NVIDIA and Japan-backed Noetra plan 140MW AI factory for physical AI models

NVIDIA says it is working with Japan-backed Noetra Corp. on a national AI factory that will anchor Japan's FRONTia Project, a government-supported program aimed at building multimodal foundation models for robotics and physical AI.

The July 16 announcement centers on a Vera Rubin AI factory using 13,750 NVIDIA Vera CPUs and 27,500 NVIDIA Rubin GPUs. NVIDIA says the system will be based on its DSX architecture, connected with Spectrum-X Ethernet networking, and sized for 140 megawatts of data center capacity. The company describes the deployment as the first national AI infrastructure specifically organized around physical AI.

Why it matters

The project is notable because it links sovereign AI infrastructure to industrial automation rather than only chatbots, office software or general cloud services. According to NVIDIA, the compute environment is intended to support open multimodal foundation models that can be used for AI agents, digital twins, robotics and other systems that must reason about real-world equipment and environments.

Japan's Ministry of Economy, Trade and Industry is supporting the FRONTia Project, formally described by NVIDIA as a program for developing multimodal foundation models for AI robotics and physical AI. NVIDIA says Noetra plans to make pretrained weights from the models broadly available to domestic model developers and enterprises, alongside software including Nemotron, Cosmos, Isaac GR00T and NeMo libraries.

The hardware scale also signals how industrial AI is moving from pilot projects toward shared national infrastructure. NVIDIA says the factory will eventually support training trillion-parameter-scale models and give organizations across Japan access to an advanced AI environment for manufacturing, logistics, healthcare, telecommunications and related sectors.

What remains uncertain

The announcement does not provide financial terms, a full construction timetable or independent performance benchmarks. It also relies on several forward-looking claims about model development, token costs and industrial adoption. Those details matter because building useful robotics models requires not just chips, but trustworthy industrial data, deployment partners and safety validation in real facilities.

Even so, the project gives Japan a concrete vehicle for combining its manufacturing base with national AI policy. If Noetra and its partners can translate the infrastructure into broadly usable models, the result could make physical AI a more practical layer for factories, logistics networks and robotics developers.

Sources

• nvidianews.nvidia.com

Cover photo by panumas nikhomkhai on Pexels, used under the Pexels License.