Can Mistral Challenge Big Tech by Doing AI Differently?

The report opens by casting the AI race as a contest that has so far been dominated by a handful of U.S. tech giants. Westin, the host, introduces the story as one about rooting for an underdog.

Europe has not seen the level of private or government capital committed to frontier AI models that the U.S. has. France-based Mistral is presented as the European underdog that is trying to redefine the competition in a way it believes can put it at the front of the pack.

When ChatGPT reached consumers in November 2022, annual investment in AI was $112 billion. By the end of 2025, that figure had risen more than 200 percent to $345 billion, and this year it is on track to exceed even those sums.

The report also points to borrowing by hyperscalers in the investment-grade market. The speaker expected such borrowing to be probably around $300 billion through 2026, and said issuance from some hyperscalers would rise another 15 percent next year.

Westin says the U.S. hyperscalers are largely competing against Chinese firms that are backed by the resources of their entire government. Those firms are pursuing lower-cost alternatives that require less compute.

Europe, by contrast, has been largely absent from this titanic struggle so far, as it has not seen comparable levels of private or government capital committed to frontier AI models.

Arthur Mensch, co-founder of Mistral, describes the AI landscape as an opposition between closed model providers and an open version of AI. In his account, closed providers promote a centralized version of AI and try to export it to the entire world, while open AI involves multiple actors contributing to an open ecosystem.

In that open model, enterprises can use the ecosystem and customize their own systems on top of an open infrastructure. Westin notes that Mensch believes Mistral has a built-in advantage over the closed-weight approach of U.S. hyperscalers, and Mensch says Mistral is now a leader in open-source models for sovereign deployment.

Mensch lists three advantages for enterprises. The first is business continuity: because the model is decoupled from the infrastructure layer, a company can bring the model into the infrastructure it wants, use multiple providers, and keep operating even if the model-providing company runs into trouble.

The second concerns sensitive data. Many AI deployments involve data that companies do not want to share with a centralized AI provider. Mensch says many Mistral customers come to the company because it deploys its models close to their data in a way that means Mistral does not see that data.

The third is cost. As AI gets more expensive, the business case for deployment has to account for it, and Mensch argues that paying a centralized provider for the model layer costs much more than running open-source models on one's own infrastructure. Asked about downsides, he said there is effectively only upside today, the only drawback being that open-source models are a bit harder to use because they came later.

Westin reports that Mistral recently released its latest open-weight model, Mistral Large 4, nicknamed "Le Chonk," and that the company raised 3 billion euros to scale its innovation.

Mensch says Mistral had been under capital and compute constraints for the last two years but still produced a very strong model despite not controlling much compute. He said this was about to change because the company had made the right investments, allowing it to scale training and inference for its customers. He described the company's approach as more frugal and said its technology was produced with much less capital than some U.S. counterparts.

Westin asks how much of Mistral's different approach is required by the limitations inherent in being a European tech company. Roxanne Varza, director of the European startup incubator Station F, says Mistral should not be seen as the exception: it is probably one of the first European companies to have reached this scale and size. Her point is that Europe's ecosystem is somewhat less mature, not that Mistral's frugal approach is simply a product of those limits.

Varza says Europe's ecosystem is somewhat less mature but is catching up and picking up pace.

Varza says Europe is served well at the early stage, where funding gaps have mostly been filled. The remaining gaps are mainly at the late stage, where companies struggle to scale, find growth capital and exit.

She also describes Europe's fragmentation as a long-standing obstacle. Each country tends to have a local dominant player that is big in its own market but not at the scale seen in the U.S. Acquisition culture is starting to pick up, but she says many acquirers still come from abroad, mainly the U.S., and possibly now Asia, with fewer in Europe.

Jacob Kirkegaard, a senior fellow at Bruegel, agrees that Europe's lack of a single deep capital market makes competing in AI a challenge, but he also sees a possible opportunity. He argues that the still not fully integrated European economy is a hindrance both to creating AI models and to seamless, harmonized diffusion of AI.

He expects significantly faster AI adoption in mostly northern European countries such as Scandinavia, the Baltic countries and the Netherlands, which have relatively flexible labor markets, while others will take longer. In his view, this unharmonized economy benefits some countries while setting others back.

Westin says Kirkegaard sees Mistral's focus on providing AI services, rather than frontier models, as potentially critical for Europe's expected productivity gains. Kirkegaard distinguishes AI producers, such as Anthropic and OpenAI, from AI users or adopters, and calls the latter the key to understanding AI as a general-purpose technology.

Economically, he says it matters little who produces the models as long as one has access to them. He notes that national security is the counterargument: Europe relying on U.S. or Chinese open-source models could face a situation in which Washington or Beijing bans exports of those models to foreign customers.

Kirkegaard describes Mistral as an outlier, arguably the most advanced and largest pure AI company in Europe, and says it was until recently close to, if not at, the frontier of model development. In his reading of its recent strategic shift, Mistral has decided not to keep investing in closed frontier models, and is now more focused on AI-related services and open-source models than on being a frontier lab like Anthropic or OpenAI.

Westin describes Mistral's latest funding round as reportedly 3 billion euros at a valuation of around 21 billion euros, maybe 24 billion U.S. dollars. That is very large by most standards but orders of magnitude smaller than the hyperscalers, and Westin asks whether Mistral can compete with the big players.

Mensch says Mistral plays in a different category. Customers sometimes use hyperscalers and closed models for generic workloads such as chatbots, but when they deploy AI for building new airplanes, better machines or finding more sources of energy, they typically want a solution that gives them the control and sovereignty they need over their deployment.