Mistral and NVIDIA partner on industrial engineering AI, Mistral Large 4 weights open by end of October
Mistral announced an NVIDIA partnership for industrial engineering AI and disclosed Mistral Large 4 specs: roughly 1 trillion total parameters, 49 billion active, trained on 3800 Grace Blackwell GPUs, targeting 1GW of compute by 2030 with weights opening end of October.
Mistral announced on October 21 a partnership with NVIDIA to bring large-model capability for computer-aided engineering and electronic design automation workflows onto NVIDIA's CUDA-X, Warp and Omniverse stacks. Mistral said the first deployment targets simulation, structural analysis and chip design flows, aimed at shortening engineering iteration cycles rather than general conversation.
The same day Mistral disclosed key specs for Mistral Large 4: roughly 1 trillion total parameters with about 49 billion active at inference, a sparse mixture-of-experts design, trained on 3800 NVIDIA Grace Blackwell GPUs. Mistral also published a compute roadmap targeting 1GW of data center capacity for the system by 2030.
Mistral committed to releasing Mistral Large 4 weights by the end of October, enabling private self-hosted deployment for enterprises that require data to stay on premises. Neither side disclosed commercial terms and NVIDIA did not issue a separate statement.