Nvidia, the Japanese government, and roughly 44 Japanese firms launched a sovereign AI initiative called Noetra during Jensen Huang's July 15–16, 2026 visit to Tokyo, backed by up to 1 trillion yen (approximately $6.2 billion) in government funding over five years TechCrunch. The consortium's core members include SoftBank, Sony, NEC, and Honda, with a broader roster spanning Fanuc, Yaskawa, Kawasaki Heavy, Fujitsu, Hitachi, Kubota, and AIRoA TechCrunch.

Nvidia and its partners characterized the announcement as the world's first national AI infrastructure Nvidia. Huang delivered a speech on July 16 at an event promoting Japan's physical AI policy, telling attendees: "The next frontier of AI is in the physical world, and this is a once-in-a-generation opportunity for Japan" TechCrunch.

The compute backbone for Noetra is a Vera Rubin AI factory data center that Nvidia is building for Japan. It is expected to launch in 2028 with 13,750 Vera CPUs and 27,500 Rubin GPUs, delivering 140 megawatts of compute TechCrunch. This sits atop an earlier commitment: SoftBank is building Japan's most powerful AI supercomputer using Nvidia Blackwell technology, announced in November 2024 Nvidia.

Noetra's roadmap unfolds in three stages. A Japanese-language reasoning model begins in fiscal 2026. An omni-modal model follows by 2028. The final stage, labeled "Real-world Native AI" for robots, targets 2030 TechCrunch.

The robotics and manufacturing thread is where this announcement gets dense. Fanuc, Yaskawa, Kawasaki Heavy, Fujitsu, Hitachi, NEC, Sony, SoftBank, Kubota, and AIRoA announced plans to build on Nvidia's Cosmos models, the open-model effort for physical AI that Nvidia started in May 2026 TechCrunch. Honda R&D and Omron are building on Nvidia's Cosmos tools as well. Some companies in the Japanese robotics coalition are already testing a shared control system based on Cosmos TechCrunch.

At the Tokyo event, Nvidia unveiled Cosmos 3 Edge, a version of Cosmos designed to run on Jetson Thor chips inside machines TechCrunch. The combination of edge inference on Jetson Thor with a shared Cosmos-based control layer points toward a coordinated robotics stack rather than isolated per-vendor systems. That several of Japan's major industrial robotics firms are already testing a shared control system suggests the collaboration has moved past the memorandum-of-understanding stage.

Beyond Noetra's core plan, SoftBank is looking to partner with Microsoft and Sakura Internet to develop AI, as reported during Huang's Tokyo visit CNBC. Separately, Japanese enterprises and startups are building industry-specialized AI using Nvidia Nemotron open models Nvidia.

What stands out here is the alignment of three normally separate vectors: national government funding, a coordinated industrial consortium spanning competing firms, and a single hardware/software vendor providing the full stack from training infrastructure (Vera Rubin data center) through open foundation models (Cosmos, Nemotron) to edge inference silicon (Jetson Thor running Cosmos 3 Edge). National AI strategies have typically focused on one or two of these layers. Japan's approach attempts all three simultaneously, with a concrete fiscal timeline and named corporate participants at each layer.

The 2028 target for the Vera Rubin AI factory and the omni-modal model is ambitious but not unprecedented for infrastructure of this scale. The 140-megawatt compute envelope is substantial, and the 27,500-GPU count puts it in the range of large commercial training clusters. Whether the fiscal 2026 Japanese-language reasoning model arrives on schedule will be an early indicator of whether Noetra operates as a functioning consortium or a coordinating body around independent efforts.

The "Real-world Native AI" stage for 2030 is the most speculative element. Japan's robotics sector is genuinely strong in industrial automation, and the involvement of Fanuc, Yaskawa, and Kawasaki Heavy gives the robotics ambition credible grounding. But bridging from language and vision models to physical control systems that operate reliably in unstructured environments is a problem that has consumed decades of robotics research. A shared Cosmos-based control layer, if it materializes, would be the mechanism that connects the AI factory to the factory floor.

For now, the pieces are in place: funding, hardware, models, edge silicon, and a consortium of companies that have signed on to build within a common stack. The execution phase starts now.