Mistral's sovereign cloud will host models it didn't train

Mistral's pitch for European AI sovereignty now includes running other labs' models. In its August 11 post on in-region inference, open models, and new compute, the company says its platform will support third-party open models, starting with Z.ai's GLM-5.2, under the same regional controls and service commitments as its own models. Z.ai is a Chinese lab, so Mistral is spelling out what sovereignty means in its offer: who runs the weights, and where. Who trained them is a separate question.

I'm not sure I fully agree with that definition, but it makes sense for Mistral to position themselves this way. A customer can inspect, adapt, and keep an open model whatever its origin. A closed API offers none of that, whether it is served from Virginia or Paris. For a European buyer, the question moves from whose model it is to whose hardware runs it and under which contract.

The contract deserves a careful read. The new Regional Endpoints let customers pin inference to Europe or the US, but Mistral notes that processing stays in-region "subject to limited, safeguarded transfers to sub-processors that may occur outside that region." That is standard cloud practice and may be harmless for most workloads. It still means choosing the EU region is where a GDPR review starts, not where it ends, and anyone moving regulated data onto this should check the sub-processor list in Mistral's Trust Center before signing. The bigger bet is the compute coalition, which turns multi-year enterprise commitments into European Compute Units. Committing to years of capacity from one provider in order to reduce dependence is an interesting trade-off for actors skeptical of US-based hyperscalers.