Mistral AI is turning European AI sovereignty into a product with service-level agreements attached. The French lab announced a three-part infrastructure expansion on August 11, backed by multi-year compute commitments from Amadeus, ASML, Capgemini, and CMA CGM.
The plan under writes 200 megawatts of European infrastructure by the end of 2027 and a full gigawatt by 2030, a buildout the company frames as a roughly $38B commitment. Customers sign five-year deals with no exit, betting that European enterprises will pay for control over where their AI workloads run.
Mistral Regional Endpoints, now generally available, let customers pin inference to Europe or the United States, and a new Priority Tier adds an uptime SLA for mission-critical work. The company is also hosting third-party open models under the same controls, starting with GLM-5.2 from Chinese lab Z.ai, positioning itself as a sovereign distribution layer for any open-weight model.
The sovereignty pitch has limits. Mistral acknowledges some tool calls, like web search, may leave the region, and executives say gating those capabilities is the answer rather than pretending they do not exist.
Microsoft looms over the strategy as both tenant and funder. A July deal has the hyperscaler renting capacity from Mistral’s European data centers, giving the lab revenue to build infrastructure it also sells to sovereignty-focused customers. The bet: in a world dominated by American and Chinese frontier labs, the durable business is selling everyone else control.
