Skip to content
NewsProduct launch

Japan and Nvidia announce 27,500-Rubin-GPU AI factory and national foundation model effort backed by $6.2B sovereign pledge

· by Pondero Newsdesk

The short version

Japan's Noetra Corp. and Nvidia confirmed a national AI factory built on 27,500 Rubin GPUs targeting 140 MW of capacity by 2028, with Toyota, Fanuc, Yaskawa, Kawasaki, and Fujitsu as industrial launch partners.

Japan and Nvidia announce 27,500-Rubin-GPU AI factory and national foundation model effort backed by $6.2B sovereign pledge

Japan committed to a specific compute target for national AI on July 16: a data center housing 27,500 Rubin GPUs and 13,750 Vera CPUs, aimed at 140 megawatts of capacity, with a 2028 operational target. The announcement named the entity responsible for building and running it.

What

Per Nvidia's July 16 press release, Japan's government is partnering with Nvidia and a national consortium called Noetra Corp. to build a Vera Rubin AI factory. The hardware configuration: 13,750 NVIDIA Vera CPUs and 27,500 NVIDIA Rubin GPUs in Vera Rubin NVL72 racks, networked on the NVIDIA DSX platform with Spectrum-X Ethernet. Noetra will use the factory to train open multimodal foundation models for industrial AI, robotics, and physical AI applications, with pretrained weights to be released broadly to domestic developers.

Japan's Ministry of Economy, Trade and Industry is backing the project under its FRONTia program, formally titled "Development of Multimodal Foundation Models with a View to AI Robotics and Physical AI." TechCrunch reported that Tokyo is committing up to 1 trillion yen ($6.2 billion) over five years to the sovereign AI effort, citing Nikkei Asia. Noetra has roughly 44 founding members, with SoftBank, Sony, NEC, and Honda at the core, per TechCrunch.

Nvidia CEO Jensen Huang spent July 15 and 16 in Tokyo. Per the NVIDIA blog, he met for lunch with the CEOs of Fujitsu, Kawasaki Heavy Industries, Fanuc, and Yaskawa, all four of which announced plans to build on Nvidia's Cosmos models. Toyota, which committed its next-generation L2++ vehicles to Nvidia's DRIVE AGX platform at CES in January 2025, added new integration work covering factory simulation, vehicle software engineering, and urban traffic intelligence.

Why it matters

The 27,500-GPU factory is sized for trillion-parameter model training. Nvidia says the factory will support that scale; Noetra's roadmap calls for models of that class to power industrial robotics. For operators watching physical AI deployment, the Cosmos coalition is the more immediate signal. Fanuc, Yaskawa, and Kawasaki are among the largest suppliers of industrial robots to automotive and electronics factories globally. Their commitment to train on Noetra's Cosmos-based models sets a path toward standardized physical AI inference at the factory level, not just in research.

The open-weight distribution plan matters. Noetra's stated intention to release pretrained weights to outside developers would put Japanese-industrial-data-trained foundation models into the ecosystem. That differs from the U.S. hyperscaler approach, where training data and model weights typically stay proprietary.

Context

TechCrunch's July 19 analysis framed Japan's approach as structurally distinct from U.S. hyperscaler-led AI development: sovereign compute owned nationally, combined with existing industrial automation leaders whose factory-floor data provides the training asset. Japan's AI Robotics Strategy, released in March, targets more than 30% of the global AI robotics market by 2040, a market Tokyo values at roughly $133 billion. The government has also set a goal of 10 million AI-equipped robots across 18 sectors by 2040.

Japan's push for compute sovereignty runs on American silicon for now. The DSX-based Vera Rubin factory hardware comes from Nvidia. What Tokyo controls is the training data, the legal entity (Noetra), and downstream deployment through companies that manufacture most of Japan's robots.

Huang appeared alongside trade minister Ryosei Akazawa at the government's physical AI launch event on July 16. Prime Minister Sanae Takaichi joined by video.

What to watch next

Noetra has outlined a three-stage model roadmap, per TechCrunch: a reasoning model heavy on Japanese-language skills in fiscal 2026, an omni-modal version handling text, images, video, and audio by 2028, and "Real-world Native AI" built to run robots by 2030. The fiscal 2026 reasoning model is the first milestone to watch. Hardware supply for the 2028 factory target also depends on TSMC's Japan fab output, which the Rubin GPU production schedule will track against.

Sources