Microsoft Azure AI Deals Add Cobalt Chips And French Capacity

Microsoft is tying AI software distribution more tightly to where enterprise workloads run, with new Databricks and Mistral agreements that extend Azure usage, Microsoft-designed processors and French-operated AI capacity.

Cloud Computing News reported the expanded partnerships after Microsoft, Databricks and Mistral detailed the arrangements.

The agreements split the AI cloud buildout across two operating paths: Databricks is moving more internal and analytics workloads onto Azure and Cobalt processors, while Mistral is offering European infrastructure that Azure customers can use without relying only on public-cloud regions run by the US company.

Databricks Adds Cobalt To Its Azure Workload Plan

Databricks will use Azure Databricks for core business operations, analytics and its unified lakehouse.

The agreement also increases Databricks' use of Azure Cobalt, Microsoft's Arm-based server processor line, for data-intensive operations and agentic AI workloads.

Databricks said the expanded Microsoft partnership will continue into the 2030s.

Microsoft said Cobalt 200 can deliver up to 50% better generational performance than Cobalt 100, depending on workload, with memory encryption enabled by default.

The processor is designed for scale-out work such as data pipelines, cloud databases and agentic AI applications, while Azure Boost handles networking and remote-storage offload.

Governance Tools Move Across Microsoft Products

The product layer is also expanding beyond compute.

Databricks Genie, a conversational analytics tool, and Unity AI Gateway, an operational-control layer for models, agents, tools and external AI services, are set for deeper integration across Microsoft services.

Unity AI Gateway centralises governance for AI services by combining access rules, spending limits, usage records, audit trails, routing controls and prompt or tool monitoring.

Enterprises using the operating layer can set spending limits, govern access, route models and keep audit records alongside identity, analytics, collaboration and Copilot workflows.

The integrations span Entra, OneLake, Power BI, Purview, Foundry, Teams and Copilot.

The deployment path puts governed AI operations beside business data and productivity tools rather than treating model access as a separate cloud service.

Mistral Adds French Capacity For Azure Customers

The Mistral agreement adds a different infrastructure route.

Microsoft committed billions of dollars to the European computing buildout, but the public terms do not give a total value or payment schedule.

Azure customers will be able to develop software using data centres in France operated by Mistral through the partnership.

Microsoft President Brad Smith put the arrangement in terms of combining American and European technology while providing continuous and assured access through Azure Local and Mistral computational capacity.

Cloud Computing News reported that Mistral is targeting one gigawatt of computing capacity by 2030.

The companies have not identified how much of that target will be covered by Microsoft's spending commitment.

The deployment routes include Azure, Azure Local and infrastructure in Europe run by Mistral.

Model access changes by product and environment, while Azure Local gives organisations a way to run selected Azure services and management tools outside Microsoft's public cloud regions, including disconnected deployments.

Model Access Still Depends On Environment And Licence

Microsoft Foundry now includes Mistral Medium 3.5 and OCR 4, and Medium 3.5 is available through Copilot Studio.

Azure Local customers will also be able to run Mistral models made available under open or open-weight licences, with modification and development rights depending on each model's licence.

Microsoft and Mistral have not said how much of Mistral's planned capacity the spending commitment will fund.