NVIDIA and Japan are building the world's first national AI factory — powered by Vera and Rubin

There’s a difference between building an AI data center for a single company and building one for an entire nation. NVIDIA and Japan just committed to the latter.

The company announced Tuesday a partnership with Japanese political and business leaders to build what it calls the world’s first national-level artificial intelligence infrastructure — a 140MW “National Physical AI Factory” that will sit at the center of Japan’s AI sovereignty strategy.

The facility, built in collaboration with Japanese tech firm Noetra on NVIDIA’s DSX platform, will run Vera Rubin NVL72 racks packing 13,750 Vera CPUs and 27,500 Rubin GPUs. The compute cluster is interconnected via NVIDIA’s Spectrum-X Ethernet fabric — the same high-speed networking backbone that links the largest AI supercomputers worldwide.

The scale isn’t accidental. Japan is using this factory to power its FRONTia initiative, a government-backed project focused on developing multimodal foundation models for AI robotics and physical AI. Physical AI — AI that can perceive, reason about, and act in the physical world — requires fundamentally different compute than the chatbot-style models that dominated the last two years. It needs models that understand 3D space, object manipulation, and real-time sensor data simultaneously.

That difference explains the hardware mix. Rubin GPUs provide the raw tensor throughput, but the Vera CPUs handle the complex simulation and physics pipelines that physical AI training demands. Together they push toward what NVIDIA calls “T-level” (trillion-parameter) model training — a regime that was barely conceivable two years ago.

Noetra’s multimodal foundation model pre-training weights will be released openly alongside NVIDIA’s own Nemotron, Cosmos, Isaac GR00T, and NeMo libraries. The goal is to give Japanese developers and enterprises access to the same foundational tools that large US AI labs work with, rather than forcing them to start from scratch.

The announcement is the clearest sign yet that sovereign AI infrastructure — government-backed compute capacity built for national strategic goals rather than corporate cloud profits — is moving from concept to construction. Japan, like the EU and a growing list of Asian nations, is deciding it cannot afford to outsource its AI compute to companies based in Silicon Valley or Beijing.

The factory won’t just serve researchers. Physical AI models trained on this infrastructure could end up in Japanese factory floors, construction sites, elder care facilities, and disaster response robots — applications where the country faces acute labor shortages and sees AI as a practical solution rather than a theoretical one.