Nvidia Forges a Physical AI Blueprint in Japan, From Robot Chips to Supercomputers

Jensen Huang unveils Japan-wide AI platform FRONTia with 27,500 Rubin GPUs, edge modules for robotics, and partnerships with Toyota, Mizuho, and Riken for physical AI.

Japan has become Nvidia’s proving ground for a future that extends far beyond the data centre. During a whirlwind visit to Tokyo this week, chief executive Jensen Huang unveiled a series of partnerships and products that collectively position the chipmaker as the architect of a nationwide artificial intelligence infrastructure — one that will span autonomous factories, smart cities, financial risk systems and quantum-classical hybrid research.

The breadth of the announcements is striking. The Japanese government, working with Noetra Corp. and several industrial groups, is building a countrywide AI platform called FRONTia. At its core will be 27,500 of Nvidia’s next-generation Rubin graphics processors and 13,750 Vera central processing units, delivering a combined 140 megawatts of compute power. Huang confirmed that the Rubin chips, equipped with HBM4 memory, are now in production and shipping — a statement aimed squarely at recent speculation about possible delays.

Nvidia’s decade-old relationship with Toyota is also being reimagined. The carmaker will deploy the chipmaker’s hardware and software for its “Woven City” smart-city project in Shizuoka prefecture, where the technology will orchestrate traffic, automate municipal processes and connect factory floors. Toyota is simultaneously adopting Nvidia’s DRIVE AGX and DRIVE OS platforms for advanced driver-assistance systems, while using Omniverse and Isaac Sim to create digital twins of its production plants. What began as an automotive tie-up has morphed into a broad platform play for what Huang calls “physical AI”.

On the hardware front, Nvidia took the wraps off two new edge computing modules — the T3000 and T2000 — at the AI Summit Japan. Built on the Thor architecture with Blackwell technology, the modules are designed to bring humanoid and autonomous robots to the mass market. Early adopters include Amazon Robotics and Boston Dynamics, while partners UBTech and Agile Robots have benefited from a memory optimisation that cuts storage requirements by as much as 15 gigabytes. The modules are slated to ship in the first quarter of 2027.

Simultaneously, the company introduced the Cosmos-3 edge model, a 4-billion-parameter neural network tailored for Nvidia’s Jetson platform and edge graphics processors. It is engineered to handle image recognition and robot control with far less computational overhead than previous versions. The Cosmos Coalition — an industrial consortium that now includes FANUC, Fujitsu, Kawasaki Heavy Industries and SoftBank — will deploy the technology in industrial robotics and shipbuilding.

Japan’s financial sector is also plugging in. Mizuho Financial Group is building AI factories with Nvidia for risk management and fraud detection, using confidential computing techniques to protect sensitive data. In academic research, the Riken Institute has brought online the Rikyu supercomputer, powered by 1,600 Blackwell GPUs, alongside the ROQUO quantum-HPC system with 540 Blackwell chips. Mitsubishi Chemical and Keio University are already using both systems for molecular analysis.

The stock market reaction has been measured. After closing at €185.24 on Wednesday, Nvidia’s shares slipped 0.95% to €183.48 on Thursday as investors digested the breadth of the news. The stock remains 8.5% below its May record of €202.50, but is up roughly 15% year to date and 25.8% over the past twelve months. The 14-day relative strength index stands at 55.4 on Thursday, down from 57.6 the previous day, suggesting neutral momentum rather than overheating. The annualised 30-day volatility of 37.75% underscores how quickly sentiment can shift in the chip space.

Away from Japan, Nvidia is navigating a carefully calibrated reopening of the China market. The US Commerce Department confirmed on July 14 that shipments of the H200 AI chip — a previous-generation product that lags the Blackwell line — have begun to Chinese customers, with each buyer undergoing an individual national security review. The move offers Nvidia a toehold in a market that has been largely closed to its most advanced silicon.

On a lighter note, Huang used a fan event in Tokyo’s Akihabara district to acknowledge a debt of gratitude. He thanked former Sega president Shoichiro Irimajiri for a $5 million investment in the late 1990s that, according to Huang, pulled Nvidia back from the brink of bankruptcy when GPU development hit a critical funding gap.

The Tokyo announcements, taken together, paint a picture of a company that is no longer merely selling chips. Nvidia is stitching together an operating system for the physical world — and Japan has become the first full-scale test bed.