A step toward European AI independence, built on data centers that still run on US-designed Nvidia chips.

Microsoft has agreed to spend billions of dollars building out Mistral's computing capacity across Europe and to carry the French startup's models further across its own cloud and software, giving heavily regulated industries an alternative to AI infrastructure that sits under US control.

The two firms unveiled the arrangement Tuesday. Under it, Microsoft Azure customers will be able to build software on Mistral's data centers in France, adding to Microsoft's European capacity; Mistral's Medium 3.5 model has been added to Copilot Studio; and companies running self-contained data centers that connect to Microsoft through Azure Local can operate Mistral's openly licensed systems, which let users treat the AI as fully their own.

Microsoft President Brad Smith said the multibillion-dollar sum pays for compute and capacity, not a fresh equity position in Mistral.

Europe's drive to build homegrown AI stretches back years, but it took on new urgency after Washington moved in June 2026 to halt overseas access to two of Anthropic's most advanced models, part of a wider push to lessen dependence on US technology and give countries more say over their own economies. Speaking to reporters together, Smith and Mistral CEO Arthur Mensch framed the tie-up as securing technological independence for Europe while keeping US software and security tools within reach; Mensch said it showed the two companies working jointly to close Europe's infrastructure gap.

Yet fully separating Europe from American technology remains difficult: Mistral's own data-center expansion depends on Nvidia processors designed in the US, and both Nvidia and Microsoft already hold stakes in the startup.