Summary: Mistral AI announced a definitive agreement to acquire Paris-based Koyeb on February 17, 2026, its first acquisition, without disclosing terms. Koyeb provides serverless GPUs, autoscaling, scale-to-zero, and isolated sandboxes, and its technology is slated to become a core component of Mistral Compute. Existing customers keep their platform while new signups start at Pro or above, reflecting an enterprise shift.
A definitive agreement to acquire Koyeb, a startup based in Paris, was announced by Mistral AI on February 17, 2026. The deal is the first acquisition in the French AI company’s history; the financial terms stayed confidential, and closing conditions still have to be satisfied before it completes. TechCrunch and Reuters both carried the story that day, citing Mistral’s last known valuation of roughly $13.8 billion and reading the purchase as support for Mistral Compute, the homegrown AI cloud it has been building.
Read on its own, the announcement is plain; its placement is the interesting part. US labs spent that stretch of the calendar publicizing giant rounds — the very same week saw Anthropic close a $30 billion Series G at a $380 billion valuation — yet Mistral pointed its cash at a 16-person infrastructure team. Two distinct expansion logics sit side by side here: one spends on compute itself, the other on people who know how to turn compute into a product.
The Transaction: A First for Mistral
For current customers, nothing breaks: the platform stays up, plans carry over unchanged, billing looks the same as before, and user data is not moving to Mistral “immediately.” Where the change lands is on new signups — the free Starter tier has shut its doors, accounts now begin at the Pro plan or above, and the company’s center of gravity moves to the enterprise.
Koyeb was started by three engineers out of Scaleway — Yann Léger, Edouard Bonlieu, and Bastien Chatelard — and its lifetime fundraising adds up to a modest $8.6 million: a $1.6 million pre-seed in 2020, followed by a $7 million seed in 2023 with Paris VC Serena in the lead. Serena principal Floriane de Maupeou, on the exit: the deal helps build “the foundations of sovereign AI infrastructure in Europe.” Whether more acquisitions are coming is something Mistral left unsaid; the price and the structure both stayed out of TechCrunch’s reach, with only the closing conditions confirmed.
A company last valued at $13.8 billion choosing to buy a seed-stage infrastructure team, rather than lease one more data center, is making a deliberate statement about where it believes the scarcity sits.
The Technology Koyeb Brings
What Koyeb ran was a serverless compute platform: serverless GPUs, niche accelerators such as Tenstorrent parts, and CPU workloads, offering sub-second scale-to-zero, autoscaling, and isolated sandboxes designed with AI agents in mind. Tens of thousands of applications lived on it, spread across bare-metal servers in ten locations around the globe.
The asset Mistral is really after is the operational layer — scheduling, isolation, the ability to scale on demand. Package a GPU cluster as a cloud service, and that turns out to be precisely the hardest software to write.
Where Koyeb Lands in Mistral Compute
Mistral Compute went live in June 2025, pitched as infrastructure much like the stack Mistral runs for itself. Koyeb’s technology is slated to become a core component of the service, and the intended uses are specific: getting models onto clients’ on-premises hardware, lifting GPU utilization, and scaling AI inference. For its part, Koyeb’s blog commits to redoubling its work on inference, sandboxes, and serverless capabilities for MCP servers.
Hardware, meanwhile, had been sorted out in advance. Per Koyeb’s announcement, the starting point is Mistral’s deployment of 40 MW of data center capacity and 18,000 NVIDIA Blackwell GPUs; days earlier, Mistral had committed $1.4 billion to Swedish data centers, pointing to customer demand for alternatives to US infrastructure. In CTO Timothée Lacroix’s words: “Koyeb’s product and expertise will accelerate our development on the Compute front.” The engineering logic is plain: Mistral holds raw capacity and lacks the software that converts capacity into a billable service with isolation guarantees.
Sovereignty Talk Meets the Balance Sheet
As for the geopolitics, little needs spelling out. At Stockholm’s Techarena, CEO Arthur Mensch made the pitch in blunt terms: a company “headquartered in Europe, that is doing frontier research in Europe.” Mistral is piecing together a full vertical stack, models at one end and machines at the other, for the European enterprises and governments unwilling to place data and workloads on US clouds.
The commercial figures carry their weight too: annual recurring revenue has passed $400 million by TechCrunch’s account, and hiring for infrastructure roles continues. Developers can draw two lessons. The first: European sovereign AI has moved past the slide deck — it now comes with data centers, a scheduling layer, and enterprise customers paying for it. The second: Koyeb’s absorption leaves one fewer independent serverless option on the market; teams running agents in Koyeb sandboxes should start putting numbers on migration and supplier-concentration risk.