Microsoft and Mistral expand European AI infrastructure pact

Microsoft and Mistral have expanded their strategic partnership through a multi-billion-dollar agreement to build AI infrastructure in Europe, with Mistral adding thousands of Nvidia Vera Rubin GPUs and Microsoft drawing on that capacity for its cloud and AI services. The companies did not disclose a precise contract value, deployment timetable, or the number of GPUs beyond “thousands,” but the deal adds more compute behind Europe’s push for AI systems that can be run with tighter data control.

Microsoft said Mistral’s Medium 3.5 and OCR 4 models are now available in Microsoft Foundry, while Medium 3.5 is also available through Copilot Studio. The companies are positioning the deployment options around enterprises and regulated sectors, including finance, healthcare and manufacturing.

The technical pitch is control over where models run. Microsoft said customers can use Mistral models through Azure Local in the cloud, in their own environments, or offline. Microsoft President Brad Smith framed the partnership as a way for Europe to use powerful AI while retaining control of data, according to Microsoft.

That framing fits Mistral’s broader positioning against fully closed AI systems. Mistral CEO Arthur Mensch has argued that companies can expose trade secrets when they rely on proprietary AI models that give outside labs insight into business processes. In Microsoft’s announcement, Mensch described the partnership as a route for Mistral to reach companies and public-sector institutions globally.

Mistral scales compute and distribution

For Mistral, the Microsoft tie-up supplies distribution and enterprise packaging at a point when the Paris-based company is trying to turn its political and technical momentum into durable revenue. The company was founded in 2023 and has said its revenue grew twentyfold to a run rate above $400 million. The company has also committed 1.2 billion euros to data centers in Sweden.

Bloomberg has reported that Mistral is in talks to raise about 3 billion euros at a valuation of roughly 20 billion euros. If completed, that round would give Mistral more capital to fund the infrastructure-heavy side of frontier AI, where access to GPUs remains one of the main constraints on model development and deployment.

The Microsoft relationship also puts Mistral models deeper into tools that enterprise buyers already use. Foundry gives developers and enterprises another route to deploy the models, while Copilot Studio availability puts Medium 3.5 closer to customers building task-specific agents and internal workflows.

Mistral has also been broadening its application layer. The company recently renamed its chatbot Le Chat as Vibe and recast it as an AI productivity product with Work Mode and Code Mode. That move suggests Mistral is trying to compete on both infrastructure and end-user software, while still using openness and data-control claims as a wedge with European buyers.

Microsoft gets another European AI partner and more regional capacity for its own services. Mistral gets cloud distribution, enterprise access and a larger hardware footprint. The unanswered questions are the same ones attached to most AI infrastructure announcements: how much usable capacity comes online, when customers can access it, and whether demand from regulated buyers matches the scale of the buildout.

This story draws on original reporting from The Decoder.