How much is Microsoft investing in the Mistral partnership?

Microsoft and Mistral described the deal as “multibillion-dollar” but did not disclose an exact figure, nor how much new GPU capacity it adds to Mistral’s existing infrastructure.

Can companies run these AI models without an internet connection?

Yes. Alongside cloud and hybrid (“Azure Local”) options, Microsoft and Mistral offer a fully disconnected deployment that runs entirely on a customer’s own servers with no live link to Microsoft.

Which Mistral models are now available through Microsoft?

Mistral Medium 3.5 and the OCR 4 document-processing model are integrated into Microsoft Foundry, with Medium 3.5 also available in Copilot Studio.

Why does AI "sovereignty" matter to European buyers?

Regulated buyers such as banks, hospitals and government agencies want to avoid depending on AI infrastructure controlled from outside their jurisdiction — a concern sharpened by earlier US export-control disruptions that pushed Ukraine toward self-hosted models.

Microsoft and Mistral Expand AI Deal for European Sovereignty

Microsoft and Mistral AI announced an expanded partnership on July 21, pairing a multibillion-dollar Microsoft investment with a buildout of Mistral’s data centers in Europe, according to a joint announcement from the two companies. The deal targets banks, hospitals, manufacturers and other regulated industries that want frontier AI without handing control of their data to an outside cloud provider.

What’s changing

Mistral is expanding its infrastructure with thousands of Nvidia Vera Rubin GPUs, and Microsoft will draw on that European capacity to serve its own Azure customers. Two Mistral models — Medium 3.5 and the OCR 4 document-processing model — are now built into Microsoft Foundry, the company’s app-development platform, with Medium 3.5 also available through Copilot Studio.

Enterprises get three ways to run the models: fully in Microsoft’s cloud, on Azure Local hardware that stays connected to Azure, or fully disconnected — running entirely on a customer’s own servers with no live link back to Microsoft. That offline option is aimed at buyers who need to keep sensitive workloads off the public internet for regulatory or security reasons. Neither company disclosed the size of the investment or how much new compute it adds to Mistral’s existing footprint.

Why it’s happening now

“Europe should have access to frontier AI without compromising control over data, operations or digital future,” Microsoft vice chair and president Brad Smith said in the announcement. Mistral chief executive Arthur Mensch said the aim is “putting frontier AI in every organization’s hands while keeping them in control.” Nvidia’s Ian Buck said the Vera Rubin systems give customers “the computing foundation for next-generation AI.”

Data control has become a recurring theme in how Microsoft and rival cloud providers sell AI in Europe, where regulators and buyers alike are wary of depending on infrastructure outside their jurisdiction. Ukraine shifted toward self-hosted models after US export controls disrupted access to American AI tools, part of a broader push toward sovereign AI that this deal is explicitly designed to answer. Mistral, for its part, has built its business around giving European customers an alternative to US-only AI stacks.

Microsoft and Mistral said they will also fund joint go-to-market efforts, proof-of-concept credits and workshops to move regulated customers onto the new setup.

What’s changing

Why it’s happening now

Frequently asked questions