Nvidia and Japan launch $6.2B national physical AI factory

This story is a big swing in the geopolitics of compute. Japan is effectively standing up a national-scale AI factory, built on Nvidia’s next-generation Vera CPUs and Rubin GPUs, to power its own sovereign models for robots, vehicles and factory automation. Noetra’s long-term roadmap — from Japanese-centric reasoning models to full multimodal “real-world native” AI for robotics — signals that frontier-scale training won’t be confined to US or Chinese stacks.

For Nvidia, this cements its position as the default hardware layer for countries that want AI sovereignty without building their own chip ecosystem. It also deeply intertwines Nvidia with Japan’s industrial base: Fanuc, Yaskawa, Kawasaki, Toyota, Honda and others lining up behind the Cosmos robotics models creates a de facto standard stack for embodied AI in manufacturing. That combination of state money, national data, and a globally dominant hardware vendor is exactly the sort of configuration that can move the needle on AGI-scale experimentation.

Strategically, this pushes the race into “physical AI”: models explicitly trained to control machines, not just text and images. If Japan hits its target of 10 million AI-equipped robots across 18 sectors by 2040, the feedback loop between real-world data, foundation models and industrial deployment will be extraordinarily tight.