Mistral enters Physical AI with Robostral Navigate at €20B valuation

Robostral Navigate is Mistral's first step into robotics and 'Physical AI,' and it targets one of the harder embodied problems: navigation from a single camera plus simple language instructions, without expensive multi-sensor rigs. Mistral reports a 76.6% success rate, beating prior single-camera approaches by 9.7 points — a meaningful margin that positions the model as a practical, low-hardware-cost entry point for robotics developers.

The launch builds on Mistral's expanding industrial footprint, including partnerships with Airbus and BMW, and arrives as the company pushes toward a €20 billion valuation and cements its status as Europe's leading sovereign-AI champion. It also comes days after Microsoft and Samsung announced multibillion-dollar infrastructure partnerships with Mistral, giving it European GPU capacity (including NVIDIA hardware) to train larger models.

Competitively, Mistral entering robotics the same week Google DeepMind shipped Gemini Robotics 2 signals that foundation-model labs now view Physical AI as a core battleground, not a side quest. Mistral's single-camera, low-cost angle differentiates it from DeepMind's whole-body humanoid focus and could appeal to industrial customers wary of heavy sensor stacks. What to watch: independent benchmarks beyond Mistral's own numbers, which robots ship Robostral Navigate first, and whether the Airbus/BMW relationships convert into deployed navigation systems.