Mistral has become the standard-bearer of European AI. In June 2026, the Paris start-up is discussing a raise of around 3 billion euros at a valuation near 20 billion, stacking up industrial partnerships and building its own data centres. But behind the momentum, two questions deserve a straight answer: can it really win over businesses, and would it survive a US "kill switch"? Here is what a Belgian leader should know, without the storytelling.
Founded in 2023, Mistral moved fast. After an 11.7 billion euro valuation in autumn 2025, with Dutch firm ASML as its largest shareholder, the company is, in June 2026, in talks to raise nearly 3 billion euros at a valuation of about 20 billion. Its recurring revenue is reported to have grown twentyfold in a year, targeting more than a billion dollars of ARR by the end of 2026. Products: Mistral Large 3, Medium 3.5, Small 4 and specialised models. Infrastructure: 830 million dollars of debt in early 2026 for a data centre near Paris, and a second site announced in Sweden.
What makes Mistral distinctive is not its benchmark scores, it is its openness. Most of its models ship as open weights, often under an Apache 2.0 licence: anyone can download, modify and host them. It is not "pure" open source (training data is not published), but it is clearly more open than OpenAI, Google or Anthropic. For a European business, the argument is concrete: run a Mistral model on your own servers, in Europe, without depending on a US cloud. Add EU data hosting, GDPR compliance and compute infrastructure built on the continent.
This is the question raised by the Fable 5 and Mythos 5 case, where a US directive cut off access to models for all customers, Europe included. The honest answer has two parts.
Once a model is downloaded and hosted by you, no provider and no state can "switch it off" remotely. This is Mistral's strongest argument. CEO Arthur Mensch summed it up after the Anthropic episode: Mistral "exists outside of state control".
For the closed services (the Le Chat assistant, the La Plateforme API), the answer is more nuanced: these are services, so subject to the same availability risks as any provider. And a deeper hardware dependency remains: like almost everyone, Mistral trains on US-made Nvidia chips. The software sovereignty of open weights does not remove hardware dependence. That is exactly the reading of the European sovereignty package (CADA): location is only the first level.
On raw performance, Mistral is not in the lead. On independent 2026 rankings, its models sit below the open-weights front-runners, dominated by Chinese labs (DeepSeek, Qwen, GLM), and behind the best US models. Mistral also remains tiny next to the giants: Anthropic and OpenAI are valued in the hundreds of billions of dollars, against around twenty billion euros for Mistral.
Finally, the headline industrial partnerships (Airbus, BMW, EDF, CMA CGM) are real and recent, but several observers note that the concrete contracts are still limited. The momentum is undeniable; dominance is not.
A small but credible challenger. Mistral hired its first US general manager in May 2026 and is expanding across the Atlantic. Its target: customers who want powerful AI without being locked into a 100% US stack. "They do not want to be forced into a stack, they want choices."
On open weights, Chinese models (DeepSeek, Qwen, Kimi, GLM) lead the benchmarks. Mistral does not win the raw-performance race, but plays another card: European jurisdiction, GDPR, Apache 2.0 licensing and trust, where a Chinese model raises governance questions for a European buyer.
Mistral is not "the best model". It is the best trade-off when sovereignty, confidentiality and self-hosting matter more than the last benchmark point.
Will Mistral win out? Probably not as the performance leader, but yes as the reference sovereignty choice in Europe, in regulated and industrial niches. For a Belgian leader, it is not "Mistral or nothing": it is Mistral as sovereignty insurance within a portfolio of models.
Molderez Consult SRL evaluates your use cases and builds a multi-model architecture that balances performance, cost, compliance and sovereignty.
Choose my AI strategyTransparence : cet article a été rédigé avec l'aide de l'intelligence artificielle, puis relu par Molderez Consult SRL. Information générale, vérifiée le 19 juin 2026 ; chiffres de levée et de valorisation rapportés par la presse et susceptibles d'évoluer.