This first ran on deanlee.info.
On September 8, Mistral said it had raised €3 billion in a Series D at a post-money valuation of more than €21 billion. Reuters put that at about $24 billion at the day's euro print. The company called it the largest equity round ever completed by a privately owned European technology firm, three years after launch. PitchBook, using its own tape, called it Europe's largest VC round and put the prior mark at €11.7 billion a year ago.
Samsung Electronics led. Scaleup Europe Fund, managed by EQT and backed by the European Commission, co-led with existing holder PSG Equity. Reuters described the three as joint leads. Mistral's own note adds Advent, BlackRock accounts, and the Grand Duchy of Luxembourg as new money. The existing list that re-upped is long: a16z, ASML, Belfius, BNP Paribas CIB, Bpifrance, Carmignac, DST Global, Eurazeo, General Catalyst, Headline, Hillspire, Index Ventures, Korelya, Lightspeed, Nvidia, Phoenix Court's Solar fund, and Salesforce Ventures. PitchBook had already flagged ASML as the Series C lead on a €1.7 billion round.
I read this as an access round first. The steelman is sitting in the CFO interview. Johan Bergqvist told Reuters that Europe needs its own AI provider because access to models has become political. He pointed, without naming the lab in that sentence, at the same episode PitchBook spells out. Earlier this year the US barred foreign nationals from two of Anthropic's most advanced models, then lifted the controls at the end of June. A ministry that cannot download Claude for a week does not care that Anthropic is valued at $965 billion. It cares that the weights were someone else's to switch off.
Mistral sells the other side of that switch. The company develops open-weight models that, Bergqvist said, more than 125 customers can download and customise on their own servers. The official post names Airbus, ASML, and HSBC. PitchBook adds Luxembourg's government and France's Ministry of Armed Forces. Open weights plus on-prem inference is the product that turns a sovereignty speech into a procurement line. Samsung on the cap table is the industrial version of the same trade. A chip-and-device company that already lives with export controls would rather own a piece of a lab whose weights it can host than rent a closed API that a foreign ministry can fence.
The money still has to buy compute. Mistral said the round funds frontier research, training capacity, infrastructure, and commercial expansion across 20 countries. PitchBook says the company is building data centres in France and Sweden and renting AI capacity out of them. That is a full-stack bet. Models, racks, and a services layer that puts forward-deployed engineers inside the customer's environment. In May it rebranded the Le Chat chatbot as Vibe, an agent platform, and bought physics startup Emmi AI at a valuation of up to €330 million. Weights on a public hub would not need forward-deployed engineers or a physics shop at that price. Those buys only make sense if Mistral intends to sit inside the plant.
Two facts sit next to that stack. Microsoft agreed in July to spend billions of dollars on Mistral's computing infrastructure in Europe. Bergqvist told Reuters that Microsoft did not participate in this equity round. And in August, PitchBook notes, Mistral said it would host external models, with China's Z.ai as the first. The same piece reminds you that the French and Swedish racks still run on Nvidia silicon, and that Nvidia is also a shareholder. The sovereignty pitch is about where the weights run and who can cut the login. The transistors still come from Santa Clara.
The people writing the €3 billion are easier to name than the lab that wins the next frontier race. Anthropic's last printed valuation in the Reuters copy is $965 billion. OpenAI's is $852 billion. Both are aiming at listings this year. Mistral at €21 billion is Europe's second-highest-valued private tech group, per Bergqvist, and still roughly a fortieth of Anthropic. The Series D leaves that gap in place. What it funds is a different bid. Stay good enough on open weights, keep the model inside the customer's wall, and sell that as insurance against the next access scare.
Revenue is the number that has to catch the mark. Bergqvist said Mistral is on track for $1 billion of annual recurring revenue by year-end, with growth in Asia and North America as well as Europe. I have the CFO quote. I do not have a backlog file, a net-retention print, or a split between hosted API, self-hosted licences, and the new compute rental. A $24 billion post-money on a path to $1 billion ARR is a growth multiple that only works if the sovereignty premium is real and sticky. If buyers treat Mistral as a cheaper open-weight substitute when Claude is available, the premium compresses. If they treat it as the model they are allowed to run in a defence ministry, it does not.
IPO talk is cheap in this tape. Bergqvist called a listing "optionality" and said there are no ongoing discussions. That is the honest line for a company that just took €3 billion from Samsung, an EU vehicle, and a pile of existing funds. Those holders can wait. The cluster cannot. Training runs and the French and Swedish halls get paid before the ARR slide does.
I keep a wide prior on the category. European governments have stood up Scaleup Europe Fund, targeting €5 billion, and the UK has a Sovereign AI Fund. Airbus and a defence ministry do not usually join a lab's story as decoration. I keep a tighter prior on this specific multiple. The round is large because the US already demonstrated that frontier access can be withdrawn. Samsung and the Commission-backed fund paid for a lab that can be hosted. Nvidia still invoices the GPUs. Z.ai still sits on the same inference board. The distribution I want is how much of that $1 billion run-rate is a customer who would leave if the Anthropic fence came down tomorrow, and how much is a customer who will still need a model they can keep when the next fence goes up.
Top comments (0)