TL;DR: The European Commission is expanding antitrust scrutiny to cover Big Tech's entire AI operations, from semiconductor chips through model training, cloud infrastructure, and end-user deployment. EU Competition Commissioner Teresa Ribiera named Nvidia, Meta, and Google by name, warning each company it needs to address specific competitive concerns. The investigation is the broadest AI competition action any regulator has mounted to date.
Table of contents
- What the EU Competition Commissioner announced
- Which companies are targeted and why
- Full AI value chain scrutiny — chips to deployment
- The WhatsApp AI lockout investigation
- Digital Markets Act amendments on the table
- Nvidia's market dominance under the microscope
- How this differs from previous EU tech enforcement
- Impact on US-EU tech relations and trade
- What this means for AI startups and competition
- Timeline and what to expect next from Brussels
What the EU Competition Commissioner announced
On March 12, 2026, EU Competition Commissioner Teresa Ribiera delivered the clearest signal yet that Brussels intends to treat AI market concentration as an urgent antitrust priority.
Ribiera's remarks, reported by Bloomberg, went beyond any previous regulatory statement on AI competition. She explicitly called out Nvidia and Meta for "entrenching corporate power" in AI markets. She told Google it must "come forward with a serious remedy" to address its dominant position. And she announced that the Commission is reviewing whether the Digital Markets Act needs amendments to address AI-specific competition risks.
The announcement came as part of the European Commission's broader competition policy agenda for 2026. But the specificity was new. Ribiera named companies, named products, and named practices. This was not a general warning about AI market concentration — it was a targeted list of concerns the Commission intends to act on.
The scope is sweeping. Ribiera indicated the investigation covers the full AI value chain: the chips used to train models, the cloud infrastructure where models run, the models themselves, and the applications deployed on top of those models. No prior EU competition investigation has attempted to regulate a technology stack end-to-end at this scale.
The Commission's concern, as articulated by Ribiera, is that a small number of American technology companies are locking in positions across every layer of AI simultaneously. If that consolidation continues unchecked, Brussels worries European AI companies will find themselves permanently dependent on US-controlled infrastructure, models, and distribution channels.
Which companies are targeted and why
Three companies were mentioned explicitly. Each faces a different set of concerns.
Nvidia is the dominant supplier of GPUs used to train and run large AI models. Its H100 and B200 chips are effectively the only hardware capable of training frontier models at scale. The Commission is examining whether Nvidia's market position in AI chips gives it the ability to distort competition across the entire AI ecosystem by allocating compute preferentially, bundling hardware with proprietary software, or pricing in ways that disadvantage European customers and competitors.
Meta faces scrutiny on two fronts. First, its AI policy for WhatsApp Business has drawn specific attention: the policy prohibits third-party AI providers from building services on top of WhatsApp Business using competing AI systems like ChatGPT. The Commission views this as a potential foreclosure of AI competition on one of the world's largest messaging platforms. Second, Meta's open-source Llama models, while freely available, are distributed with commercial restrictions that Brussels is examining for potential anti-competitive effects.
Google is being asked to provide remedies for its position across AI search, cloud infrastructure (Google Cloud, TPUs), and AI model distribution (Gemini). Ribiera's phrasing — "come forward with a serious remedy" — is the language Brussels uses when informal resolution is expected before formal charges. Google has already faced multiple EU antitrust fines exceeding €8 billion. A new AI-related enforcement action would dwarf previous proceedings in scope.
Full AI value chain scrutiny — chips to deployment
The most significant aspect of Ribiera's announcement is not the specific companies named — it is the declared scope of investigation. The Commission is treating the AI stack as a single competitive market rather than separate sectors.
Previous EU tech regulation drew clean lines: chips were hardware, models were software, apps were services. Each had its own regulatory framework. The DMA focused on platform gatekeepers. EU semiconductor policy addressed chips through industrial subsidies. Model safety regulation was addressed by the EU AI Act. Competition law applied to specific markets.
Ribiera's framing collapses those distinctions. By stating that scrutiny covers the entire value chain "from chips to deployment," the Commission is signaling that it will examine how dominance in one layer reinforces dominance in another. A company that controls AI chips, cloud infrastructure, model training, and app distribution simultaneously creates a vertically integrated monopoly that is harder to regulate layer by layer.
This end-to-end framing is borrowed from the Commission's analysis of the smartphone market, where Apple and Google's control of both hardware and software ecosystems prompted DMA designation. Applied to AI, it creates a much larger regulatory surface.
The practical implication: companies that are dominant in only one AI layer (a chip maker, a cloud provider, a foundation model company) may still face antitrust scrutiny if their dominance in one layer creates leverage across adjacent layers. That expands the potential scope well beyond Nvidia, Meta, and Google to include Microsoft, Amazon Web Services, and potentially Anthropic and OpenAI.
The WhatsApp AI lockout investigation
The Meta WhatsApp investigation deserves specific attention because it illustrates exactly the kind of platform leverage Brussels is most concerned about.
Bloomberg's reporting revealed that Meta's current WhatsApp Business API terms prohibit AI service providers from using the platform to power competing AI assistants. Concretely: a company building a customer service chatbot using OpenAI or Anthropic's models cannot deploy that chatbot via WhatsApp Business if the chatbot's AI capabilities compete with Meta's own AI products.
This is a direct analog to the browser choice and search choice restrictions that produced the Commission's largest prior fines against Google. Google was penalized for pre-installing Chrome and Google Search on Android devices, making it harder for competing browsers and search engines to reach users. Meta is accused of doing the same thing with AI: using WhatsApp's 3 billion users as a lever to foreclose competing AI providers from one of the world's most important business communication channels.
The WhatsApp Business platform handles customer service interactions for hundreds of thousands of businesses globally. If Meta can restrict which AI systems power those interactions, it gains a structural advantage for its own AI products (Meta AI, Llama-based tools) that has nothing to do with the quality of those products.
Under the DMA, Meta was designated as a gatekeeper for WhatsApp in 2023. The DMA explicitly prohibits gatekeepers from using their platform position to advantage their own services over third parties. The WhatsApp AI lockout appears to be a direct violation of this obligation, which may be why Ribiera singled it out.
Digital Markets Act amendments on the table
The most consequential long-term element of Ribiera's statement may be the least dramatic: she indicated the Commission is actively considering amendments to the Digital Markets Act to address AI-specific competition risks.
The DMA was passed in 2022 and took effect in 2023. It was designed primarily around existing platform behaviors — app stores, search, messaging, browsers. It was not written with foundation models, AI chips, or inference infrastructure in mind.
The core gatekeeper designation process in the DMA focuses on end-user-facing platforms with high user counts and revenue thresholds. A company like Nvidia, which sells hardware to data centers rather than consumers, does not fit the DMA's current gatekeeper definition despite controlling a resource (AI compute) that every AI company depends on.
Possible amendments being discussed include:
- AI infrastructure gatekeeper designation: Extending gatekeeper rules to companies controlling critical AI infrastructure (chips, cloud compute) above market-concentration thresholds.
- Foundation model interoperability requirements: Requiring dominant foundation model providers to publish APIs and allow third-party fine-tuning and deployment without restrictive commercial terms.
- Data access obligations: Requiring companies that use proprietary data advantages (search logs, social graph data, enterprise data) to train AI models to provide competing AI companies with equivalent data access under regulated terms.
- AI distribution fairness rules: Preventing companies from using dominant distribution channels (app stores, messaging platforms, search results) to preference their own AI products.
DMA amendments require approval from EU member states and the European Parliament. The process typically takes 18-24 months. But the Commission can begin enforcement under existing DMA rules while amendments proceed.
Nvidia's market dominance under the microscope
Of the three companies named, Nvidia is arguably the most structurally significant and the least prepared for EU antitrust scrutiny.
Nvidia controls an estimated 80-85% of the market for AI training chips. Its CUDA software platform, which AI models depend on for hardware acceleration, is a proprietary ecosystem with no competitive alternative at scale. AMD's MI300X chips offer comparable performance on some benchmarks, but software ecosystem compatibility has kept Nvidia dominant.
The EU's specific concerns about Nvidia are not yet fully public, but based on Ribiera's "entrenching corporate power" language, several practices are likely under examination:
Compute allocation: During chip shortages in 2023-2024, Nvidia prioritized supply to large US hyperscalers (Microsoft, Google, Amazon, Meta) over European cloud providers and AI companies. European firms found themselves at the back of a queue for hardware that US competitors had already secured through multi-year supply agreements.
CUDA lock-in: Nvidia's CUDA platform is designed to run only on Nvidia hardware. Models optimized for CUDA perform significantly worse on competing chips. This creates switching costs that make it costly for AI companies to migrate away from Nvidia even as alternative hardware becomes available.
Bundling: Nvidia's AI Enterprise software suite, DGX systems, and networking products (Mellanox/InfiniBand) are often sold as integrated packages with pricing structures that disadvantage component-level competitors.
How this differs from previous EU tech enforcement
The EU's track record on tech antitrust spans two decades. It has fined Google, Microsoft, Apple, and Meta billions of euros across dozens of proceedings. But this AI investigation represents a meaningful departure from previous enforcement patterns.
Previous EU tech enforcement was mostly reactive. The Commission opened cases after markets had already consolidated around dominant players. Google's shopping search case concluded 10 years after the conduct began. Apple's App Store investigation dragged over multiple product cycles. The enforcement timeline meant that by the time remedies were imposed, market structure had already calcified.
Ribiera's announcement is explicitly pre-emptive. She said Brussels is acting now, while AI market structure is still being established, rather than waiting until concentration is irreversible. The phrase "entrenching corporate power" implies the Commission is trying to prevent a market structure from forming, not unwind one that already exists.
This is a harder legal case to make. EU competition law generally requires demonstrating existing market dominance and specific anti-competitive effects. Arguing that a market is being monopolized in real time, before dominance is fully established, requires a different evidentiary standard. The Commission will need to demonstrate both current market power and likely foreclosure effects — a higher bar than simply pointing to an existing monopoly.
The other difference: the Commission is coordinating with the UK Competition and Markets Authority (CMA) and the US Federal Trade Commission on AI competition concerns. Previous EU enforcement was often conducted unilaterally, producing remedies that applied only in Europe. A coordinated multi-regulator approach could produce global structural changes.
Impact on US-EU tech relations and trade
Ribiera's announcement lands in a politically sensitive moment. US-EU trade tensions were already elevated in early 2026 over tariff disputes. Targeting Nvidia, Meta, and Google with antitrust investigations will likely be framed in Washington as discriminatory treatment of American technology companies.
This narrative is not new. US officials have argued for years that EU tech regulation is protectionist industrial policy dressed in competition law clothing. The European Commission has consistently denied this, pointing to the fact that most fines are levied against conduct that disadvantages European consumers and competitors, not simply against American companies.
The AI antitrust push does not change this fundamental disagreement, but it will intensify it. Nvidia's GPU dominance has become a US national security and industrial policy priority. The Commerce Department has placed export controls on advanced chips specifically to maintain US AI leadership. EU antitrust action that constrains Nvidia's ability to prioritize US customers or maintain CUDA's competitive advantages could create a direct conflict between EU competition law and US technology policy.
Three possible scenarios for US-EU AI tech relations:
- Negotiated settlement: Large fines are avoided in exchange for structural commitments (interoperability, data access, chip supply fairness). This was the resolution for most prior Google cases.
- Escalation: If the Commission opens formal proceedings, US tech companies may lobby Washington for retaliatory trade measures, escalating into a broader digital trade conflict.
- Divergent regimes: EU and US regulatory frameworks diverge permanently, requiring AI companies to maintain separate compliance architectures for each jurisdiction — increasing costs and potentially fragmenting AI development.
The most likely outcome is negotiated settlement, as it has been in nearly every prior EU tech case. But the AI context makes the negotiations more complex because the stakes — control of the next-generation technology platform — are higher than previous disputes over app stores or search rankings.
What this means for AI startups and competition
For European AI startups, Ribiera's announcement has a straightforward implication: Brussels is trying to create conditions under which they can compete against hyperscaler-backed AI products.
The practical barriers facing European AI companies are significant.
Access to AI compute is the most acute problem. Training a competitive frontier model requires thousands of Nvidia GPUs running continuously for months. The cost of that compute runs into the tens of millions of dollars at cloud market rates. European AI startups have found it harder to secure GPU allocations from Nvidia and hyperscalers than their US counterparts, which have direct supply relationships and preferential pricing.
Distribution is the second constraint. If WhatsApp, Google Search, and iOS/Android app stores can preference the AI products of Meta, Google, and Apple, European AI startups face a structural distribution disadvantage that no amount of technical innovation can overcome. The DMA was supposed to address this for app distribution. The WhatsApp AI lockout investigation suggests it has not fully achieved that goal.
Data is the third barrier. Companies with proprietary search logs, social graphs, and enterprise data have training data advantages that are structurally inaccessible to startups. A European company building a search AI product cannot replicate Google's three-decade accumulation of search behavior data. Regulatory data access requirements could partially close this gap.
If Ribiera's investigation produces real remedies — fair chip allocation, open AI distribution, data access — the EU could succeed in creating a viable European AI competitive tier. If the investigation produces only fines and symbolic commitments, as some prior EU tech cases have, the structural advantages of US hyperscalers will continue to compound.
Timeline and what to expect next from Brussels
EU competition investigations follow a predictable sequence: informal inquiry, formal investigation, statement of objections, remedies, and potential fines. The full process can take three to five years for complex cases.
Based on Ribiera's public statements, here is a realistic timeline for what comes next:
Q2 2026 (April-June): Formal information requests sent to Nvidia, Meta, and Google. Companies are required to provide detailed data on pricing, market shares, supply chain relationships, and AI product architectures. This phase is not public but sets the evidentiary foundation for any subsequent proceedings.
Q3-Q4 2026 (July-December): The Commission assesses responses and decides whether to open formal proceedings. Google's instruction to "come forward with a serious remedy" suggests informal settlement discussions may run in parallel, potentially avoiding formal charges if acceptable commitments are made.
2027: If formal proceedings are opened, statements of objections are issued. Companies have the right to respond, request oral hearings, and provide additional evidence. This phase typically adds 12-18 months.
2027-2028: Remedies and potential fines. DMA violations can attract fines up to 10% of global annual turnover. Nvidia's annual revenue was approximately $130 billion in fiscal 2026. A maximum DMA fine would be $13 billion — larger than any prior EU tech penalty.
The DMA amendment process runs separately. Ribiera's comments suggest a Commission proposal could come in late 2026, with legislation possible by 2028.
What to watch: Whether Google proactively offers AI-related remedies in the coming months will be an early signal of how the Commission intends to proceed. Google settled its 2023 DMA concerns about Android interoperability with commitments rather than fines. A similar dynamic in AI would suggest Brussels prefers structural change over punitive enforcement.
Frequently asked questions
What is the EU investigating in Big Tech AI?
The European Commission is examining whether Nvidia, Meta, and Google have used their dominant positions across AI chips, cloud infrastructure, models, and distribution to foreclose competition across the entire AI value chain.
Why did the EU name Nvidia in the antitrust investigation?
Nvidia controls an estimated 80-85% of the AI training chip market. EU Competition Commissioner Teresa Ribiera said Nvidia and Meta are "entrenching corporate power" in AI markets. The Commission is examining chip allocation practices, CUDA software lock-in, and bundling strategies.
What is the WhatsApp AI lockout?
Meta's WhatsApp Business API terms currently prohibit third-party AI providers from using the platform to power AI assistants built on competing AI systems. The Commission views this as using WhatsApp's 3 billion users to foreclose competing AI providers.
What is the Digital Markets Act and how does it apply to AI?
The DMA is EU legislation requiring large platform gatekeepers to maintain fair, open markets. It was passed in 2022 and designated Meta, Google, Apple, and others as gatekeepers. Ribiera announced potential DMA amendments to extend gatekeeper obligations to AI infrastructure, foundation models, and AI distribution.
DMA violations can result in fines up to 10% of global annual turnover. Nvidia's fiscal 2026 revenue was approximately $130 billion, making a maximum fine around $13 billion. Meta and Google have global revenues of $160 billion and $350 billion respectively.
How long will the EU AI antitrust investigation take?
Complex EU competition investigations typically take three to five years from formal proceedings to final decision. Informal settlements may be reached faster — potentially within 12-18 months if companies offer acceptable structural remedies.
Does this affect AI companies outside the EU?
Yes. EU competition remedies have global effects because they often require structural changes to products and business practices that apply company-wide, not only in European markets. Prior Google remedies required changes to Android that applied internationally.
What does this mean for European AI startups?
If the investigation produces real remedies (fair chip allocation, open AI distribution, data access requirements), European AI startups will face a more level competitive landscape against US hyperscaler-backed AI products. Fines alone would not achieve this.
Key takeaways
- EU Competition Commissioner Ribiera named Nvidia, Meta, and Google by name, citing "entrenched corporate power" in AI markets — the strongest language Brussels has used on AI competition.
- The investigation covers the full AI stack: chips, cloud infrastructure, foundation models, and deployment — not any single layer in isolation.
- Meta's WhatsApp AI lockout (blocking competing AI systems from the WhatsApp Business platform) is a specific DMA violation concern that could produce fast enforcement under existing rules.
- Google was told to "come forward with a serious remedy" — language that typically precedes informal settlement negotiations rather than immediate formal charges.
- The Commission is actively considering DMA amendments to extend gatekeeper obligations to AI infrastructure providers like Nvidia that don't fit the current gatekeeper definition.
- DMA fines can reach 10% of global annual revenue — at Nvidia's current revenue, that is a potential $13 billion penalty.
- The investigation is deliberately pre-emptive: Brussels is trying to prevent AI market concentration from calcifying, not unwind it after the fact.
- The most likely resolution follows previous EU tech cases: large companies offer structural commitments in exchange for avoiding maximum fines, producing partial but imperfect competitive outcomes.