The office was called "le frigo." The fridge. Not because it was cool — because it had no heating.
That is where Mistral AI started in April 2023: a forty-square-metre room in central Paris, four people, and a conviction that Europe needed its own AI company. Arthur Mensch had just left DeepMind. Guillaume Lample and Timothée Lacroix had just left Meta. They had worked on some of the decade's most consequential models — Lample was one of the architects of Meta's Llama. They decided to build the next ones independently, in France, and on different terms than the American labs were offering.
On September 8, 2026, Mistral announced a €3 billion Series D at a post-money valuation of €21 billion — approximately $24 billion. The round was led by Samsung Electronics, the EU-backed Scaleup Europe Fund managed by EQT, and existing investor PSG Equity. It is the largest equity funding round ever completed by a privately owned European technology company.
Three years from le frigo to Europe's highest-valued private AI company. The speed is worth sitting with for a moment.
Who Built This and Where They Came From
Arthur Mensch is 34 years old. He grew up near Paris, studied at École Polytechnique — France's most prestigious technical institution — took a PhD in machine learning and brain imaging at Inria, then spent time at NYU studying reinforcement learning under Joan Bruna, before joining DeepMind Paris in 2020.
For three years at DeepMind he worked on large language models, multimodal systems, and architectures combining generation with information retrieval. He was inside the room where some of the foundational work on modern AI was happening. He watched ChatGPT launch in November 2022 and understood what it meant — not just technically, but strategically. Europe had no equivalent. No company building at that level, with that kind of compute and research talent, outside of American or Chinese control.
He left DeepMind in early 2023. Guillaume Lample and Timothée Lacroix were leaving Meta at the same time. Lample had co-created Llama, Meta's open-source model that had become one of the most widely used foundations for AI development globally. They had worked together before. They agreed on the same thing: the moment was now, and Europe needed to be in it.
Headline VC backed them four weeks after the company was founded. The founding team had no product. They had a conviction and a track record. That was enough.
The Open Source Bet
Mistral made a choice early that defined everything that followed: open weights. Where OpenAI and Anthropic build proprietary closed models that users access through APIs, Mistral releases its models openly — allowing companies to download the weights, customise the models on their own infrastructure, and run them without sending data to Mistral's servers.
This is not just a technical decision. It is a sovereignty decision. For European governments, financial institutions, healthcare systems, and enterprise clients who cannot or will not send sensitive data to American cloud providers, an open-weight model that runs on your own hardware is a qualitatively different proposition. You are not dependent on Mistral's API remaining available, on Mistral's pricing not changing, or on Mistral's data policies being compatible with your regulatory environment.
The sovereignty argument became dramatically more visible in June 2026, when the US government limited foreign access to two advanced Anthropic models. For European AI buyers who had been abstractly concerned about dependency on American AI infrastructure, that decision made the concern concrete. Mistral's CFO Johan Bergqvist was direct in his response: "The fact that the EU or Europe have to have their own kind of AI provider in the game is important. We have seen in the past that there is politics involved in the access of these solutions."
Mistral now has more than 125 enterprise customers and is on track to reach $1 billion in annual recurring revenue by the end of 2026. The customer base is increasingly international — growth is strongest in Asia and North America, not just Europe.
The Competitive Position
At €21 billion, Mistral is Europe's second-highest-valued private technology company and its most valuable AI business. It is also, by any honest measure, significantly smaller than its primary American competitors. Anthropic is valued at $965 billion. OpenAI at $852 billion. Mistral at $24 billion is approximately 40 times smaller than Anthropic.
The comparison is not entirely fair — Mistral is three years old and the American labs have had longer runways, larger compute budgets, and deeper relationships with US cloud providers. But the gap is real and the question it raises is honest: can Mistral compete at the frontier of AI research against organisations with 40 times the capital?
Mensch's answer is that frontier competition is not the only game. The next wave of AI adoption will not be driven by the largest model alone. It will be driven by whoever makes intelligence easiest, cheapest, and safest to deploy — for the specific regulatory and operational contexts that European and Asian enterprise customers operate in. Mistral's position is not to out-compute OpenAI. It is to build the distribution layer for AI in markets where American models carry political or regulatory risk.
The Chinese competition complicates this. Models from DeepSeek, Qwen, and others have demonstrated frontier-level capability at significantly lower cost than American alternatives. Mensch acknowledges the technical quality directly — but notes that Chinese labs are not operating with enterprise customers outside China, and that European companies cannot rely on long-term support, upgrade paths, or export stability from Chinese AI providers. "The volatility in this space is actually quite extreme," he said.
Mistral's infrastructure ambition is part of the answer. The €3 billion will go toward building out Mistral's own data centres — the company aims to double its owned compute capacity in the next five years. Microsoft has committed to spending billions on Mistral's European AI infrastructure in a deal announced in July 2026. Own the compute, own the infrastructure, own the relationship: it is the same thesis that made AWS a more durable business than any individual application running on top of it.
ElevenLabs and the Valuation Context
Mistral becomes Europe's most valuable private AI company today — but the title may be temporary. ElevenLabs, the AI voice technology company, is reportedly in discussions around a tender offer valuation approaching $22 billion. If that transaction completes at or above current estimates, Europe will have two AI companies in the same valuation territory within months of each other.
That matters. It suggests that the European AI ecosystem is not producing one exceptional outlier — it is producing a cohort. Mistral from Paris, ElevenLabs from London, a cluster of infrastructure and application companies building on open models. The ecosystem is beginning to look like one.
For context on how AI companies are moving from private to public markets, the FreeMalta IPO Watch tracks the companies navigating that transition — Mistral's CFO described an IPO as "always of course an optionality for us going forward" while declining to give a timeline. The more substantive story for now is the Series D and what it enables.
From Le Frigo to Here
The founding story of Mistral is one of the cleaner ones in recent AI history. Three researchers who had worked on the most important models of the decade, who understood exactly what the moment required, who made a decision to build independently rather than stay inside the labs, and who started in a cold room in Paris because that is what was available.
The Mistral company story in The Garage covers the founding history in detail — the DeepMind years, the Meta research, the Llama architecture, the decision to go independent. The trajectory from April 2023 to September 2026 is three and a half years. Le frigo to €21 billion.
It is not the American story. It was never supposed to be. That is precisely the point.
FreeMalta covers company founding stories and AI developments at freemalta.com. This article does not constitute investment advice.
