Nvidia, the Japanese government, and about 44 Japanese companies have launched a joint AI project called Noetra. The announcement came during Nvidia CEO Jensen Huang's visit to Tokyo on July 15–16, 2026. The Japanese government is committing up to 1 trillion yen (about $6.2 billion) over five years TechCrunch. Key participants include SoftBank, Sony, NEC, and Honda, along with Fanuc, Yaskawa, Kawasaki Heavy, Fujitsu, Hitachi, Kubota, and AIRoA TechCrunch.

Nvidia and its partners called it the world's first national AI infrastructure Nvidia. Speaking at a Tokyo event on July 16, Huang said: "The next frontier of AI is in the physical world, and this is a once-in-a-generation opportunity for Japan" TechCrunch.

At the center of Noetra is a massive new data center that Nvidia will build for Japan, called a Vera Rubin AI factory. Think of it as a giant facility filled with specialized computers designed specifically to train AI. It is planned to open in 2028 with 13,750 Vera CPUs and 27,500 Rubin GPUs, using 140 megawatts of power, which is roughly the electricity needed for 100,000 homes TechCrunch. This builds on an earlier plan from November 2024, when SoftBank announced it would build Japan's most powerful AI supercomputer using Nvidia technology Nvidia.

Noetra's plan unfolds in three stages. First, a Japanese-language AI model starts in fiscal 2026. Second, an "omni-modal" model arrives by 2028, meaning it can understand text, images, audio, and video all at once. Third, by 2030, the project aims to create what it calls "Real-world Native AI" for robots TechCrunch.

The robotics part involves many of Japan's biggest manufacturing and robotics companies. Fanuc, Yaskawa, Kawasaki Heavy, Fujitsu, Hitachi, NEC, Sony, SoftBank, Kubota, and AIRoA plan to build on Nvidia's Cosmos models, which are open AI tools designed for physical machines that Nvidia started in May 2026 TechCrunch. Honda R&D and Omron are also using Cosmos tools. Some companies in the group are already testing a shared control system based on Cosmos TechCrunch.

At the Tokyo event, Nvidia also introduced Cosmos 3 Edge, a version of its Cosmos software designed to run on Jetson Thor chips inside physical machines TechCrunch. In plain terms, this means AI processing happens inside the robot or machine itself, instead of relying on a faraway data center. The combination of on-device processing with a shared control system points toward robots from different companies working on the same software platform, rather than each company building its own isolated system. The fact that several major robotics firms are already testing this shared system suggests the collaboration is real, not just a press release.

Beyond Noetra's main plan, SoftBank is looking to partner with Microsoft and Sakura Internet to develop AI, as reported during Huang's Tokyo visit CNBC. Separately, Japanese companies and startups are building specialized AI tools for specific industries using Nvidia's Nemotron open models Nvidia.

What makes this unusual is how three different pieces are being lined up together. Normally, national AI efforts focus on one or two layers at a time, such as funding research or building data centers. Japan's approach tries to cover all three at once: government money, a group of competing companies working together, and a single technology partner (Nvidia) supplying everything from the data center to the AI models to the chips inside the robots.

The 2028 target for the data center and the omni-modal model is ambitious but not unheard of for projects of this scale. The real early test will be whether the Japanese-language AI model arrives on schedule in fiscal 2026. That will indicate whether Noetra is working as a real collaboration or just a coordinating body around separate efforts.

The 2030 robotics stage is the most speculative part. Japan has a genuinely strong robotics industry, and having companies like Fanuc, Yaskawa, and Kawasaki Heavy involved gives the ambition real credibility. But teaching AI to control physical machines in unpredictable, real-world environments is a problem researchers have been working on for decades. A shared control system based on Cosmos, if it actually works, would be what connects the AI data center to the factory floor.

For now, the pieces are in place: funding, hardware, software, chips, and a group of companies that have agreed to build on a common platform. The execution phase starts now.