Japan plans 27,500 Rubin GPUs — 44 partners back a 140MW robotics AI factory
TL;DR
Noetra and 44 partners plan a 140MW AI platform with 27,500 Nvidia Rubin GPUs, due online in June 2028.
Japan's Noetra announced a plan on July 16 for about 27,500 Nvidia Rubin GPUs, with 44 companies and organizations backing a 140MW physical-AI computing platform. Construction is scheduled to start in April 2027 and operations in June 2028. No GPU purchase price was disclosed.
Sony, SoftBank, NEC, and Honda are Noetra's core members and investors. The official participant list also includes FANUC, Fujitsu, Hitachi, Kawasaki Heavy Industries, OMRON, Preferred Networks, Sakana AI, and Tokyo Electron. Japan's AIST research institute is part of the development organization.
TNW reported that the platform also specifies 13,750 Nvidia Vera CPUs, Vera Rubin NVL72 racks, Spectrum-X Ethernet, and BlueField DPUs across 140MW of data-center capacity. Noetra is a private company, not a Japanese government agency. Nvidia working with Noetra on the facility does not mean the government directly purchased every chip.
The model roadmap has three stages. Noetra plans a Japanese-language understanding, reasoning, and instruction-following model in fiscal 2026; an omni-modal model processing text, images, video, and audio in fiscal 2028; and 'Real-world Native AI' with spatial and physical understanding in fiscal 2030 for AI agents, digital twins, and robots.
TNW reported that the related METI program has an upper limit of ¥1 trillion across fiscal 2026 to 2030, with ¥387.3 billion committed for the first two years and later funding subject to annual review. The 27,500 Rubin figure is planned capacity; Nvidia and Noetra disclosed no contract value.
via TNW / Noetra / NEC / The Japan Times
Sony, SoftBank, NEC, and Honda are Noetra's core members and investors. The official participant list also includes FANUC, Fujitsu, Hitachi, Kawasaki Heavy Industries, OMRON, Preferred Networks, Sakana AI, and Tokyo Electron. Japan's AIST research institute is part of the development organization.
TNW reported that the platform also specifies 13,750 Nvidia Vera CPUs, Vera Rubin NVL72 racks, Spectrum-X Ethernet, and BlueField DPUs across 140MW of data-center capacity. Noetra is a private company, not a Japanese government agency. Nvidia working with Noetra on the facility does not mean the government directly purchased every chip.
The model roadmap has three stages. Noetra plans a Japanese-language understanding, reasoning, and instruction-following model in fiscal 2026; an omni-modal model processing text, images, video, and audio in fiscal 2028; and 'Real-world Native AI' with spatial and physical understanding in fiscal 2030 for AI agents, digital twins, and robots.
TNW reported that the related METI program has an upper limit of ¥1 trillion across fiscal 2026 to 2030, with ¥387.3 billion committed for the first two years and later funding subject to annual review. The 27,500 Rubin figure is planned capacity; Nvidia and Noetra disclosed no contract value.
via TNW / Noetra / NEC / The Japan Times
