Mistral AI and Nvidia announce Mistral Compute, a French AI cloud

2025-06-11 · Publisher : Mistral AI · Domain : Cloud & DevOps

At VivaTech, Mistral AI and Nvidia announce Mistral Compute, a full AI platform (GPUs, orchestration, APIs) hosted in Europe, with 18,000 Blackwell chips according to the press, and partners such as BNP Paribas, SNCF and Orange.

### Background On 11 June 2025, at VivaTech, Jensen Huang (Nvidia), together with Mistral AI, unveils **Mistral Compute**. Mistral describes it as a private, integrated stack: GPUs, orchestration, APIs, products and services. It turns the company, until then a model developer, into a full AI infrastructure provider, with a discourse of European sovereignty and decarbonized energy. ### Key facts - Nvidia is the main partner, with its latest reference architectures; Mistral speaks of "tens of thousands of GPUs", the business press of 18,000 Grace Blackwell superchips. - Launch partners: Black Forest Labs, BNP Paribas, Kyutai, Mirakl, Orange, Schneider Electric, SLB, SNCF, Thales and Veolia. - The platform lets customers train and serve any AI workload, with targeted uses for defense, pharmaceuticals and financial markets. - Nvidia says it wants to multiply European AI computing capacity tenfold within two years. ### Why it matters Mistral Compute is the first European attempt to build an integrated "AI cloud" around a model lab rather than a hyperscaler. For European companies, it is an alternative to US providers for training and inference, with the same questions about commitment and availability. Actual capacity and sites will have to be confirmed by deliveries. One should still distinguish announcement from delivery: Mistral's release speaks of "tens of thousands of GPUs", the 18,000-chip figure comes from the press. ### Takeaway The French lab also becomes a compute provider, with Nvidia and French industrial partners.

Tags : Souveraineté, NVIDIA, IA, Puces, Europe