Jensen Huang's Tokyo Trip: A Turning Point for Japan's Physical AI Strategy

From Sega’s $5 Billion to Trillions of Yen: Japan’s Thirty-Year Cycle with NVIDIA

In the 1990s, a nearly bankrupt graphics chip company received a life-saving $5 million investment from Japanese game company Sega; thirty years later, that company is worth over $3 trillion, and its CEO, Jensen Huang, was in Tokyo discussing with Japan’s political and business elite how to use AI to reshape manufacturing. This visit from July 15 to 16 left behind not just simple business contracts, but a roadmap for Japan’s industrial competitiveness over the next two decades.

Huang’s itinerary was packed and purpose-driven: lunch with industrial giants such as Toyota, Fanuc, and Yaskawa Electric; raising glasses with supply chain executives at a Kanda izakaya; and jointly announcing with Japan’s Minister of Economy, Trade and Industry, Toshimitsu Akazawa, “the world’s first national-level AI infrastructure”—a data center powered by NVIDIA’s Vera Rubin chips. The Japanese government has committed up to 1 trillion yen (approximately $62 billion) over five years to promote domestic physical AI foundation models.

This is not merely a business collaboration; it is a national-level industrial strategy shift. Japan is positioning “physical AI”—AI systems capable of operating robots, vehicles, and factory equipment—as the core lever for its manufacturing revival.

The Strategic Logic Behind the Three Agreements

The core outcomes of Huang’s trip are reflected in three dimensions, each targeting the pain points and ambitions of Japan’s industrial system.

1. Noetra: The Hardware Foundation of Sovereign AI

Noetra, jointly established by 44 domestic companies including SoftBank, Sony, NEC, and Honda, will operate a 140-megawatt AI data center. According to the plan, the facility will advance in three phases: launching an inference model with strong Japanese-language capabilities in 2026; upgrading to a multimodal model in 2028; and achieving “reality-native AI running on robots” by 2030.

This timeline mirrors Japan’s anxiety over sovereign AI. In the field of general-purpose large language models, the United States and China have already taken a commanding lead, while Japan has only a handful of LLMs. However, Japan believes that in the niche of “physical world AI,” its accumulated manufacturing data and factory automation experience could form a barrier to entry. The essence of Noetra is: train Japan’s AI with American chips, keeping data sovereignty and model ownership within the country.

2. Robot Alliance: Cosmos Lands in Japan

NVIDIA has released the Cosmos 3 Edge model, a physical AI iteration optimized for edge devices that can run directly on Jetson Thor chips. More than ten companies and research institutions—including Fanuc, Yaskawa, Kawasaki Heavy Industries, Fujitsu, Hitachi, NEC, Sony, SoftBank, and Kubota—have announced the development of robot systems based on Cosmos. These companies are not only cooperating to test shared control systems, but Honda R&D and Omron have already begun actual development using this tool.Japan's robotics industry holds a globally leading share in industrial robots, but has long relied on external solutions for the AI software layer. Cosmos' open model strategy provides Japanese companies with a relatively low-cost, standardized AI integration path. More importantly, NVIDIA has optimized the model to run locally on robots, reducing the latency and dependency on cloud computing—crucial for real-time control on production lines.

3. Toyota: Comprehensive AI transformation from driving to manufacturing

Toyota has long been a customer of the NVIDIA Drive platform, but this collaboration extends to the manufacturing process. Toyota's next-generation vehicles will feature advanced driver-assistance systems (ADAS), while its production line design will leverage NVIDIA's simulation technology. Vehicle software and traffic analysis systems will also be driven by NVIDIA chips. Toyota adopts a conservative, incremental approach, differing from Waymo and Tesla's direct L4/L5 solutions, but with broader coverage—digitizing the entire chain from production to operations.

Urgency Behind the Numbers

Japan's Ministry of Economy, Trade and Industry released the "AI Robot Strategy" in March this year, setting ambitious goals: deploying 10 million AI-driven robots across 18 industries by 2040; total public and private investment reaching $6.5 billion; and domestic companies capturing over 30% of the global AI robot market (currently around 10%).

Behind these numbers lie existential pressures from Japan's demographic structure. Over 29% of Japan's population is aged 65 or older, and the labor force continues to shrink. Automation is no longer a choice but a necessity. Japanese political circles have begun serious discussions on "robot taxes" and "AI national income," but the more urgent task is to make robots truly "smart"—which is why Tokyo is willing to invest 1 trillion yen in building its own foundation models.

Notably, the video call between Jensen Huang and Prime Minister Yoshihide Suga, along with the personal endorsement by the Minister of Economy, Trade and Industry, indicates that this strategy has risen to the highest national decision-making level. The Suga administration has listed AI and semiconductors as core elements of 17 strategic growth areas, aiming to achieve 370 trillion yen ($2.3 trillion) in public and private investment by 2040. Noetra's AI factory is the first milestone in this blueprint.

Walking a Tightrope Between Dependence and Independence

However, an undeniable contradiction is that Japan pursues technological sovereignty, but this effort currently relies entirely on American chips. Noetra's AI factory will use NVIDIA's GPUs, and the Cosmos model runs on NVIDIA's Jetson platform. Japan's domestic chip manufacturers, such as Rapidus (Japan's next-generation semiconductor foundry project), still need several years to achieve mass production of 2nm process nodes.From a broader perspective, Japan’s bet on physical AI is essentially a counterattack against China’s competition in manufacturing AI applications. China possesses a more extensive factory data pool, more aggressive AI deployment policies, and a more complete supply chain. Japan is attempting to leverage its own advantages in robotics hardware and precision manufacturing expertise to gain a first-mover position in defining the standards for physical AI. NVIDIA, meanwhile, plays the role of an arms dealer—supplying chips to both sides while ensuring that Japan’s camp remains deeply embedded in its ecosystem.

During his speech in Tokyo, Jensen Huang said: “Japan invented modern manufacturing. Now it has the opportunity to reinvent it for the age of intelligent industry.” This statement is both a tribute to history and a contract for the future. But the real test lies in whether Japan, as it frantically catches up in the AI software layer, can avoid turning its industrial IoT into a “colony” of American chips while advancing its hardware to higher levels.

The turning point for Japan’s physical AI strategy has arrived. The next five years will determine whether it can break free from the historical fate of being a “manufacturing powerhouse but a software lightweight.”