Mistral adds AI endpoints, third-party model support and gigawatt-scale infrastructure

Mistral positions itself as the core infrastructure that will enable Europe to preserve AI sovereignty. The company offers organisations and states full control over models, execution infrastructure and compute resources while guaranteeing compliance with local regulatory requirements. As part of the broader effort, Mistral expands access to open models from external sources, launches regional endpoints and dedicated service tiers, and creates a coalition to secure future production capacity. The plan includes building infrastructure up to one gigawatt by 2030, intended to deliver scalable capacity that keeps value and control with the customers themselves.

A large share of the company’s customers currently run their models in cloud environments and private data centres. As AI becomes more deeply embedded in production systems, demand for reliability and resource availability under regional oversight is growing. According to the company, for some customers the appropriate solution is to augment their own infrastructure with production capacity operated and managed directly by Mistral.

In the execution layer this means keeping data and processing within the selected region, together with stable resource access during peak load. The Mistral Regional Endpoints service, now officially available, lets customers choose whether execution occurs in Europe or the United States to satisfy data-residency, regulatory and latency requirements. Processing is performed in the chosen region, except for limited, secured sub-processor transfers outside the region as described in the company’s trust centre. In addition, the new Mistral Priority Tier service, now in public preview, provides binding service levels for critical workloads, including custom rate-limit settings and availability commitments. Mistral notes it is the only European AI processor offering both a regional processing choice and a formal service commitment. The company’s models will continue to be accessible through external partners.

Control over execution location is only one part of the equation. Customers also need control over the models they run on the infrastructure. Modern AI systems no longer rely on a single model but on a full suite of capabilities: extended reasoning, high-throughput processing for manufacturing, or workloads tailored to a company’s internal data. Mistral develops models for each of these uses, ranging from cutting-edge offerings to specialised models such as Mistral OCR and Voxtral, as well as custom models trained on a client’s own knowledge.

Customers value the company primarily for its leadership in open models. Open models give critical workloads exactly what is required: the ability to inspect the model, adapt it, and retain the knowledge built with it. For this reason the company contributes to the Open Secure AI Alliance and the Nvidia Nemotron Coalition. It is now extending this openness beyond its own models. The platform will support third-party open models, beginning with GLM-5.2 from Z.ai. That model and future open models will run on the same infrastructure, under the same regional controls and service commitments as Mistral models, allowing customers to broaden their model selection.