EU Orders Google to Open AI and Search Data

Google's AI Flywheel Just Hit a Wall in Brussels — and Its US Rivals Stand to Gain

The European Commission's July 16 specification decisions force Google to open Android's AI plumbing and its search-data vault to competitors. The biggest beneficiaries will not be European startups.

On July 16, 2026, the European Commission adopted two binding decisions that do something no regulator has attempted before: they order Google to dismantle the self-reinforcing loop between its search monopoly and its AI ambitions before that loop becomes unbreakable.

The first decision forces Alphabet to give rival AI assistants, from OpenAI's ChatGPT to Anthropic's Claude, the same access to Android operating system features that Google's own Gemini enjoys. The second compels the company to share its search data (the ranking, query, click, and view data generated by billions of European users) with competing search engines, including AI chatbots that offer search functions. Both are legally binding. Failure means fines of up to 10% of global annual turnover, which for a company that reported $307 billion in revenue in 2024 would approach $31 billion per violation.

The measures land barely two weeks after the European Court of Justice dismissed Google's final appeal of its €4.1 billion Android antitrust fine, according to BBC News, bringing the company's cumulative EU antitrust bill to over €10 billion. But the specification decisions are more consequential than any fine. They target the architecture of Google's next monopoly, not the last one.

The Flywheel Brussels Is Trying to Break

The Commission's logic, laid out in the official specification documents published on the DMA portal, runs on a simple diagnosis: Google's AI services are not winning on merit. They are winning because they inherit an ecosystem nobody else can replicate.

Android powers 60% of EU smartphones. Google's Gemini assistant has unrestricted access to 11 specific Android features: voice invocation via "Hey Google," screen-context awareness through Circle to Search, the ability to read and write data across Google apps like Gmail, Calendar, Keep, and Drive, and the capacity to execute actions inside third-party apps. Rival AI assistants get none of this. A user cannot say "Hey ChatGPT" and have the assistant book a taxi through an app or search their calendar for an upcoming trip. The result, the Commission found, is that "third-party AI assistants are not competing on an equal footing" and are "less attractive" to European users.

On the search side, the problem is even more structural. Google's search engine processes hundreds of billions of queries annually across every access point — Chrome's omnibox, the Google Search widget, Google Assistant, Gemini, Google Lens, Circle to Search, and Android's text search. This generates an unmatched stream of query, click, and ranking data that Google uses to continuously optimize its own search services and, critically, its AI products. As the DOJ argued in its parallel US antitrust case, detailed by NPR, Google's search index — over 100,000,000 gigabytes and hundreds of billions of webpages — gives Gemini a training and grounding advantage no competitor can match through organic growth alone.

The Commission's July 16 decisions are an attempt to break this flywheel at two points simultaneously.

What the AI Interoperability Decision Actually Requires

The specification decision under Article 6(7) of the DMA — case DMA.100220 — covers 11 Android features and imposes a concrete implementation timetable.

By August 1, 2027, when Android 18 ships, Google must provide rival AI services with equally effective access to: on-screen context (the ability to read what a user is viewing), voice invocation via hardware buttons or navigation handles, centralized read/write access to user data across Google apps, the ability to execute tasks in third-party apps, and access to Google's own services, including Gmail, Maps, Calendar, YouTube, Drive, and the Phone app, through operating-system-level integration channels equivalent to what Gemini uses.

By August 1, 2028, with Android 19, Google must enable concurrent hotword detection — meaning a user can say "Hey ChatGPT" or "Hey Claude" alongside "Hey Google," and the appropriate assistant will wake.

Critically, all of this must be provided free of charge, across the entire Android ecosystem including devices from Samsung, Xiaomi, and other manufacturers. Google must publish complete documentation, allow third parties to test their implementations, and provide technical assistance. An independent certification process — with Google publishing draft eligibility terms by February 1, 2027 and final terms by May 1, 2027 — will determine which AI services qualify.

The Commission explicitly states, in the specification proceedings page, that the decision "does not specify the exact technical implementation." Google retains flexibility on how to open Android. It does not retain flexibility on whether.

The Search Data Decision: AI Chatbots Are the Real Target

The second decision, under Article 6(11), is where the AI implications sharpen into focus.

The DMA already required Google to share anonymized ranking, query, click, and view data with rival search engines on fair, reasonable, and non-discriminatory (FRAND) terms. But the Commission found that Google's existing offer was "ineffective." The July 16 decision specifies, with granular detail, what must change.

AI chatbots with search functionality are explicitly eligible to receive the shared data. This is the load-bearing provision. It means OpenAI (ChatGPT Search), Anthropic (Claude with search), Perplexity, Microsoft (Copilot/Bing), and any future rival can access the same query-and-click stream that Google uses to train and ground its own AI search products. They cannot use the data to train general-purpose AI models or for advertising (the purpose limitation is strict), but they can use it to optimize retrieval, ranking, indexing, query analysis, spell-checking, auto-suggestion, and auto-completion for their search services.

The data scope is comprehensive: queries from every Google Search access point, query metadata (language, device type), URLs viewed, user interaction data with search results, and information on ranking positions. Google must share this data via an API with latency no worse than what it uses internally, for up to five years per beneficiary.

The pricing formula must be FRAND. The anonymization requirements, developed in consultation with the European Data Protection Board, mandate a multi-layered method that the Commission argues preserves data utility while protecting privacy — though this remains the most contested part of the decision. DuckDuckGo has previously criticized Google's proposed anonymization as producing "a data set that is unusable for search training purposes," according to Brookings.

The Transatlantic Context: Parallel Actions, Different Logics

The EU's moves do not happen in a vacuum. US District Judge Amit Mehta, in his remedies ruling in the DOJ's search antitrust case, also ordered Google to share certain search index and user interaction data — though not advertising data — with competitors, as reported by BBC News. The UK's Competition and Markets Authority has separately designated Google with "strategic market status" and ordered transparency on search rankings, according to Al Jazeera. And the BBC reported that the EU has opened a parallel investigation into whether Google's AI Overviews and AI Mode use publisher content without "appropriate compensation."

But the July 16 specification decisions differ from all of these in one decisive respect: they are ex ante and self-executing. Under the DMA, the burden is on Google to demonstrate compliance, not on the regulator to prove violation. The Commission does not need to wait for harm to materialize.

This regulatory architecture is precisely what has drawn fire from Washington. As CSIS documented, the Trump administration imposed visa restrictions on five EU officials involved in drafting the DMA and DSA, and has embedded anti-digital-regulation clauses into bilateral trade agreements with Malaysia, Indonesia, Cambodia, Argentina, Guatemala, and El Salvador. The administration's strategy, CSIS argues, is "containment" — unable to roll back EU rules at their source, it aims to prevent their spread.

Yet Brussels has shown no sign of retreat. Commission President Ursula von der Leyen has called the DMA and DSA "untouchable" in trade negotiations, Brookings reports. The specification decisions adopted on July 16 are the clearest demonstration yet that the Commission intends to use every tool the DMA provides — not just fines, but structural remedies that reshape market architecture.

Who Wins, Who Loses

The immediate beneficiaries are not European companies. They are OpenAI, Anthropic, and Perplexity — the American AI firms that have been building search-capable chatbots without access to the underlying data flywheel Google enjoys. According to Comscore data from December 2025, published via FT Markets, ChatGPT reached 34.5 million mobile unique visitors in the US, with Google Gemini at 12.8 million. In Europe, where Gemini is more deeply integrated into Android but AI Overviews are not available due to regulatory constraints, the competitive dynamics are different — and the specification decisions tilt them further.

For European AI startups — Mistral, Aleph Alpha, and others — the Android interoperability measures lower the barrier to building assistant experiences, but the search data sharing benefits them less directly, since few operate general-purpose search engines. The Commission's bet is that access will stimulate a broader ecosystem of privacy-focused and specialized search alternatives over time.

Google loses something more valuable than cash: time. The two-year implementation window for Android interoperability (August 2027–2028) gives the company room to adapt, but the search-data sharing timeline is far tighter. Within 1.5 months of the decision, by end of August 2026, Google must publish technical specifications. Within 4 months, by November 2026, it must share its anonymization methodology with the Commission. The API must be operational shortly thereafter. By early 2027, rival search engines and AI chatbots will have access to Google's most valuable data asset on a continuous, real-time basis.

Diplomat View

The July 16 specification decisions represent the most ambitious use of the DMA's structural remedy powers to date — and they reveal a Commission that has learned from the limitations of its earlier approach. Fining Google €10 billion over a decade did not change the search market. The DMA's self-executing compliance model, backed by specification proceedings that produce granular, binding technical measures, might.

The thesis: These decisions will not create a European Google. That is not the point. They are designed to ensure that the next AI-driven search market is not won by default — that the company with the largest existing data hoard and the most locked-in mobile ecosystem cannot simply convert its search monopoly into an AI monopoly without competitors getting a genuine shot. The measures are imperfect: the anonymization requirements risk producing data too noisy to be useful, the implementation timelines give Google a head start, and the Android interoperability rules still allow Google to control the certification process.

But the direction of travel is unambiguous. Brussels is no longer content to fine after the fact. It is intervening in the architecture of competition before the game is over. For OpenAI, Anthropic, and any company building AI that needs to understand what users actually search for and click on, that intervention is a regulatory windfall. For Google, it is a structural tax on its most durable advantage.

What to watch:

• By end of August 2026: Google must publish the technical specifications for the search data sharing API — the first public test of whether the company will deliver workable access or comply in form while defeating the remedy in substance.
• By November 2026: Google's anonymization methodology and latency specifications are due to the Commission — a critical juncture where privacy requirements and data utility will collide.
• February–May 2027: Google publishes the draft and final eligibility terms for AI assistant certification on Android — the moment that determines whether the door is genuinely open or just appears so.

The Bottom Line

The European Commission has concluded that Google's next monopoly, in AI, will not be fought on the merits of its models alone but on who controls the operating system features and search data those models need. By ordering Google to open both, Brussels is running an experiment no other jurisdiction has attempted: can you regulate a data flywheel out of existence before it locks in the next market? The answer will become visible not in courtroom rulings but in whether, by 2028, a European Android user can say "Hey Claude" and get an answer as useful as Gemini's.