Brussels takes aim at Google's Android grip as Gemini 3.5 stalls

The European Commission on 16 July 2026 ordered Alphabet's Google to open its Android mobile operating system and to hand rivals a slice of its search data, escalating a fight that now sits at the intersection of two regulatory wars the bloc has been running for years. Hours earlier, Google had confirmed it was pushing back the release of Gemini 3.5 Pro, its next frontier large language model, so the company could sharpen the product's coding capabilities before launch. Read together, the two moves tell a clear story. Brussels is picking its moment. And Google, for the first time in this cycle, is on the back foot in both places at once.

The Commission's order lands inside a Digital Markets Act framework that already treats Android and Google Search as "core platform services" subject to mandatory interoperability and self-preferencing bans. What is new is the bite. Under the latest directive, Google must give competing AI assistants real-time access to search index data on non-discriminatory terms, and must allow rival app stores and engines to sit at the same level of the Android stack that its own products enjoy. The Commission framed the move as a structural correction, not a fine. For American and Asian AI challengers who have complained for two years that they cannot match Google's distribution and data gravity, the door has just been pushed open, and pushed open from the EU side rather than from Washington.

What the order actually requires

The text of the Commission decision, as summarised in the Telegram wire that surfaced on 16 July at 19:17 UTC, obliges Google to grant "effective, real-time access" to search results, ranking signals, and click-and-query data to third-party AI assistants that meet objective technical standards. The same directive forces Android to accommodate alternative app marketplaces and engines at the system level, on terms comparable to Google's own. A second Digital Markets Act proceeding, still pending, sits on top of this one: an antitrust probe opened in 2024 into whether Google favours its own Gemini models in Search and on Pixel devices. None of that is incidental. The Commission is layering remedies so the cost of non-compliance compounds with each quarter Google drags its feet. Past DMA fines against the company have run into the billions of euros; the new order is more dangerous because it touches the product itself.

Google's public response has been procedural rather than substantive. The company says it will study the order and engage with the Commission, the same posture it took when the DMA was first enforced. That posture has bought it time before, but the political weather in Brussels has changed. France, Germany, and the Netherlands have all signalled in recent months that they want the Commission to use DMA powers more aggressively. The German federal cartel office's case against Google in 2025 set the tone: structural remedies, not fines. Berlin has been pushing the same line in BEREC consultations on telecom interoperability. The current Commission has an institutional interest in proving the DMA bites before parliamentary renewal debates in 2027.

The Gemini 3.5 delay, and what it signals

On the product side, Google is in a different kind of trouble. The decision to delay Gemini 3.5 Pro, first reported on 16 July at 19:29 UTC, was framed by the company as a quality decision. Coding has become the AI industry's most commercially consequential benchmark. Anthropic's Claude 4.5, OpenAI's GPT-5 family, and a string of open-weight models from Chinese labs have all staked ground there over the last six months. Google is signalling that it does not want to ship a frontier model that loses on a benchmark developers actually use to pick their default provider. The risk of a soft launch is reputational as much as technical. Once a developer picks a default model for code completion, switching costs compound inside the IDE. Google's internal calculus, if the public framing is to be believed, is that a delayed clean release beats a punctual embarrassment.

There is also a corporate-finance read. Reuters and Bloomberg have spent the better part of 2026 documenting how Alphabet is repricing its AI capex in front of Wall Street. Slower model releases let the company defer depreciation on the latest training cluster by a quarter. That is not the official reason, and it should not be presented as one, but it is the kind of consideration a CFO raises in any release-timing conversation at a firm spending tens of billions a year on inference and training infrastructure.

The structural frame, in plain prose

What we are watching is not a single antitrust case. It is the collision of two regulatory regimes with two different theories of harm. The American model, still dominated by case-by-case FTC and DOJ litigation, treats dominance as a question of consumer prices and output. The European model, codified in the DMA and the AI Act, treats dominance as a question of structural access. The Brussels theory is that a platform's data and distribution are themselves essential facilities, and that downstream markets for AI assistants, app stores, and search cannot function competitively while those facilities are privately held. The order against Google is the clearest test yet of whether that theory works in practice for AI. If it does, expect Brussels to reach next for Microsoft's bundling of Copilot into Office, and for Apple's grip on iOS distribution in Europe.

A counter-reading deserves airtime. American and some British commentators will frame the order as protectionism dressed up as competition policy. The argument: European AI labs are behind, so Brussels is using regulation to do what subsidies could not. There is some truth in the optics. Mistral, Aleph Alpha, and the newer German and French model shops cannot match the frontier-lab spend of US hyperscalers. But the DMA's logic does not depend on European firms being competitive. It depends on the platform being open. That distinction matters. The order is structural, not industrial. Whether that structural remedy produces European AI champions is a separate question, and one the Commission cannot answer by decree.

What to watch next

The procedural clock now starts. Google has until late summer 2026 to file its formal compliance plan; the Commission has until autumn to either accept the plan or move to enforcement, which under the DMA can mean fines of up to ten percent of global turnover and, in the most serious cases, structural break-up. Watch three dates. First, the September 2026 deadline for Google's compliance filing. Second, the autumn ruling from the EU's General Court in the parallel appeal Google lodged against its earlier DMA designation. Third, the EU AI Act's general-purpose-model obligations, which come into force in stages through 2027 and which will force Google to disclose training-data provenance for Gemini-class models. A company that has just been told to share its search index with rivals will find the AI Act's transparency requirements harder to stonewall than it has in Washington.

The Gemini 3.5 delay is the smaller story in volume terms but the larger one in narrative terms. It tells the market that the frontier is not a fixed point; it is a contest with a moving finish line. Brussels is betting that contest stays open long enough for its structural remedies to bite. Alphabet is betting it can ship fast enough to make the remedies moot. The next six months will tell us which of those bets was the mistake.

Desk note: Monexus framed this as a two-front story, EU regulation and US product delay, because the wires ran them as separate threads on the same day. The structural connection is the piece's own contribution.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://t.me/CryptoBriefing
• https://t.me/CryptoBriefing
• https://t.me/CryptoBriefing