Japan is making its biggest bet yet on sovereign artificial intelligence. NVIDIA announced on July 16 that it is partnering with Noetra Corp., a newly formed Japanese consortium, to build a Vera Rubin AI factory packing 27,500 NVIDIA Rubin GPUs and 13,750 Vera CPUs — what the companies describe as the world's first national AI infrastructure dedicated to physical AI. The 140-megawatt facility will anchor a government-backed program worth up to 1 trillion yen, or roughly $6.2 billion, over five years.
A Consortium Built by Japan's Industrial Giants
Noetra is not a typical AI startup. The consortium was founded by SoftBank Corp., Sony, NEC and Honda, with investment from 44 companies and organizations spanning Japan's industrial base. Japan's Ministry of Economy, Trade and Industry (METI) and its innovation funding arm NEDO have jointly commissioned Noetra and AIST, the national research institute, to deliver the FRONTia Project — formally titled "Development of Multimodal Foundation Models with a View to AI Robotics and Physical AI."
The funding structure is deliberate. An initial tranche of 387.3 billion yen (about $2.4 billion) is committed for fiscal year 2026, financed through GX Economy Transition Bonds, while funding beyond the first two years is subject to annual stage-gate reviews. The full $6.2 billion is a ceiling, not a guarantee — a discipline mechanism rarely seen in headline AI announcements.
NVIDIA CEO Jensen Huang, who attended the Tokyo announcement, framed the project in historical terms, saying Japan invented modern manufacturing and is now building the AI factories to power the next industrial revolution.
Inside the 140MW Machine
The factory's specifications place it firmly at the frontier of AI infrastructure:
- The chip counts divide exactly into 382 Vera Rubin NVL72 racks, each housing 72 Rubin GPUs and 36 Vera CPUs.
- The racks sit on NVIDIA's DSX reference data center platform, stitched together with Spectrum-X Ethernet networking.
- The 140-megawatt envelope is sized to train trillion-parameter-scale multimodal foundation models.
Neither party disclosed total project cost, but with VR200 NVL72 systems currently quoted between $5 million and $7 million apiece, industry observers estimate the rack hardware alone at $1.9 billion to $2.7 billion. No firm deployment timeline was given; Rubin racks only reach volume production in the second half of 2026, pointing to a phased ramp.
From Reasoning Models to Real-World Native AI
FRONTia's roadmap runs on a three-stage clock. Noetra targets a reasoning foundation model in fiscal 2026, an omni-modal model that processes text, images, video and audio by fiscal 2028, and "real-world native AI" with genuine spatial awareness by fiscal 2030 — the kind of model that could drive humanoid robots, autonomous factory equipment and mobility systems.
Crucially, the pretrained weights of Noetra's foundation models will be made broadly available to domestic developers and enterprises, alongside NVIDIA's Nemotron open models, Cosmos world models, Isaac GR00T robotics models and NeMo libraries. The announcement capped a week of Japan-focused NVIDIA news, including July 15 disclosures that the country's physical AI leaders are building on the Cosmos, Isaac, Metropolis and Jetson platforms across manufacturing and mobility.
Why It Matters
This deal crystallizes three trends reshaping the AI landscape at once. First, sovereign AI is going physical: while most national AI initiatives have chased chatbots and language models, Japan is aiming its compute directly at robotics — a domain where its manufacturing giants like Honda, Sony and NEC still hold deep advantages. Second, it is a template for state-industrial AI financing, blending transition bonds, stage-gate reviews and a 44-member consortium rather than a single national champion. Third, it hands NVIDIA a marquee win for its Vera Rubin generation before volume production even begins, reinforcing the company's grip on the full AI stack from GPUs to networking to robotics software.
For the global open-model ecosystem, the commitment to release FRONTia's pretrained weights could eventually make Japan a significant contributor to open multimodal AI — an area currently dominated by Chinese labs. And for every country debating how to fund AI sovereignty, the 140-megawatt factory rising under Mount Fuji's shadow will be the experiment to watch: proof, or disproof, that national AI infrastructure can be built on industrial policy rather than hyperscaler balance sheets.
The stakes are as much strategic as economic. General-purpose robots are moving from research labs toward mass-market deployment, and the winners of that shift will be the nations whose models understand the physical world — not just the internet's text.
