AI-Economy

Microsoft invests billions in Mistral's expansion in Europe

3 min read
Microsoft Vice President Brad Smith and Mistral CEO Arthur Mensch seal the expanded AI partnership in front of a European data center Image generated with GPT Image 2
Microsoft Vice President Brad Smith and Mistral CEO Arthur Mensch seal the expanded AI partnership in front of a European data center

TL;DR Too Long; Didn’t read

Microsoft and French AI provider Mistral expanded their partnership on July 21, 2026, with a billion-dollar infrastructure deal. Mistral is expanding European data centers with thousands of Nvidia Vera Rubin chips, and Microsoft is integrating the Medium 3.5 and OCR 4 models into Foundry and Copilot Studio. Targets include banks, clinics, and industrial firms with high data protection needs.

Key takeaways

  • Mistral is expanding European data centers with thousands of Nvidia Vera Rubin chips, aiming for one gigawatt of capacity by 2030.
  • Medium 3.5 and OCR 4 are now running in Microsoft Foundry, with Medium 3.5 also in Copilot Studio.
  • The target audience includes banks, clinics, and industrial companies with strict data protection and export rules.
  • Prices for inference and a timeline for the new chip capacity have not yet been published.
  • Mistral is reportedly negotiating a funding round of around three billion euros in parallel.
  • Microsoft had already integrated Mistral Large as its first model in Azure back in February 2024.

Microsoft and the French AI provider Mistral significantly expanded their strategic partnership on July 21, 2026. A multibillion-dollar agreement gives Microsoft access to Mistral’s European data center capacity, which is growing with thousands of Nvidia Vera Rubin chips. At the same time, Mistral’s current models, Medium 3.5 and OCR 4, are now available in Microsoft’s cloud services.

Vera Rubin chips power Mistral’s European data centers

The core of the agreement is the expansion of Mistral’s data centers in Europe with thousands of Nvidia Vera Rubin graphics chips. According to Unite.AI, capacity should rise to one gigawatt by 2030 – roughly the output of a large power plant. Microsoft is renting into this infrastructure, letting it serve part of its cloud demand from European rather than its own US data centers.

Both companies market the offering as a “sovereign cloud”: a combination of Mistral’s European models and Microsoft’s security and compliance infrastructure meant to let customers keep control over their data. The target audience explicitly includes banks, industrial companies, and hospitals that are unable or unwilling to run US models in US data centers for data protection or export control reasons. According to Unite.AI, the French military is already among Mistral’s customers – a sign of demand from the defense sector.

The deal fits into Microsoft’s own multibillion-dollar investments in European data centers in recent years, a response to growing demands for digital sovereignty in the EU. For Mistral, it provides access to Microsoft’s global sales channels and cloud customer base without having to finance its own data centers at the same pace alone.

Foundry and Copilot Studio gain new Mistral models

Mistral’s current models, Medium 3.5 and OCR 4 – a model specialized in automatically reading text from documents – are now available in Microsoft Foundry, with Medium 3.5 also available in Copilot Studio. Companies can choose between three deployment modes: fully in the cloud, cloud-connected with local data storage, or fully disconnected from the internet – the latter relevant for government agencies or defense contractors. Copilot Studio additionally lets companies build their own AI agents for internal workflows such as contract or invoice review on top of the Mistral models.

“Europe should have access to the world’s most capable AI without compromising control,” Microsoft Vice Chair and President Brad Smith said in the official announcement. Mistral co-founder and CEO Arthur Mensch stresses that the company wants to put frontier AI in the hands of every organization. The partnership is not new: Microsoft first integrated Mistral Large into Azure back in February 2024, then purely as model distribution without an infrastructure investment. The current step expands that relationship from a licensing deal into a joint infrastructure and sovereignty strategy.

Analysts want to see real customer contracts

It remains unclear what access to Mistral’s models through Microsoft will cost – neither company comments on inference pricing. The exact timeline for the additional Vera Rubin chips is not yet set either. Mistral is reportedly negotiating a separate funding round of around three billion euros at a twenty-billion-euro valuation – independently unverified.

For Microsoft, the deal fits its own strategy of reducing dependence on individual AI providers: CEO Satya Nadella had warned that companies pay twice for commercial models – with money and with disclosed expertise. Mistral CEO Mensch himself had previously warned companies about dependence on closed models. That does not contradict the new partnership with cloud giant Microsoft, since Mistral’s models remain selectable through Foundry rather than being exclusively bound to it.

What matters now is whether banks, hospitals, and industrial companies turn the announcement into real production contracts or whether it stays at the pilot stage. The models are already live, Unite.AI notes, but the pledged capital has so far mostly been just that – a pledge. Whether that changes should become clear once the first concrete customer names from regulated industries go public.

Frequently asked questions

What does the partnership mean for companies in Germany and the EU?

Companies can use Mistral's models through Microsoft Foundry or Copilot Studio and choose whether data is processed in the cloud, partially locally, or completely disconnected from the internet. This targets industries with strict data protection or export rules.

What will access to Mistral's models via Microsoft cost?

Neither company has announced inference pricing yet. It is also unclear whether costs will differ from using Mistral directly.

What are Nvidia Vera Rubin chips?

They are Nvidia's newest generation of AI accelerators for training and running large language models, which Mistral is using to expand its European data centers.

Since when have Microsoft and Mistral been working together?

Microsoft first integrated Mistral Large into Azure in February 2024. The new agreement expands that licensing relationship into joint infrastructure and a sovereignty offering for regulated industries.

When will the additional data center capacity be available?

Microsoft and Mistral have not published a concrete timeline. Reports suggest Mistral is targeting around one gigawatt of compute capacity by 2030.

Sources

  1. Microsoft and Mistral expand strategic partnership to give enterprises and regulated industries frontier AI they can control
  2. Microsoft Widens Mistral Deal to Court Regulated AI Buyers

← Back to the blog