Mistral partners with Nvidia to accelerate industrial design and engineering simulation with physical AI

The two companies plan to combine Mistral Large 4 and Mistral's physical AI with Nvidia's CUDA-X, the Warp software library, and Omniverse for 3D simulation, aiming to connect design, simulation, and AI training into a continuous workflow for aerospace, automotive, and semiconductor industries. They also intend to release public benchmarks and open datasets for testing and comparing physical models and AI agents working with engineering software. Mistral Large 4 has roughly 1 trillion parameters, was trained in Europe on 3,800 Grace Blackwell GPUs, and is slated for release as an open-weights model by the end of October. Mistral also aims to scale compute to 1 GW by 2030. The two firms previously co-launched Mistral NeMo and are advancing an open foundation model under the Nemotron Coalition. For simulation engineers in aerospace, automotive, and semiconductors, the direct value is bringing modeling, simulation, and AI training into a single pipeline; for developers automating engineering software, the upcoming benchmarks and datasets will determine whether physical models and agents can be compared head-to-head.