Mistral raises €3B at €21B, led by Samsung

Mistral has raised €3 billion and says it will keep publishing model weights anyone can download. The French AI company announced the Series D round on September 8, 2026, at a post-money valuation above €21 billion. That matters to developers because the largest European lab just bought itself years of runway without changing its open-weight strategy.

Post-money valuation means what the company is worth counting the new cash. Mistral's previous round, a €1.7 billion Series C, was announced on September 9, 2025. The new round is close to double that, one year later.

Who put in the money

Samsung Electronics led the round. Two co-leads joined it: the Scaleup Europe Fund, managed by EQT, and PSG Equity.

Several existing backers stayed in, including a16z, Nvidia and Salesforce Ventures. New names on the list include Advent, BlackRock and the Grand Duchy of Luxembourg. A national government investing directly is unusual, and it fits the pitch Mistral is making.

France's President Macron described the effort as "building a third way in AI," according to TechCrunch's report. The third way means neither American nor Chinese.

Where the money goes

Mistral says the cash will scale compute capacity, build infrastructure, speed up commercial growth and widen its international footprint.

The concrete number is power. The company plans 1 GW of compute capacity in Europe by 2030. A gigawatt is the electrical draw of a large data-center campus, and it is now the standard way AI labs describe their scale.

It is also building tools that let a customer choose which regions process their queries. Inference is the work a model does when it answers you. Deciding where that happens is the practical core of what Mistral calls sovereignty, and it is a contract requirement for many European public bodies.

The company says it operates in 20 countries, aimed at governments and large corporations. Its Microsoft partnership was significantly expanded in July 2026.

The open-weight part

Open weights mean the trained model file is published, so you can download it and run it on your own machines. It is not the same as open source, because the training data and code usually stay private.

Mistral calls "sovereign, open-weight AI" its core mission and lists four models it released in 2026: Mistral Small 4, Medium 3.5, Devstral 2 and Leanstral 1.5. Alongside those sit several commercial products. They include Forge for enterprise customization, Studio for agentic development and Vibe for long-horizon work. The list also covers OCR 4, a set of physics models and a security line called Shieldstral.

One detail stands out. Mistral says it will host third-party open-weight models, including Chinese ones, on its own infrastructure. We have covered Chinese and IBM labs shipping open-weight reasoning models this year; Mistral is offering to run them for European customers who cannot send data abroad.

What this means for developers

The useful signal here is continuity, not the headline number. A lab that raises €3 billion on an explicitly open-weight pitch has committed to that strategy in front of its new investors. If you build on Mistral Small or Devstral, that is a reason to expect the next version to be downloadable too.

The region-selection tooling is worth watching if you ship to European customers. Data residency is usually handled at the storage layer, and inference has been the awkward gap. Check whether the guarantees are contractual or best-effort before you rely on them.

Treat the 1 GW figure as a plan, not capacity you can rent today. It is a 2030 target, and the sources give no interim milestones. Nothing about your current rate limits changes because of this round.

Finally, note the pattern. DeepSeek was reported nearing a $7.4 billion raise days before this, at a $74 billion valuation. Open-weight labs are now raising sums that used to belong to closed ones, which is the healthiest development for anyone who wants to run models locally.