An idea has gained ground in Europe: we should leave frontier AI models to the Americans and the Chinese. It is advocated by serious people, whose reasoning has its logic. A frontier model costs hundreds of millions of euros per generation, and soon billions. Better, they argue, to follow the internet playbook, where value was created at the application layer, and to focus our efforts on what Europe does well: its industrial data, its small and medium-sized businesses, its domain expertise. For everything else, they say, open models released by others should be enough.
This is a fundamental mistake. It will end up costing Europe what remains of its place in the world and lock it into a regime of economic dependence unprecedented in the history of civilizations.
The Cloud and software value has gone elsewhere
Europe followed the internet playbook. We developed the applications and left the platforms to others. Twenty years later, the value has gone elsewhere. European companies pay more than €260 billion every year to American cloud and software providers. Europe has become the continent that consumes what others build.
Whoever controls the model sets the terms
AI is not the internet. The internet ran on open protocols that nobody owned. Closed AI models, by contrast, have an owner. What an application or an agent can do is decided by the model, and whoever controls the model sets the terms: they change the price, restrict uses, withdraw a version. Every decision made upstream is imposed on the entire chain. Those who build on top bear the consequences, with no say in the matter.
The risk is anything but theoretical. Last June, an American lab suspended access to its most advanced models. Access was restored. But the decision was not made in Europe, nor for European reasons. Nobody asked for our opinion.
A vital matter of security and progress
Some will reply that open models solve the problem. That is partly true. An open model can be downloaded, self-hosted, and audited. But without actively contributing to the ecosystem, you cannot choose what gets released, nor when, nor under what license the next version will ship. Dependence on others, even on their openness, is still dependence.
Building frontier AI systems is becoming a vital matter of security. Models are now taking on a central role in cyber defense and in protecting critical infrastructure. A state must be able to audit the systems it uses, adapt them to its needs, and be certain it will still have access to them tomorrow. With a model designed and controlled elsewhere, none of these three conditions is guaranteed over time.
The question is also becoming salient for fundamental research in Europe. Frontier models are now indispensable to scientific progress in every field of research. They are increasingly used exclusively by certain companies, which have every interest in guarding the research advantages they confer without sharing access. Using AI for AI research accelerates progress for us and for our competitors alike, which points to very significant threshold effects.
The trap of pitting research against deployment
Pitting frontier AI research against industrial AI deployment is therefore a trap. Of course, part of Europe's growth will come from AI tailored to industry. But that requires building a European AI industry, not accepting an AI industry in which we would be mere tenants. We see it with our customers: For the most demanding uses, mastering the model is not a luxury; it is a prerequisite. You do not learn to modify and use a model effectively simply by importing it.
What it takes to build frontier AI
Europe has the means to build these systems. Black Forest Labs and ElevenLabs, two European companies, produce the best models in their category. At Mistral, our researchers and engineers are based in Paris, London, Munich, Barcelona and Warsaw, as well as in the United States, where we export our products. With fewer resources, we train general-purpose reasoning models that now rank among the best open models, along with the world's best specialized models in several use categories.
Furthermore, we know exactly how to improve our performance. We are about to increase our compute capacity sixfold over the coming year, building on the confidence we have earned in our methods and operations. We can move faster with more capital, and we will raise it.
Breaking with pessimism
Closing the gap demands first and foremost a break with pessimism. We keep being told the game is over and that Europe has lost. That story suits those who are ahead. It does not need to be true to be effective: all it takes is for us to believe it. At Mistral, this story has been held up against us for three years, and for three years we have been proving it wrong.
If Europe resigns itself, AI will go on being built elsewhere, and we will only have access to second-rate export versions, in exchange for fees running into the trillions of euros. That will wipe out any prospect of growth or power. It could even prevent us from continuing to hold together as a society.
Build or surrender: for now, the choice is still ours.