Jensen Huang left Tokyo with deals spanning Japan's entire tech ecosystem, and the stakes go far beyond chips

Japan is building its own AI infrastructure. Every layer of it depends on Nvidia.

A government-backed coalition of 44 Japanese companies, with SoftBank, Sony, NEC, and Honda in anchor roles, has formed under a new entity called Noetra Corp. 1. The project carries a five-year budget of up to ¥1 trillion, about $6.2 billion, to develop foundation models for robotics, automotive systems, and factory automation 1. Noetra's mandate: train models on Japanese industrial data, house them in Japanese facilities, deploy them through Japanese machines. The compute powering that pipeline is Nvidia's.

Jensen Huang spent July 15 and 16 in Tokyo 2. He left with Nvidia embedded at three levels of Japan's plan. Noetra will operate a Vera Rubin AI factory configured with 13,750 Vera CPUs and 27,500 Rubin GPUs 2. The facility is rated at 140 megawatts, with a target operational date in 2028 2. Nvidia describes the installation as the first national-scale AI infrastructure built anywhere 2. Down at the device level, Jetson Thor modules are built to sit inside the physical machines that will run Noetra's models 3.

The dependency runs from data center to factory floor. Models train on Nvidia GPUs, execute on Nvidia edge processors, and run software built on Nvidia's Cosmos platform 3. Each tier locks in the next. Once a robot fleet is trained on Rubin silicon and deployed on Jetson hardware, switching to a competitor's platform means rebuilding the inference pipeline and re-engineering firmware from scratch. That switching cost becomes structural, compounding with every product generation.

Noetra's technical roadmap escalates across three capability tiers. Initial output targets a reasoning model optimized for Japanese-language tasks, scheduled for fiscal 2026. The second tier, around 2028, aims for a model that processes written text, visuals, video, and audio as combined inputs. The capstone, designated "Real-world Native AI," is engineered to control autonomous machines by 2030 1. Models at that level push into the trillions of parameters, workloads that only Vera Rubin-scale infrastructure can sustain.

Japan's manufacturing sector is already committing. Fanuc and Yaskawa in factory automation, Kawasaki Heavy Industries and Hitachi in heavy industry, Sony, NEC, and Fujitsu in electronics, SoftBank and Kubota in systems integration, and the robotics coalition AIRoA have all pledged to build on Cosmos 3. Nvidia released Cosmos 3 Edge during the visit, a variant engineered to execute directly on Jetson Thor silicon inside machines 3. Toyota, which announced at CES in January 2025 that its next vehicle generation would adopt Nvidia's Drive platform, is extending that relationship into factory simulation and vehicle software 1.

The engine behind the spending is demographic. Japan's labor pool is contracting, and the government has set a goal of distributing 10 million robots with AI capabilities across 18 industrial categories before 2040. Public and private funding committed to the effort totals $65 billion 1. Prime Minister Sanae Takaichi and trade minister Ryosei Akazawa appeared alongside Huang at the July 16 launch, signaling that AI and semiconductor development anchor a growth strategy targeting ¥370 trillion, approximately $2.3 trillion, in combined investment by 2040 1.

Noetra exists because Tokyo judged dependence on foreign AI for its industrial base an unacceptable strategic risk. The logic holds: a country whose manufacturing data is a national asset cannot route that data through infrastructure controlled by another government. But the sovereignty Japan secures at the model layer is purchased with dependence at the silicon layer. Tokyo will own its data, its models, and its training pipeline. It will not own the processors underneath any of it.

Nvidia's position is generational. A nation that commits its full AI stack to a single vendor creates a relationship that tightens with every hardware cycle. New models require compatible training infrastructure. New robot platforms require compatible edge silicon. Three decades ago, Sega placed $5 million into a company on the verge of collapse. That capital kept Nvidia alive 1. The dependency has inverted. Japan's sovereign AI ambition now rests on the company that once relied on Japan to survive.