Nvidia’s Japan Deals Turn Sovereign AI Into a Silicon Dependency Test

Jensen Huang spent two days in Tokyo on July 15 and 16, then left with a stack of deals that put Nvidia at the center of Japan’s push to build AI it can call its own. A national AI factory, a coalition of the country’s biggest robotics and manufacturing names, and an extension of Nvidia’s automotive platform into Toyota’s next generation of vehicles all point the same direction. Japan wants sovereignty over its AI future, and for now it is building that sovereignty on American silicon.

The stop followed earlier visits to Taiwan and South Korea this year, according to TechCrunch. Three stops in a row suggest a pattern: Nvidia is becoming the default infrastructure layer for national AI ambitions across Asia. What makes Japan’s version notable is how explicit the tension is. As TechCrunch put it, the country doesn’t want to run its factories and robots on American or Chinese AI, yet every initiative announced in Tokyo runs on hardware that only one company currently supplies at the scale Japan needs.

A National AI Factory Built Around One Vendor’s Chips

The centerpiece is Noetra, a new entity that Japan’s Ministry of Economy, Trade and Industry (METI) is backing as part of its FRONTia project for multimodal foundation models aimed at AI robotics and physical AI, according to Nvidia’s own announcement. SoftBank, Sony, NEC, and Honda anchor the roughly 44 firms funding Noetra alongside Japan’s National Institute of Advanced Industrial Science and Technology (AIST). METI first floated the plan on July 1, when Minister Ryosei Akazawa unveiled an updated national robotics strategy; The Register reported at the time that Noetra would be majority-owned by that same core group, with Fujitsu and Rakuten weighing whether to join.

Nvidia will supply the hardware for that ambition directly. The companies plan to build a Vera Rubin AI factory packing 13,750 Nvidia Vera CPUs and 27,500 Rubin GPUs, delivering 140 megawatts of capacity on Nvidia’s DSX platform, a specification confirmed independently by both Nvidia’s newsroom and TechCrunch’s reporting. Tokyo has committed roughly 1 trillion yen, about $6.2 billion, over five years to the effort. “Japan invented modern manufacturing,” Huang said in Nvidia’s announcement. “Now, it is building the AI factories that will power the next industrial revolution.”

A Three-Stage Timeline, on Nvidia’s Clock

TechCrunch’s reporting lays out Noetra’s plan in three stages: a reasoning model with strong Japanese-language skills starting in fiscal 2026, an omni-modal version handling text, images, video, and audio by 2028, and a “real-world native AI” built to run robots by 2030, released to outside developers in phases. Every stage of that roadmap depends on the Vera Rubin factory actually shipping on schedule, which ties Japan’s model-development timeline directly to Nvidia’s chip production and allocation decisions.

A Robotics Coalition Built on Nvidia’s Cosmos Models

Alongside the factory news, Nvidia announced that Japan’s robotics and manufacturing leaders are joining its Cosmos coalition, an open effort to build frontier physical AI models. Nvidia named 22 companies intending to join, including FANUC, Yaskawa Electric, Kawasaki Heavy Industries, Fujitsu, Hitachi, Kubota, Honda R&D, NEC, Sony, SoftBank, and the robotics research group AIRoA. SiliconANGLE’s coverage of the announcement noted that Fujitsu is already building a shared control platform with FANUC, Yaskawa, and Kawasaki Heavy Industries, three companies that together account for a large share of Japan’s installed industrial robot base.

The technical piece underpinning the coalition is Cosmos 3 Edge, a 4-billion-parameter model built on Nvidia’s Nemotron architecture. Unlike cloud-dependent AI, it runs entirely on-device, on Nvidia RTX GPUs, DGX systems, or Jetson modules bolted onto factory-floor hardware, letting a robot perceive its surroundings and generate actions without a network connection. Nvidia says the model can be adapted to a new robot or environment in about a day. “The next frontier of AI is in the physical world, and this is a once-in-a-generation opportunity for Japan,” Huang said in the announcement.

Toyota’s Parallel Bet on the Same Platform

Toyota’s presence in Huang’s Tokyo itinerary extends a relationship that predates this trip. The automaker committed its next-generation vehicles to Nvidia’s Drive platform at CES in January 2025, and TechCrunch reports that the partnership has since grown into Nvidia-powered simulation for designing Toyota’s production lines, software running inside its vehicles, and systems that read road traffic. Toyota’s driver-assistance systems remain more conservative than Waymo’s or Tesla’s, still requiring a human driver, but the underlying compute stack is the same one Noetra and the Cosmos coalition are building on.

Sovereignty That Runs on Someone Else’s Silicon

Set side by side, the three announcements describe a full stack: Nvidia chips inside the national AI factory, Nvidia edge silicon running Cosmos 3 Edge on factory floors, and Nvidia’s Drive platform inside Toyota’s cars and plants. What Japan is actually building sovereignty over is the layer above that stack: Japanese-language reasoning models, foundation models trained on domestic manufacturer data, and governance held by a Japan-based, Japan-funded consortium. What it has not built sovereignty over is the compute underneath all of it, which still comes from a single American supplier.

That is not a uniquely Japanese problem. The Register noted that South Korea announced its own robotics-powerhouse plan the same week Japan updated its strategy, part of a regional pattern of national AI programs branded as sovereign while remaining dependent on the same GPU vendor. For enterprise and infrastructure teams evaluating any government’s or vendor’s sovereign AI claims, the practical test is not where a data center physically sits or which local entity holds equity in it. It is whether a real alternative exists if that GPU supplier changes pricing, allocation priorities, or export terms. Keeping model development and governance domestic while sourcing compute from Nvidia, which is what Japan is doing, is a reasonable hedge against some of that risk. It does not remove the underlying dependency; it just moves it up one layer of the stack.

Huang left Tokyo with commitments spanning that entire stack, from the chips in the data center to the silicon inside a welding robot’s controller. Japan left with a roadmap toward the AI independence it says it wants. Whether that independence arrives on schedule depends on a supply chain Japan does not control, no matter how many domestic firms sit on Noetra’s cap table.