Mistral raises €3B in Europe's largest tech round to push sovereign AI
Three years after launch, Mistral has closed the biggest private funding round in European tech history — and it's spending it on data centers it owns, not just models.
Mistral raised €3 billion ($3.5 billion) at a post-money valuation above €21 billion, the largest equity round ever completed by a European technology company. Samsung Electronics led the round, joined by co-leads Scaleup Europe Fund — the EU-backed vehicle managed by EQT — and existing investor PSG Equity. Advent, funds managed by BlackRock, and the Grand Duchy of Luxembourg came in as new investors; Nvidia, ASML, a16z, Bpifrance, DST Global, General Catalyst, Index Ventures, Lightspeed and Salesforce Ventures all participated. CEO Arthur Mensch told CNBC the money goes into building infrastructure, including Mistral's own data centers, with the share of compute Mistral owns itself roughly doubling over the next five years.
The strategy is coherent in a way most "sovereign AI" pitches aren't. Mistral is selling control across four layers — data that stays inside the customer's boundary, models that can be customized, compute that's private and predictable, and production systems that stay auditable — and it's the only major lab actually shipping all four with open weights attached. That's a real differentiator against OpenAI and Anthropic for European enterprises and governments. Mensch also took an unusually direct shot at the alternative: Chinese open models can be hosted on Mistral's infrastructure, he said, but volatile model quality and the risk of future export restrictions make them hard to build a multi-year plan on.
The catch is that "sovereign" is now a crowded label, and €3 billion doesn't buy frontier scale by US or Chinese standards. Mistral's pitch is that institutional buyers won't optimize for the frontier — they'll optimize for the guarantee that a supplier will still be shipping better models in 2029. That's a bet on procurement behavior, not capability. Mensch says new models arrive "very soon" and expects to beat a previously flagged $1 billion annual recurring revenue target; we covered the company's infrastructure push in Mistral rallies ASML and partners behind 1GW European AI compute.
Huawei is investing in Chinese lithography companies and brokering their deals with fabs including SMIC, according to people familiar with the matter. The Financial Times reports Huawei is putting money into domestic equipment and component makers while coordinating suppliers, foundries and integrators to get homegrown DUV lithography tools into real production lines — Bernstein analysts reportedly describe Huawei as the "project manager" of China's DUV effort. Shanghai Yuliangsheng is furthest along: its tools have been tested at fabs including SMIC and Huawei's own lines, with a target of producing 12 units by the end of this year. Initially those machines would make simpler chips; the endgame is advanced AI silicon.
It's still not a clean break from foreign supply. Yuliangsheng's projection lenses reportedly still come from Germany's Zeiss, and high-power light sources remain the key bottleneck — which is exactly why Huawei has been buying into optics and light-source specialists. We covered the demand side of this build-out in DeepSeek plans a 160,000-chip Huawei cluster in Inner Mongolia.
Arm used its platform launch to push two things at once: a semi-custom server blueprint and its own data-center chip. Neoverse CSS N4, codenamed Ranger, lets cloud customers configure 8 to 128 Neoverse N4 cores per die at up to 3.8 GHz on TSMC's N3P node, with up to 256 MB of shared L3, DDR5 or LPDDR6 memory, 128 lanes of PCIe 7 or 6, and CXL 4.0. Arm claims twice the socket performance, 1.25x the performance per watt and 1.75x the memory bandwidth of its N3-based predecessor — vendor reference numbers, not shipping hardware, and Arm hasn't named a customer yet.
The more concrete news is the AGI CPU Arm is actually selling: two dies, up to 136 cores, 272 MB of L3, up to 6 TB of memory per chip, and a claimed sub-100-nanosecond memory latency. Oracle and ByteDance are deploying it, joining Meta, Lenovo, SAP, OpenAI and Cloudflare. Arm says the chip delivers more than twice the performance per rack of the latest x86 systems — its own internal estimate, with no independent benchmarks behind it.
Is a €21 billion Mistral a genuine third pole in AI, or Europe buying insurance? Tell us in the comments.
