Two Announcements, One Press Release
On 11 August 2026, Mistral published a single announcement that bundled two different products under one sovereign-AI banner. The first is Regional Endpoints, now generally available: a customer picks Europe or the United States, and Mistral states plainly that inference and its associated processing take place in the selected region, with only limited, safeguarded transfers to sub-processors permitted outside it under the company's own Trust Center terms. This is infrastructure that exists today and can be tested against a contract this week.
The second product is European Compute Units, or ECUs - multi-year forward purchases of data-center capacity that Mistral has not finished building. The company's stated goal is 200 megawatts of European compute by the end of 2027, rising to as much as a gigawatt by 2030. Five anchor customers have already committed: ASML, Amadeus, Capgemini, Caisse des Dépôts and CMA CGM. Rather than raising capital first and selling capacity later, Mistral is doing the reverse - selling multi-year capacity commitments now to help finance the build.
What You Can Actually Verify Today
Regional Endpoints is the part of this announcement an EU compliance officer can act on immediately. It is a live product: choose the EU region, and processing stays there under Mistral's published terms, which is the kind of concrete data-residency commitment that GDPR-conscious procurement teams have been asking European model providers for since well before this year. Mistral also opened a Priority Tier into public preview alongside it - custom rate limits and an uptime SLA for workloads a business cannot afford to have degrade, again something that can be checked against a contract rather than a roadmap.
None of that requires trusting a construction timeline. A company evaluating Mistral purely on Regional Endpoints and Priority Tier is buying a service that runs on infrastructure that already exists, governed by terms it can read before it signs.
What You Would Be Financing, Not Buying
An ECU is a different instrument entirely, and the difference matters at the negotiating table. It is a claim on capacity across a build-out that reaches only 200 megawatts by 2027 and does not reach its full gigawatt until 2030 - four years out. Commitments are reported at roughly five years with no early exit, meaning a signatory is locked in through construction delays, cost overruns, or a shift in Mistral's own priorities, none of which a customer controls. ASML, Amadeus, Capgemini, Caisse des Dépôts and CMA CGM are large enough to absorb that risk and are buying influence over a strategic European asset as much as future compute; a mid-sized enterprise offered the same structure is taking on a materially different risk for the same sovereignty story.
The two products get marketed together because they share a sovereignty narrative, but only one of them is something Mistral can be held to this quarter.
Before You Sign Anything Called Sovereign Compute
Ask any vendor pitching an EU AI infrastructure deal, Mistral or otherwise, to separate the two questions in writing: what is live and auditable today, and what is a forward commitment on capacity still under construction. Price a Regional Endpoints-style guarantee at its compliance value - it solves a real GDPR and data-residency problem now. Price an ECU-style forward purchase at construction-delay and vendor-execution risk, and get an explicit answer on exit terms before treating a five-year, no-early-exit commitment as equivalent to a service contract you can walk away from next renewal.
Sovereign AI is a real and growing procurement category in Europe. The companies getting the best terms from it will be the ones who can tell a live guarantee from a bet on a construction schedule - not the ones most persuaded by which one comes with a bigger number attached.
Read next: ASML and CMA CGM Now Fund Europe's AI Compute | Atos Sets a New Bar for Sovereign Cloud Claims
