NVIDIA used its founder’s Tokyo schedule to demonstrate how the company intends to turn Japan’s existing manufacturing and component strengths into a production engine for physical AI systems. Rather than announcing new chips or robots, the emphasis fell on open-model developer events and closed supply-chain meetings that lock in upstream partners.
The most visible signal came at Studio Koku inside the Happo-en garden, where Huang made an unannounced appearance at the Build-a-Claw session. Developers had spent the day training open models on NVIDIA’s platform to control simple grippers. Huang handed two autographed DGX Spark personal AI supercomputers to contest winners, framing the devices as tools for individuals to run their own agents.
Later the same day Huang told reporters that Japan’s historical edge in precision and large-scale manufacturing now has a new variable. “You can combine the two technologies and create robotics,” he said, adding that AI can run continuously to augment existing workers and lift national productivity.
A separate dinner in Tokyo’s Kanda district brought together more than 30 executives from 16 companies that already dominate global semiconductor equipment, materials and electronic components. Participants included Tokyo Electron, ADVANTEST Corporation, KYOCERA Corporation, Mitsubishi Electric, Murata Manufacturing, Panasonic, Renesas Electronics Corporation, Sumitomo Electric Industries, TAIYO YUDEN, TDK Corporation, Kioxia Corporation, Mitsui, Asahi Kasei, Nittobo, Shin-Etsu Chemical and Shibaura.
The gathering underscored NVIDIA’s recognition that scaling physical AI requires more than models; it needs reliable, high-volume supply of sensors, memory, substrates and test equipment. Japan’s firms already sit at multiple points in that chain.
Huang’s itinerary also included a CEO-level discussion at Tonkatsu Fumizen focused on inserting physical AI directly into Japanese factories. The session examined how existing mechatronics know-how can be paired with NVIDIA’s simulation, training and inference stack to shorten deployment cycles.
For investors and competitors, the pattern is clear: NVIDIA is not trying to build every robot or factory line itself. Instead it is using Japan’s industrial base to create reference implementations and component standards that other regions can adopt, while keeping the software and model layer inside its own ecosystem.
The immediate commercial implication is that Japanese hardware companies gain early access to NVIDIA’s physical AI roadmap in exchange for design wins and data feedback loops. Builders outside Japan receive working examples of how open models plus DGX Spark-class hardware can move from bench to production line without starting from scratch.