In-region inference, open models, and new European infrastructure for sovereign AI

What changed

Source summary

Mistral made its Regional Endpoints generally available, letting customers choose whether their AI inference runs in Europe or the US, and launched a new Priority Tier (public preview) offering committed service levels, custom rate limits, and an uptime SLA. It also opened its platform to third-party open models beyond its own, starting with Z.ai's GLM-5.2, under the same regional and service guarantees. Separately, Mistral announced a European Compute Units initiative with enterprises including ASML, CMA CGM, Amadeus, and Caisse des Depots to build up to 1 GW of European AI compute capacity by 2030.

Why it matters to your work

Practical interpretation

If data residency or EU compliance matters for your organisation, you can now choose to keep AI processing inside Europe rather than the US, with a paid tier offering guaranteed uptime and response times. This is mainly relevant if you are evaluating Mistral for a business deployment with data-sovereignty requirements — it does not add anything for casual chatbot users.

When this could help

Relevant now for any organisation evaluating Mistral that needs to guarantee data stays in a specific region, or tracking European AI sovereignty initiatives.

Mistral now offers in-region inference with a choice of Europe or the US, so processing stays in the selected region. Open-weights models let customers inspect, adapt, and keep what they build. Third-party open models (starting with Z.ai GLM-5.2) run on the same regional infrastructure. A Priority Tier with committed SLAs is in public preview, and European Compute Units (ECUs) offer multi-year compute commitments toward a planned 1 GW of European capacity by 2030.

What to check

• Only Europe and the US are available as regions. Limited transfers to sub-processors outside the chosen region can still occur under safeguards. Source ↗
• Priority Tier is public preview only. No pricing is published for inference or ECUs. Only one third-party open model (GLM-5.2) is available so far. Source ↗