Every famous name in frontier AI sits in America: OpenAI in San Francisco, Anthropic in the same city, Google in Mountain View. There is one meaningful exception, and it is deliberately European. Mistral AI was founded in Paris in April 2023, and in three years it went from a seed-stage wager to the largest equity fundraise by any European technology company — with its models still downloadable by anyone. Here is how that happened.
Three researchers walk out of DeepMind and Meta
Mistral was started by three French researchers who had seen frontier AI from the inside. Arthur Mensch, the CEO, came from Google DeepMind, where he contributed to the Retro, Flamingo and Chinchilla projects. Guillaume Lample and Timothée Lacroix came from Meta, where Lample led the development of the original LLaMA models. All three had met as students at École Polytechnique.
Their thesis, laid out on Mistral's own about page: innovation was advancing, but the big labs were closing off. Mistral would combine frontier capability with openness, transparency and cost efficiency — a different kind of AI company, headquartered in Europe.
The open-weights bet
Mistral announced itself in September 2023 with Mistral 7B, a small model released under the permissive Apache 2.0 license — distributed, famously, via a plain BitTorrent link. By Mistral's own benchmarks it beat every open model up to 13 billion parameters, including Llama 2 13B, while running at a fraction of the cost of proprietary systems.
Three months later came Mixtral 8x7B, a sparse mixture-of-experts model that helped popularize that architecture and, by Mistral's claims, beat GPT-3.5 and Llama 70B on most benchmarks. The pattern held: Mistral kept releasing open weights (Mistral Small, Devstral, Voxtral for speech, Magistral for reasoning) while reserving its largest proprietary models — Mistral Medium, Mistral Large — for paying customers. As of 2026, the company has more than 1,000 employees and a model catalog spanning text, code, vision, audio and OCR.
From Le Chat to Mistral Vibe
Mistral's consumer assistant, Le Chat, became Mistral Vibe in 2026 — no longer just a chatbot but an agent platform: web search, document and spreadsheet analysis, connectors into email, calendars, GitHub and Jira, voice input, image generation, and a coding agent that can write, test and open changes across a codebase. For European users it is the closest thing to a home-grown ChatGPT — and, fittingly for this company, it can also run Mistral's open models on European infrastructure.
The money: a record every year
The funding trajectory is the clearest measure of how fast belief in Mistral grew:
- June 2023 — €105 million seed round led by Lightspeed, then a record for Europe
- December 2023 — €385 million Series A at a ~$2 billion valuation
- June 2024 — €600 million at a ~$6 billion valuation
- September 2025 — €1.7 billion Series C at €11.7 billion ($13.8 billion), more than doubling the price
- September 2026 — a round Mistral described as the largest equity fundraise ever completed by a European technology company, at a valuation close to double the year before
The Series C had a headline investor that says everything about Europe's AI predicament: ASML, the Dutch chip-equipment monopoly, put in €1.3 billion for an 11 percent stake, alongside Nvidia, a16z, DST Global, Bpifrance and others. Revenue disclosed around that round was in the region of $400 million.
Why a chip company bought an AI lab
ASML's investment was billed as a first-of-its-kind partnership, not just a check: Mistral's models are being integrated into ASML's lithography systems, research and operations. For ASML, it is a way to embed AI in the most delicate manufacturing process on earth. For Europe, it quietly connects its two most strategic technology assets — the company that makes chipmaking machines and the company trying to keep frontier AI European. It also arms Mistral with compute-adjacent leverage at a moment when access to chips decides who gets to train what.
The honest picture
Mistral is Europe's best, not the world's biggest. Its revenue is a rounding error next to OpenAI's, its valuation is a fraction of the American leaders', and the open-weights strategy means its best work is copied as fast as it ships. Its independence is real but not absolute — the cap table includes Samsung, Nvidia, ASML and Silicon Valley venture firms. And the frontier gap from the US labs, measured in funding and compute, is the same gap described in our earlier piece on Europe and the AI race.
Still: three researchers with a torrent link built the only European lab that gets named in the same sentence as the Americans. That is not nothing.
No comments yet