Nvidia has built its empire on the chips that power generative AI, and that engine is still on the rise. The company posted a record $81.6 billion in quarterly revenue, an 85 percent increase year over year. But with a $5 trillion valuation factoring in years of growth, CEO Jensen Huang is under pressure to prove that expansion can extend beyond the US data center boom. His latest presentation focuses on Japan, where the government and major industrial players are preparing to deploy artificial intelligence in robots and factories.
During a two-day visit to Tokyo last week, Hwang rallied Japan’s manufacturing giants, robotics innovators and software developers behind “physical AI,” the technology that allows robots and other machines to perceive their surroundings, make decisions and act in the real world. Hwang claims that Japan has a natural advantage for such a transformation. “Japan has historically been very good at microfabrication and very large-scale manufacturing, but now we have artificial intelligence, where you can combine the two technologies and create robots,” he told reporters during a July 15 event.
Japan has strong incentives to embrace this partnership. It boasts world-class industrial equipment but faces a worsening labor shortage. Prime Minister Sanae Takaishi, who has made semiconductors and artificial intelligence a key component of her growth agenda, is seeking to pair that hardware power with more advanced software. Japan accounts for about 70 percent of the global industrial robot market, but just over 10 percent of the service robot market, according to government data. Its updated strategy targets more than 30% of the emerging AI robotics market by 2040, representing $133 billion in business.
During his visit to Tokyo, he presented his vision to the country’s industrial elite. At a bar in Tokyo’s Kanda district, Huang gathered more than 30 executives from 16 major companies — including Tokyo Electron, Panasonic, and Mitsubishi Electric, among others — to discuss how Japan’s semiconductor supply chain can support AI-led expansion. During lunch, he also met with executives from Fujitsu, Kawasaki Heavy Industries, Fanuc and Yaskawa. According to Nvidia, the four companies are now developing physical AI systems on their platform.
Tokyo’s reception contrasts with Nvidia’s experience in China. Huang He visited the country in January He returned in May as part of President Trump’s delegation, trying to rebuild Nvidia’s standing in a market squeezed by US export controls and Beijing’s support for domestic chip manufacturers.
In comparison, Japan puts government support behind infrastructure built around Nvidia’s technology. On the second day of his trip, Huang appeared alongside Japanese Industry Minister Ryusei Akazawa to launch the country’s new government-backed physical AI initiative. AAs part of that start, Huang announced that Nvidia is teaming up with Noetra Corpa Japanese AI consortium backed by Sony, SoftBank, Honda and nearly 40 other companies, to build what the chipmaker calls “the world’s first national AI infrastructure for physical AI.”
“Japan must own, improve, secure and deploy Japanese AI,” Huang said during his keynote speech. “After 15 years of work, physical AI is here, it is the foundation of the next industrial revolution, and it should be made in Japan.”
The project focuses on an “AI Factory” powered by 13,750 Nvidia Vera CPUs and 27,500 next-generation Rubin GPUs. The facility is expected to provide 140 megawatts of data center capacity, and will provide the computing required to train complex physical AI models. Construction is scheduled to begin in April 2027, with operations to begin in June 2028.
The infrastructure will serve as the computing backbone for FRONTia, a government project that supports multimodal models of physical Noetra AI and will lead the development of local models running on the network. According to Nnoitra’s roadmap, the group will prioritize understanding and reasoning Japanese before expanding into text, image, video and audio capabilities in 2028. By 2030, it aims to deploy “authentic, real-world AI” for robots and autonomous machines.
This week, Nvidia brought the same message to SIGGRAPH, the annual computer graphics conference currently underway In Los Angeles. During Monday KeywordHuang directly links Nvidia’s roots in computer graphics to its AI ambitions. “Whether it’s gaming, cinema, robotics or digital twins in factories, the goal is the same: to create virtual worlds that behave faithfully and realistically to the physical world,” he said in a pre-recorded video.
Before a robot can navigate its surroundings, it needs a model of how the physical world behaves. At SIGGRAPH, Nvidia released Cosmos 3 Edge, a compact version of its Cosmos World model designed to run directly on hardware inside robots, vehicles, and peripherals. By processing data locally rather than routing queries to a remote data center, edge models enable real-time decision making. Companies currently evaluating the framework include Agile Robots, Doosan Robotics, Siemens, and Skild AI.
However, not all training has to happen in a physical space. During the keynote, Mingyu Liu, vice president of Cosmos Lab at Nvidia, unveiled Cosmos-Dreams, a set of virtual testing environments for physical AI. In one demonstration, the software created an entire driving scene from a single video frame.
Liu said Cosmos was trained to interpret a scene, predict outcomes, simulate consequences and choose action. These capabilities can be applied to machines ranging from humanoid robots to mechanical arms and autonomous vehicles. “Each incarnation speaks a different language,” Liu said. “Our solution is to build a common vocabulary.”
