Nvidia said it will work with Japan’s Noetra Corp. on a 140-megawatt AI factory intended to serve as the compute base for FRONTia, a Japanese government-funded program focused on physical AI. Nvidia describes the project as national AI infrastructure, a label that is doing a lot of diplomatic work, but the hardware plan is concrete: 27,500 Rubin GPUs and 13,750 Vera CPUs.
The system is slated to use Nvidia’s Vera Rubin NVL72 racks on the company’s DSX reference design, linked with Spectrum-X Ethernet. Nvidia said the facility will train open multimodal foundation models for robotics, digital twins and factory automation, with pretrained weights made broadly available to developers in Japan.
Jensen Huang, Nvidia’s founder and chief executive, framed the project as a continuation of Japan’s manufacturing history, saying in the company’s announcement that Japan is building “AI factories” for the next industrial shift. Strip away the podium language and the idea is straightforward: Japan wants domestic compute for models that understand machines, spaces and industrial processes, rather than renting all of that capacity elsewhere.
What Nvidia and Noetra are building
The chip counts map cleanly onto 382 Vera Rubin NVL72 racks. Each rack contains 72 Rubin GPUs and 36 Vera CPUs, so the published totals are not a vague capacity target. They describe a specific rack-scale buildout.
Nvidia and Noetra did not disclose the cost of the facility. Tom’s Hardware has reported that VR200 NVL72 racks are currently quoted at $5 million to $7 million each, which would put the rack hardware alone at roughly $1.9 billion to $2.7 billion. Morgan Stanley estimates that Nvidia will charge about $55,000 per Rubin GPU in volume, implying around $1.5 billion for the GPU silicon before memory, networking, cooling and the rest of the expensive plumbing.
Nvidia did not give a deployment date. Rubin rack systems are expected to enter volume production in the second half of this year, and Nvidia said the factory will support trillion-parameter model training as it expands. That wording points to a staged rollout rather than a single switch-flip opening.
The consortium behind the project
Noetra is a new consortium founded by SoftBank Corp., Sony, NEC and Honda. NEC said the group has backing from 44 companies and organizations. Noetra and Japan’s National Institute of Advanced Industrial Science and Technology, known as AIST, won a New Energy and Industrial Technology Development Organization public tender on June 30 to run FRONTia from fiscal 2026 through fiscal 2030, according to Asia Times.
Asia Times reported that the program has ¥387.3 billion, about $2.4 billion, in first-year funding and as much as ¥1 trillion, about $6.1 billion, over five years. The later money is not automatic, because funding beyond the first two years depends on annual stage-gate reviews.
NEC said Noetra’s roadmap calls for a reasoning foundation model in fiscal 2026, an omni-modal model that can process text, images, video and audio by fiscal 2028, and “real-world native AI” with spatial awareness by fiscal 2030.
The FRONTia factory follows SoftBank’s Blackwell-based DGX supercomputer, announced in 2024, and FugakuNEXT, a $740 million RIKEN, Fujitsu and Nvidia zetta-scale system planned for around 2030. The distinction here is procurement: FRONTia is a state-tendered national infrastructure project, not a corporate cluster or a scientific supercomputer. Japan’s AI Robotics Strategy, released in March, targets more than 30% of the global AI robotics market by 2040, a market the government values at $133 billion.
This story draws on original reporting from Tom's Hardware.
