Japan's problem is not ambition. It is arithmetic.
The country's population peaked in 2010 and has been falling since. Roughly 30% of Japanese people will be over 65 by 2030. Immigration policy has not filled the gap. The government's own projections put the shortfall in AI and robotics specialists at over 3 million people by 2040. When you cannot hire enough workers, you build them.
That is the context behind Jensen Huang's Tokyo visit this week. The deals he signed are large, but the numbers that explain them are the demographic ones.
At the centre of everything is Noetra, a new company that began operations on July 16. It is majority-owned by SoftBank, Sony, Honda, and NEC, with roughly 44 Japanese firms involved. The Japanese government is committing up to 1 trillion yen, about $6.2 billion, over five years to fund it. The goal is to build AI software that can run robots, not just answer questions or generate text. Construction on the data centre that will train these systems starts in April 2027, with operations expected in June 2028.
Nvidia's role is to supply the computing power. The planned facility will hold 27,500 of Nvidia's latest Rubin processors. Japan will own the software being built on top. The hardware comes from California.
This is the tension at the heart of the whole project. Japan wants independence from foreign AI systems, especially American and Chinese ones. But the chips running its sovereign AI factory are American. Huang's playbook, and he has now run it in South Korea and Taiwan in recent months, is to position Nvidia as the essential infrastructure layer beneath every country's domestic AI ambitions. It is working.
Beyond Noetra, the robotics partnerships matter enormously. Japan's top robot makers, Fanuc, Yaskawa, and Kawasaki, collectively account for over 40% of global industrial robot shipments. These are the yellow arms you see welding car frames in almost every major manufacturing country. They are extremely reliable. What they have not historically been is smart in the way AI enables. They follow programmed instructions. They do not adapt.
Nvidia's Cosmos platform is designed to change that. It gives robots the ability to learn from simulated environments before they are deployed on real factory floors, which dramatically reduces the cost and time of training a robot to do something new. Fanuc and Yaskawa are now building on this platform. Nvidia also unveiled a version of Cosmos that runs directly inside the machines themselves, rather than relying on a remote data centre.
Toyota adds a different dimension. It is using Nvidia's systems for car software, for traffic-reading tools, and now for simulating production lines before they are physically built. That last application is quietly significant for any manufacturer: you can model an entire assembly process in software, test it, and only then spend money on physical equipment.
The broader Japanese strategy, targeting 30% of the global AI robotics market by 2040 in a market Tokyo values at roughly $133 billion, is the kind of industrial ambition that tends to pull other countries into competitive response. South Korea announced its own robotics push within days of Japan's confirmation.
For business operators outside Japan, the relevant question is not whether to care about Japanese government policy. It is simpler than that. The companies that make the robots in your suppliers' factories, or the machines in your logistics chain, are now tying their software roadmaps to a specific AI platform. The choices being made in Tokyo this week will shape what capabilities are available, and at what cost, in industrial automation for years to come. Understanding which AI infrastructure underpins your suppliers is going to matter as much as understanding their hardware.