Arthur Mensch, the co-founder and chief executive of Mistral AI, has warned that Europe risks losing meaningful control over artificial intelligence if American companies lock up the chips, energy and data-center capacity needed to run advanced models.

The comments, reported Sunday by The Times of India and earlier by Business Insider, were drawn from Mensch’s testimony before a French National Assembly inquiry into structural digital dependencies and systemic vulnerabilities. The hearing took place on May 12 in Paris, where lawmakers questioned Mistral representatives as part of a broader investigation into France’s technological independence.

“Once supply is monopolized by American players, suddenly we no longer have supply and we can no longer transform electrons into tokens,” Mensch said, according to Business Insider’s translation of his remarks. The phrase captured the central strategic concern now shaping AI policy on both sides of the Atlantic: the most advanced models require not only software talent, but access to scarce accelerators, low-cost power, data-center capacity and cloud-scale deployment networks.

Mistral, founded in 2023 and headquartered in Paris, has become the most visible European challenger to U.S. AI companies. It develops large language models and has promoted an open-weight strategy in contrast with the more closed systems associated with many leading American platforms. The company’s rise has made Mensch one of the most closely watched executives in the global AI debate, especially as European governments look for ways to reduce dependence on Silicon Valley infrastructure without walling themselves off from U.S. technology.

The warning lands at a moment when the AI industry is increasingly defined by infrastructure commitments. U.S. companies and their partners have been racing to secure long-term access to Nvidia chips, custom accelerators, cloud capacity, power generation and specialized data centers. That shift has changed the competitive map. Model performance remains important, but so does the ability to guarantee reliable inference at scale for millions of users and enterprise customers.

For U.S. technology companies, the dynamic is a strategic advantage. American firms dominate much of the cloud market, lead the most valuable AI model companies, and sit close to the chip ecosystem that powers the current generation of generative AI. Nvidia remains the central supplier of high-end AI accelerators, while Microsoft, Google, Amazon and Oracle have become critical infrastructure providers for model developers and enterprise customers. OpenAI, Anthropic, Meta and Google DeepMind have also helped set the pace for frontier AI development.

Mensch’s argument is that this concentration could leave Europe with limited negotiating power if it does not build more of its own stack. In his framing, the race is not only about whether European startups can produce competitive models. It is about whether Europe can obtain enough compute and energy to keep operating them independently if supply tightens or commercial terms change.

That concern is closely tied to “sovereign AI,” a term used by governments and companies to describe AI systems that can be deployed under local legal, security and data-governance conditions. In practice, sovereign AI often means local hosting, domestic or regional control over sensitive data, and reduced reliance on foreign cloud or model providers. The idea has gained traction in Europe as banks, public agencies, defense organizations and regulated industries consider where their data is processed and who controls the underlying systems.

Mistral has tried to turn that concern into a business strategy. Reuters reported in March that the company secured $830 million in debt financing to buy 13,800 Nvidia chips for a large-scale data center near Paris, with financing from a consortium of banks including BNP Paribas, Crédit Agricole CIB, HSBC and MUFG. The data center in Bruyères-le-Châtel was scheduled to go live in the second quarter of 2026, according to that report.

The financing marked a shift for Mistral from pure model development toward infrastructure. Reuters also reported that the company planned further expansion in Sweden and aimed to reach 200 megawatts of computing capacity across Europe by the end of 2027. Those figures remain far smaller than the most ambitious U.S. infrastructure programs, but they show how European AI companies are beginning to finance compute as a strategic asset rather than simply renting capacity from U.S. cloud providers.

Mistral has also pushed deeper into institutional partnerships. A recent partnership with Groupe Caisse des Dépôts and its subsidiaries, including La Banque Postale and Bpifrance, was framed around digital sovereignty, generative AI deployment and GPU computing capabilities. For European governments and regulated businesses, such partnerships are designed to make AI adoption less dependent on foreign-controlled infrastructure while still using globally sourced hardware where necessary.

The paradox is that even Europe’s sovereignty push currently depends in part on American technology. Mistral’s data-center plan relies on Nvidia chips, and many European AI deployments still interact with U.S. software, cloud or semiconductor ecosystems. That makes the issue less a clean break from the United States than an attempt to avoid single-vendor or single-region dependence.

For Silicon Valley, Mensch’s warning is a sign that U.S. AI dominance is producing both commercial opportunity and political resistance. American cloud providers and chip companies stand to benefit from global demand for AI capacity, but the same dominance can spur governments to fund domestic alternatives, tighten procurement rules, scrutinize cloud concentration and favor local providers for sensitive workloads.

The European debate also overlaps with competition policy. Regulators in the United States, the European Union and the United Kingdom have already examined whether partnerships, cloud credits, model distribution and access to critical inputs could entrench a small group of firms. AI infrastructure concentration raises questions that resemble earlier platform-era fights, but with higher capital requirements and more direct links to energy systems and national security.

Mensch’s testimony did not suggest that Europe can replace the U.S. AI stack overnight. Instead, his message was that the next two years may determine whether Europe can preserve enough capacity to negotiate, compete and deploy models on its own terms. In that view, dependence becomes dangerous not when foreign tools are used, but when there is no credible alternative supply.

The U.S. relevance is direct. If Europe accelerates sovereign AI procurement, American companies could face a more fragmented global market, with more demand for local hosting, open models and jurisdiction-specific compliance. If Europe fails, U.S. firms could deepen their hold over the global AI economy, making American infrastructure the default layer for much of the world’s digital activity.

Mistral’s warning therefore reflects more than a European industrial-policy debate. It points to the next stage of the AI race, where national leverage may depend on who can turn electricity into tokens at scale, and who controls the supply chain when demand exceeds capacity.