Mistral pushes sovereign AI infrastructure
Archive item — written before sources were shown.
Mistral made Regional Endpoints GA across Europe and the US, added Z.ai's GLM-5.2 to its platform, and outlined plans for up to 1GW of capacity by 2030.
Mistral made Regional Endpoints generally available, letting customers commit their inference to Europe or the US specifically, and opened a public preview of a Priority Tier with committed service levels and an uptime SLA. The company also added support for Z.ai’s GLM-5.2 on its platform, alongside partnerships tied to Nvidia’s Open Secure AI Alliance and Nemotron Coalition.
On infrastructure, Mistral outlined plans to build up to 1 gigawatt of compute capacity by 2030 and introduced European Compute Units, a multi-year commitment model for accessing that capacity. The announcement carried quotes from customers including ASML, CMA CGM, Amadeus, Caisse des Dépôts, and Factory, framing the push around European enterprises that need AI infrastructure they control rather than rent from a US hyperscaler.
What it means for you
For teams with genuine EU data-residency or sovereignty requirements, this converts Mistral’s regional promise from a policy statement into a shipping product: Regional Endpoints are now GA, not preview. For everyone else, the more interesting detail is Mistral running a third-party open model, GLM-5.2, on its own platform, one more sign that model providers are becoming routing layers as much as model builders, a shift covered in ai routers become a shipped feature. If sovereignty or in-region processing shows up in your compliance checklist, this is worth a direct evaluation now that it’s generally available.
- 01In-region inference, open models, and new European infrastructure for sovereign AImistral.ai · primary
