Nvidia will supply the compute backbone for Noetra, Japan’s newly launched domestic-AI venture, as the consortium moves to build multimodal models for robots, factories and other “physical AI” systems. Nvidia CEO Jensen Huang announced the collaboration in Tokyo on July 16, alongside Japanese government and industry representatives.
According to Nvidia, the planned AI factory will use 27,500 Rubin GPUs and 13,750 Vera CPUs in a 140-megawatt deployment based on Nvidia’s DSX platform. Construction is scheduled to start in April 2027, with operations due to begin in June 2028.
Noetra is backed by core investors SoftBank, Sony Group, NEC and Honda, with 44 participating companies and organizations in total. SoftBank’s announcement says the group will draw on engineers from its member companies, Japan’s National Institute of Advanced Industrial Science and Technology, Preferred Networks and other partners.
Noetra’s assignment is not simply to stand up another local large-language model. The group intends to build Japanese-developed reasoning and multimodal foundation models that can process text, images, video and audio, then apply that capability to robots and industrial systems operating in real-world environments.
The published roadmap calls for a reasoning model with Japanese-language understanding, logical reasoning and instruction-following capabilities beginning in fiscal 2026. An omni-modal model is targeted for fiscal 2028, while “Real-world Native AI” — models designed to understand spatial and physical properties — is the fiscal 2030 goal.
Nvidia described the effort as the computing foundation for Japan’s FRONTia project, a Ministry of Economy, Trade and Industry-backed program for multimodal models focused on AI robotics and physical AI. The company said Noetra’s pretrained model weights are expected to be made broadly available to Japanese model developers and enterprises, alongside Nvidia software and open-model components including Nemotron, Cosmos, Isaac GR00T and NeMo.
That open-availability pledge matters more than the ceremonial launch. If Noetra follows through, Japanese manufacturers, logistics operators and robotics vendors could gain a domestic base model and compute environment tuned for local industrial data and Japanese-language workflows, rather than having to build every vertical system from scratch on overseas foundation models.
For enterprise IT teams, the project reinforces Nvidia’s wider push to bundle accelerators, networking, DPUs, reference architecture and AI software into turnkey “AI factory” deployments. Nvidia says the Noetra installation will use Vera Rubin NVL72 racks, Spectrum-X Ethernet and BlueField DPUs, making it a high-profile example of that full-stack approach.
Japan’s consortium will begin model work using existing Japan-based compute resources before the Rubin facility is online in June 2028.
According to Nvidia, the planned AI factory will use 27,500 Rubin GPUs and 13,750 Vera CPUs in a 140-megawatt deployment based on Nvidia’s DSX platform. Construction is scheduled to start in April 2027, with operations due to begin in June 2028.
Noetra is backed by core investors SoftBank, Sony Group, NEC and Honda, with 44 participating companies and organizations in total. SoftBank’s announcement says the group will draw on engineers from its member companies, Japan’s National Institute of Advanced Industrial Science and Technology, Preferred Networks and other partners.
A Japanese foundation-model project aimed at robots
Noetra’s assignment is not simply to stand up another local large-language model. The group intends to build Japanese-developed reasoning and multimodal foundation models that can process text, images, video and audio, then apply that capability to robots and industrial systems operating in real-world environments.The published roadmap calls for a reasoning model with Japanese-language understanding, logical reasoning and instruction-following capabilities beginning in fiscal 2026. An omni-modal model is targeted for fiscal 2028, while “Real-world Native AI” — models designed to understand spatial and physical properties — is the fiscal 2030 goal.
Nvidia described the effort as the computing foundation for Japan’s FRONTia project, a Ministry of Economy, Trade and Industry-backed program for multimodal models focused on AI robotics and physical AI. The company said Noetra’s pretrained model weights are expected to be made broadly available to Japanese model developers and enterprises, alongside Nvidia software and open-model components including Nemotron, Cosmos, Isaac GR00T and NeMo.
That open-availability pledge matters more than the ceremonial launch. If Noetra follows through, Japanese manufacturers, logistics operators and robotics vendors could gain a domestic base model and compute environment tuned for local industrial data and Japanese-language workflows, rather than having to build every vertical system from scratch on overseas foundation models.
The practical Nvidia angle
This is a large future infrastructure commitment, not a new product release for Windows PCs or a near-term GeForce upgrade. Rubin is Nvidia’s next-generation data-center platform, and the Noetra deployment is aimed at frontier-model training and industrial AI workloads rather than consumer desktops.For enterprise IT teams, the project reinforces Nvidia’s wider push to bundle accelerators, networking, DPUs, reference architecture and AI software into turnkey “AI factory” deployments. Nvidia says the Noetra installation will use Vera Rubin NVL72 racks, Spectrum-X Ethernet and BlueField DPUs, making it a high-profile example of that full-stack approach.
Japan’s consortium will begin model work using existing Japan-based compute resources before the Rubin facility is online in June 2028.