Mistral opens platform to third-party models and targets 1GW European compute by 2030

Mistral is repositioning itself as a sovereign European AI infrastructure and distribution layer rather than just a model maker. The company began hosting third-party open models on its platform, starting with Z.ai's GLM-5.2, letting enterprises and governments run a wider array of models while keeping regional control and service commitments — a notable move for a lab hosting a Chinese rival's weights.

On infrastructure, Mistral announced a European Compute Units coalition with ASML, CMA CGM, Amadeus, and Capgemini, targeting 200 MW by the end of 2027 and 1 GW by 2030. It framed the strategy around three pillars: choice of intelligence (which model), choice of inference location (where it runs), and assured compute (guaranteed capacity).

Operationally, Mistral made its Regional Endpoints generally available — pinning inference to EU or US regions with data-residency guarantees — and introduced a Priority Tier in public preview offering a 99.5% uptime SLA for critical workloads.

The competitive logic is sovereignty: European enterprises and regulators increasingly want AI that stays on-continent, and Mistral is betting that data residency plus assured capacity is a durable differentiator against US hyperscalers. Hosting GLM-5.2 also hedges Mistral's own model roadmap, giving customers frontier-class Chinese open weights under European governance. The open question is capital — 1 GW by 2030 is an enormous buildout for a company far smaller than the hyperscalers announcing hundreds of billions in financing this same week, and execution risk is high.