PARIS — Microsoft and French artificial intelligence company Mistral are expanding their partnership through a multibillion-dollar agreement focused on European computing capacity and AI deployments that customers can operate under tighter controls.
The companies said Microsoft will purchase capacity from Mistral’s expanding graphics-processor infrastructure in Europe and distribute Mistral models through Microsoft’s enterprise AI products. They did not disclose the total value or term of the agreement.
The partnership is aimed in part at companies and public institutions in regulated industries that must control where data is processed, who operates the underlying systems and whether workloads can continue without a connection to a public cloud.
Microsoft and Mistral said customers will be able to deploy the same models and development tools in Microsoft’s public cloud, in customer-controlled systems connected to the cloud or in fully disconnected environments.
Models move into Microsoft’s enterprise products
Mistral Medium 3.5 and OCR 4 are now available through Microsoft Foundry, the company’s platform for building and managing AI applications. Medium 3.5 is also available in Copilot Studio, which organizations use to build assistants and automated workflows.
The companies said developers can use Microsoft Foundry and Foundry Local to build applications with common tools and then choose where those applications run. The approach is intended to reduce the need to redesign software for separate cloud and on-premises environments.
Potential users include banks, manufacturers, health care organizations and government agencies that face rules governing data residency, operational resilience or access to sensitive information.
The arrangement also adds computing capacity for Microsoft in Europe. Reuters reported that Microsoft agreed to spend billions of dollars on Mistral’s infrastructure while expanding distribution of the French company’s technology.
What sovereign AI means in practice
“Sovereign AI” has become a common industry term, but it does not describe a single technical standard. Companies generally use it to refer to combinations of local data storage, control over infrastructure, model choice, regulatory jurisdiction and the ability to operate systems independently.
Those requirements can vary significantly. Keeping data inside a country does not necessarily determine who owns the model, controls software updates or can access the infrastructure. A fully disconnected deployment offers stronger operational separation but may also require customers to manage more hardware, security and maintenance themselves.
The Microsoft-Mistral agreement combines a European model developer and European computing infrastructure with the software, security tools and distribution network of a U.S.-based cloud provider. Microsoft President Brad Smith and Mistral CEO Arthur Mensch told Reuters that the companies were seeking to provide local control while retaining access to Microsoft’s software and security capabilities.
The companies said their expanded relationship includes a joint sales plan for enterprise customers in Europe and other markets.
Europe seeks more control over computing capacity
The partnership comes as the European Union tries to reduce strategic dependence on foreign technology while expanding domestic AI development.
The European Commission has proposed a Cloud and AI Development Act intended to increase energy-efficient data-center capacity and support European cloud and AI services. The Commission has also backed AI factories, shared supercomputing resources and pilot programs for public agencies.
Three pilot projects that began July 1 are developing reusable generative-AI systems for public administrations. The EuropAI program is testing systems designed to meet European requirements for security, transparency and data governance.
Microsoft and Mistral’s original partnership drew regulatory attention in 2024, though Britain’s Competition and Markets Authority determined that it did not qualify for a merger investigation.
The expanded agreement does not identify specific enterprise customers or provide a detailed timetable for Mistral’s additional European capacity. Adoption will depend on whether organizations conclude that the deployment options meet their legal, security and cost requirements.
The agreement is another indication that model performance is no longer the only factor in enterprise AI purchasing. The location of computing resources, control over operations and the ability to move workloads among environments are becoming part of the product decision.