EU Antitrust Expands to Nvidia and AMD AI Chip Dominance

The Digital Markets Act Foundation

The DMA, which came into force in 2023, gave the European Commission new enforcement authority over what it calls "gatekeepers": companies with sufficient market power in core platform services that their behavior can effectively determine access conditions for competitors, business users, and consumers across the European digital economy. The DMA was explicitly designed to reach situations where market structure problems were not being solved quickly enough by traditional competition law, which requires proving market harm that has already occurred rather than preventing it prospectively.

The DMA's initial focus on consumer-facing platforms (app stores, search engines, social networks, online marketplaces) reflected the most visible and politically salient forms of tech market power. Extending it to semiconductor companies requires a slightly different analytical frame. Nvidia does not run a platform in the traditional sense. It manufactures chips. But the argument the EU is developing is that control over the hardware infrastructure on which AI systems run is functionally equivalent to control over a core platform service: if you cannot access competitive AI chips, you cannot compete in AI markets, regardless of your software capabilities or business model.

The CUDA (Compute Unified Device Architecture) ecosystem is central to this argument. CUDA is Nvidia's proprietary programming framework that allows developers to write software that runs efficiently on Nvidia GPUs. Over more than a decade, CUDA has become the dominant software layer for AI development: the vast majority of AI research papers, frameworks, and production systems are written to run on CUDA. This creates a lock-in dynamic that extends beyond hardware procurement. Even if AMD's hardware specifications were competitive (and the MI300X series has closed the gap in some benchmarks), switching to AMD requires rewriting or porting software that was built for CUDA. The switching cost is not just the hardware cost; it is the accumulated development investment in a software ecosystem.

Why AMD Is Also Named

AMD's inclusion alongside Nvidia in the EU's expanding scrutiny reflects a different concern. If the investigation focused solely on Nvidia, it would read as a straightforward dominant-position case. Including AMD suggests the EU is looking at the AI chip market structure more broadly, examining whether the duopoly structure (Nvidia dominant, AMD as the only meaningful alternative) itself creates conditions that disadvantage European buyers and AI developers.

AMD's ROCm software ecosystem, while open source, has historically lagged CUDA in maturity, documentation, and developer adoption. That gap has narrowed as AMD has invested heavily in ROCm development and as more AI frameworks have added native ROCm support. But the practical situation for a European AI startup or research institution in 2026 is that CUDA is significantly easier to use for AI development than any alternative. That ease of use is not an accident; it reflects over a decade of deliberate ecosystem investment by Nvidia.

The EU's concern with AMD may also relate to pricing dynamics. In markets with a single dominant player and one meaningful competitor, the competitor can price at a level that maintains the dominant player's premium pricing rather than competing aggressively enough to drive prices down to competitive levels. Whether that characterization applies to Nvidia and AMD in AI chips is precisely the kind of market analysis that the EU investigation would examine.

European Tech Sovereignty: The Political Urgency

The timing of the expanded antitrust scrutiny is not coincidental. Europe's strategic dependence on US technology has been a topic of policy discussion for years, but the Iran conflict has moved it from a medium-term concern to an immediate one. The war has disrupted global supply chains, made visible the ways in which geopolitical events can rapidly constrain technology access, and reinforced the argument of European technology nationalists who have been warning that dependence on US platforms for critical digital infrastructure is an unacceptable strategic vulnerability.

A Computerworld analysis from March 13 noted that European companies had been warning the EU about US tech dependence with increasing urgency. The concern is not hypothetical: if US-EU relations deteriorated significantly, or if US technology export controls were extended to cover European customers in ways they currently do not, European organizations that have built critical operations on US cloud infrastructure and US AI chips would face significant disruption with few immediately available alternatives.

The DMA investigations into Apple, Google, Meta, and Microsoft have already produced concrete behavioral changes: Apple has opened app sideloading in the EU, Google has modified its search results presentation in European markets, and Meta has offered a subscription tier as an alternative to its advertising data model. The EU has demonstrated that its regulatory tools can produce outcomes, which makes the extension of that toolkit to chip market dynamics a credible threat rather than a paper exercise.

For European AI companies, the implications are direct. If Nvidia is found to have maintained its market position through anti-competitive practices tied to CUDA lock-in, the remedies available to the EU could include requirements to license CUDA more openly, restrictions on bundling CUDA licenses with hardware purchases at discounted rates, or interoperability mandates that make it easier for AI software to run across Nvidia and AMD hardware without significant development overhead. Any of those outcomes would benefit European AI developers competing against US hyperscalers with deeper Nvidia relationships. For context on the scale of Nvidia's current order backlog and what it signals about market concentration, see Nvidia's Vera Rubin architecture announcement and its $1 trillion order backlog.

Nvidia's Defense and the Supply Chain Argument

Nvidia has not yet publicly responded in detail to the EU's expanded scrutiny, which is consistent with standard practice for companies under regulatory investigation. The company's likely defense is a combination of arguments that the AI chip market is relatively young and evolving rapidly (making dominant position claims premature), that CUDA's adoption reflects developer preference rather than forced lock-in, and that Nvidia's investment in research and development is what created the market performance advantage that its customers are purchasing.

The R&D investment argument has merit. Nvidia's current position in AI chips is not the result of acquiring competitors or regulatory lobbying. It is the result of a decade-long bet on GPU computing for parallel workloads that most of the technology industry, including Intel, initially dismissed. The company invested heavily in CUDA long before AI was the dominant use case, betting that general-purpose GPU computing would eventually find killer applications. When deep learning emerged as that application, Nvidia was the only company with a mature parallel computing software ecosystem ready to support it. That is competitive success, not abuse.

The counterargument, which the EU is likely developing, is that however the dominant position was acquired, its current maintenance involves practices (CUDA bundling, pricing structures, supply chain preferencing) that foreclose competition in ways that harm European buyers and the broader innovation ecosystem. The distinction between "deserved dominance" and "sustained dominance through exclusionary practices" is precisely the boundary that antitrust law is designed to examine.

The supply chain dimension adds another layer. Nvidia's chips are manufactured primarily at TSMC in Taiwan. EU regulators concerned about technology sovereignty cannot easily require Nvidia to manufacture in Europe or to transfer manufacturing technology to European companies; those decisions are outside the EU's direct regulatory authority. But the EU can affect the market conditions under which Nvidia's chips are sold and distributed in European markets, and it can require behavioral changes that benefit European AI companies even without touching the manufacturing supply chain directly.

The Broader DMA Enforcement Pattern

The EU's track record under the DMA and predecessor competition frameworks provides context for what the Nvidia and AMD investigation is likely to look like procedurally. The Apple App Store investigation produced its first major binding ruling approximately 18 months after the DMA came into force. The Google search investigation has been ongoing in various forms for over a decade under prior competition law and has produced fines, settlements, and behavioral commitments across multiple proceedings.

A semiconductor investigation will proceed more slowly than a software platform investigation, partly because the relevant market definition work is more complex (AI chips are a subset of GPUs, which are a subset of semiconductors, and each boundary matters for the analysis) and partly because the behavioral remedies in hardware markets are less straightforwardly applicable than in software markets. You cannot easily mandate an interoperability standard for chips the way you can mandate an API disclosure requirement for a social network.

What the EU can do is investigate, publish preliminary findings, issue statements of objection that the companies must respond to, and ultimately impose fines and behavioral commitments. The fines available under the DMA can reach 10 percent of global annual revenue, a figure that for Nvidia would represent billions of euros. That number gets attention in boardrooms even when the underlying investigation has years left to run.

The investigation also has a signaling function that operates independently of its ultimate legal outcome. By naming Nvidia and AMD explicitly, the EU is communicating to European AI companies that the Commission is aware of their dependence concerns and is taking action. That communication has value even before a single fine is imposed, because it affects the negotiating dynamics between the named companies and their European customers and partners. For coverage of the related cybersecurity threat environment that is shaping EU tech sovereignty urgency, see IRGC's cyberattack threat against US tech firms.

Implications for AI Investment in Europe

The investigation's implications for the European AI industry are not straightforward. European AI startups and research institutions are Nvidia's customers. If the EU's investigation produces remedies that lower CUDA switching costs or reduce Nvidia's pricing power, those customers benefit directly. If the investigation creates uncertainty about Nvidia's product roadmap or supply commitments in European markets (for example, if Nvidia responds to regulatory pressure by deprioritizing European customer relationships), those customers could be disadvantaged during the investigation period.

The broader EU political project around AI competitiveness is pursuing multiple tracks simultaneously. The AI Act, which is the EU's primary regulatory framework for AI system deployment, imposes compliance requirements that some argue disadvantage European AI developers relative to US and Chinese competitors who face less stringent domestic regulation. The DMA investigation of Nvidia and AMD could be read as an attempt to compensate for that competitive disadvantage by ensuring European developers have access to AI chips on terms that are not systematically worse than what US hyperscalers receive.

Whether those two regulatory efforts are actually coherent with each other (one adds compliance cost to European AI deployment, the other attempts to reduce chip access barriers) is a question that European AI policy observers have been raising with increasing volume. The EU has significant regulatory ambition in the technology sector. Whether that ambition translates into a coherent industrial strategy that benefits European AI development, rather than simply adding regulatory overhead to an already challenging competitive environment, is the central question the Commission's critics are asking.

The answer will depend in part on the pace and outcome of the Nvidia-AMD investigation, and on how the EU uses whatever behavioral commitments it extracts. It is a question worth watching closely over the next 24 months.