You’re never better defended than when you’re at the helm of one of Europe’s leading AI companies.
In an op-ed published by Le Monde and republished in English on LinkedIn, Arthur Mensch takes aim at a notion that, he says, “has taken root in Europe”: leaving cutting-edge AI models in the hands of Americans and the Chinese.
He attributes it to “serious people, who have their logic”: a frontier model costs hundreds of millions of euros per generation, soon billions. Better, according to them, to replicate the Internet playbook, create value in applications, and focus on industrial data, SMEs, and domain expertise, with open models published by others deemed sufficient for the rest.
The head of Mistral AI sees this as a “fundamental error” that “will seal Europe’s place in the world.”
“Letting AI models to Americans and Chinese would place Europe in an unprecedented dependency in the history of civilizations,” he argues in the headline of his op-ed
“Produce or Suffer”
The Mistral boss sees as a “fundamental error” the idea that Europe should leave frontier models to the United States and China, while concentrating on industrial data, SMEs, and expertise. This choice “will seal Europe’s place in the world as something to lose,” he says.
And he concludes: “Produce or suffer: the choice is ours for the moment.”
Arthur Mensch rejects the Internet analogy, often invoked to justify a European specialization in use cases. “The Internet relied on open protocols that no one owned. Closed AI models, by contrast, have a proprietor,” he writes.
He emphasizes that control of the model determines access conditions across the entire value chain: “Whoever controls the model sets the terms: they change the price, restrict uses, remove a version.”
To illustrate this risk, he mentions the temporary suspension, in June, of global access to the most advanced Anthropic models at the request of the U.S. government. “The decision wasn’t taken in Europe, nor for European reasons. No one asked for our input,” he asserts.
Mistral AI has long defended open models, but Arthur Mensch refuses to treat them as a sufficient solution. “An open model can be downloaded, hosted, audited,” he concedes, before adding: “Relying on others, even on their openness, is still dependency.”
His argument is that Europe does not control the publication schedule, the features, or the licences of future versions unless it actively participates in their development.
Security, Research and Industry
The Mistral chief broadens the issue beyond the economy. For cyber defense and critical infrastructure, he argues that a state must be able to “audit the systems it uses, adapt them to its needs and be certain of having access to them tomorrow.”
He adds that cutting-edge models have become indispensable to scientific research, including AI research itself, and that their concentration in a few companies could create “very significant tipping points.”
Arthur Mensch also rejects the dichotomy between fundamental research and industrial deployment. “For the most demanding uses, mastering the model isn’t a luxury, it’s a necessity,” he writes.
“Break with Pessimism”
Mistral believes Europe still has the skills needed to catch up. Mensch cites Black Forest Labs and ElevenLabs, then presents Mistral as proof that European models can sit among the best open and specialized models.
He announces that the company “is poised to multiply [its] computing capacity by six in the coming year,” while calling for more capital to accelerate.
And he warns against the narrative of a Europe that is already surpassed and “persuading those who are ahead.” “It doesn’t have to be true to be effective: it’s enough that we believe it to give our competitors the edge,” he cautions.
In other words, the race for global AI dominance is underway and every trick is fair game to win, including at the highest political levels.
Mistral’s Position on Regulation
This op-ed aligns with Mistral AI’s usual stance: in favor of regulating AI, but only for high-risk applications and uses.
Regarding the AI Act, Mistral argued against imposing heavy obligations on general-use models, fearing they would penalize European startups against American giants. Mensch later described the final text as “entirely manageable.”
Mistral AI also advocates transparency around training data, provided it does not force disclosure of industrial secrets. This stance has drawn criticism from NGOs and European officials, who argue that Mistral helped, together with Aleph Alpha and the French and German governments, to ease the Act’s initial proposal.
The French-scale-up nevertheless remains in favor of certain rules, such as a mandatory contribution, tied to the revenue of model providers operating in Europe, intended to remunerate creators whose works are used in training. It calls for this contribution to also apply to foreign providers.
