Mistral opens its platform to third-party models and pitches European AI infrastructure

Mistral AI announced a set of infrastructure products aimed at enterprise and government customers who want their AI workloads to stay within Europe, alongside a pledge, backed by a coalition of European firms, to build up to 1GW of compute capacity by 2030.

The announcement bundled three things. Regional inference endpoints, now generally available, let customers choose whether their inference runs in Europe or the US, keeping processing in-region for data residency purposes. A Priority Tier, in public preview, adds committed rate limits and an uptime service-level agreement — which Mistral said made it the only European AI lab offering both regional choice and SLA-backed commitments. And Mistral began hosting third-party open models on the same infrastructure, starting with Zhipu’s GLM-5.2 — a rival lab’s model, not Mistral’s own.

The compute pledge is a coalition of what Mistral called anchor enterprises committing to multi-year capacity agreements in exchange for “European Compute Units” giving them access to the infrastructure. Named participants include Amadeus, ASML, Capgemini, Caisse des Dépôts and CMA CGM.

Taken together, the announcement marks a shift in Mistral’s business from training and releasing open-weight models toward selling regional and sovereign inference infrastructure — a market position built on European data-residency rules and doubts, sharpened by geopolitics, about routing sensitive workloads through US or Chinese cloud providers. Hosting a competitor’s model is a departure from the industry norm of vertically integrated model-plus-inference offerings, betting that operating the regional infrastructure layer is worth more to Mistral than exclusivity over which models run on it.