Nvidia Bets on Japan to Lead 'Physical AI'

Jensen Huang arrived in Tokyo this week with a message Japanese officials were glad to hear. At a July 16 event linked to Japan's Ministry of Economy, Trade and Industry, the Nvidia chief executive cast his company as the engine behind the country's bid to build artificial intelligence it can call its own. He kept returning to one phrase, "physical AI," the software that lets machines perceive, move and act in the real world, and he argued that no country is better placed to lead it than Japan.

This was Huang's first official visit to Japan since October 2025, and it came with signed partnerships rather than the usual talk of potential. It also followed grumbling in the Japanese press about "Japan passing," the worry that Nvidia had been lavishing attention on rivals elsewhere in Asia while Tokyo waited.

Open models for a homegrown stack

The anchor announcement, made on July 15, was a roster of Japanese institutions building their own systems on Nvidia's Nemotron open models. The Institute of Science Tokyo trained its Swallow foundation models on Nemotron datasets for financial-document work. SoftBank's SB Intuitions built the Sarashina series, and its Sarashina3 mini has already been picked up by Japan's Digital Agency. Stockmark, avatarin, ENEOS, NTT DATA, Hitachi and Sakana AI each described projects of their own, from airport service avatars running on Nvidia Jetson to catalyst research driven by AI agents.

Huang's framing was blunt. "Every nation and every company should own and control its intelligence infrastructure," he said. Open weights, in that telling, are not a threat to Nvidia's business but a way to sell more of the hardware those models run on.

The Noetra project and a plan for ten million robots

Behind the corporate deals sits a national wager. Japan has committed to a sovereign AI programme built around Noetra, a consortium that includes SoftBank, NEC, Sony and Honda and is charged with developing a domestically produced multimodal foundation model. METI has pledged up to one trillion yen, roughly 6.2 billion dollars, over five years to back the effort, according to Nikkei Asia.

The ambition stretches well past chatbots. The government wants ten million AI-equipped robots deployed across 18 sectors, including restaurants, food manufacturing and medical care, by 2040. For a country with a shrinking and ageing workforce, the robots are less a novelty than a staffing plan.

Physical AI on the ground

Some of that vision is already being wired up. Nvidia said it would supply technology for traffic management at Woven City, the experimental town Toyota is running in Shizuoka prefecture, and it named further tie-ups with Kawasaki Heavy and Mitsubishi Heavy. Hitachi's plan to pair Nemotron with Nvidia's Cosmos platform points the same way, connecting back-office systems to the machines on a factory floor.

A pattern, not a one-off

Japan's push is part of a wider scramble by governments to keep their most sensitive AI work at home rather than renting it from a handful of American clouds. China is making the same case in louder terms, with Xi Jinping set to headline its flagship AI summit for the first time. Nvidia, sitting in the middle, is content to sell the shovels to every side.

The robot numbers deserve a note of caution. Grand deployment targets have a way of slipping, and the distance between a slick demonstration and a machine that reliably works a hospital ward is still wide, as we have written before. What changed this week is less the timeline than the money and the names now attached to it.