any previous point.

The European Perspective