Europe's leading AI developers, France's Mistral, are pushing back against US frontier labs that have urged the industry to ease the pace of AI capability breakthroughs. Their objection centers less on safety than on timing: Europe already trails the United States and China in notable AI models, computing power, and development funding, and its challengers fear that coordinated pacing by the biggest labs would only make the competitive divide harder to close.

The pushback arrives as Brussels pairs rulemaking with large-scale investment, and as fresh data quantify how far the continent sits from the frontier. According to Prosus, Europe accounted for 11% of global AI venture funding as of 2023, compared with 77% captured by the United States — a gap that widens at later stages of funding. The venture firm's findings underline the fundraising imbalance that European challengers say leaves them at a structural disadvantage.

'Totally self-serving,' say Europe's challengers

The dispute stems from a proposal by Anthropic CEO Dario Amodei to slow the rate of development of new AI and to seek an antitrust exemption that would let laboratories cooperate on safety matters. Reuters reported that Mistral's representatives believe established players are "pushing for regulation designed to favour them over competitors."

Raphael Auphan, chief operating officer of Proton, was more blunt, calling the initiatives "totally self-serving" and saying they serve the interests of US companies by keeping customers dependent on their services. Black Forest Labs argued that "arbitrary thresholds" can impede innovation, while Hugging Face chief Clément Delangue said Europe should keep advancing, though he backed Amodei's call for independent assessment.

What Amodei actually proposed

In a September 12 essay titled "We Must Pace Frontier", Amodei advocated slowing advances in AI capability while product releases and training continue. He laid out a three-step approach: installing external evaluators with employee-level access to labs' development processes, coordinating among laboratories in democratic countries, and eventually extending the arrangement to international collaboration.

The idea drew swift backing from Elon Musk and OpenAI CEO Sam Altman. Altman endorsed Anthropic's concept of embedded evaluators while cautioning that pacing should not be understood as stopping development.

According to Forbes, the debate spilled into markets on September 14, when shares of major AI companies fell: Intel dropped roughly 7% to around $96, AMD slid about 6% to $487, and Nvidia declined about 3% to $212.

The gap Europe is trying to close

Europe's stance reflects a substantial disparity in model progress and the infrastructure behind it. In a September 14 address, European Central Bank President Christine Lagarde noted that the United States produced 59 notable AI models in 2025, China released 35, and France and the United Kingdom developed one each. She added that the US accounts for roughly 75% of global AI data-center computing capacity, while Europe holds just 5% — remarks delivered two days after Amodei's essay.

The most recent data from Epoch AI point to a capability gap at the top of the field as well. US-made models lead Epoch's ECI ranking, headed by GPT-6 Astra at 166 points. China's Kimi K3 ranks 11th at 158 points, eight behind the US leader, while France's Mistral Medium 3.5, released in April, scored 142.5 points.

Europe lags far less in adoption, however. Microsoft's Global AI Diffusion report found that generative AI usage among working-age individuals stood at 48.6% in Norway, 48.4% in Ireland, and 47.8% in France, compared with just 31.3% in the United States. The figures suggest Europe is advancing faster in using AI than in building cutting-edge models and the infrastructure to train them.

Brussels is spending, not retreating

The EU is pairing regulation with investment. Its AI Continent Action Plan aims to mobilize €200 billion for AI, including a €20 billion InvestAI facility to finance up to five AI gigafactories, alongside at least 19 AI factories. The European Commission also notes that only 13.5% of EU companies currently use AI — a gap its Apply AI strategy is meant to narrow. Those programs amount to the EU's direct answer to the compute and adoption shortfalls its own officials describe, and each stage of their rollout — financing, construction, operation — offers a measurable checkpoint for whether those shortfalls begin to close.

Who pays for a slowdown

The competition concern is not purely theoretical. An OECD report found that concentration in AI innovation is associated with higher sales concentration, while AI startups attracting venture capital are still frequently acquired by large incumbents. That does not prove pacing rules would protect today's leaders, but it helps explain why smaller developers worry that new requirements could become another cost of entry. The antitrust exemption request sits at the center of that worry: it would formalize cooperation among the very incumbents that challengers such as Mistral and Proton are trying to catch.

The stakes extend beyond Europe. Cryptopolitan reported that Chinese state media framed the US slowdown push as a "Cold War" tactic aimed at preserving Washington's lead. If US frontier labs slow while European and Chinese challengers keep pushing ahead, the result may be a shift in capital, talent, and open-model development rather than a uniform global slowdown. The near-term checkpoints are concrete: how competition authorities respond to the antitrust exemption request, how quickly the EU's €200 billion mobilization and its gigafactory and Apply AI programs move from announcement to operating capacity, and successive Epoch AI rankings show about the capability gap as new models ship.