Nvidia is building, together with the Japanese corporate consortium Noetra and the government in Tokyo, what it claims to be the world’s first state-funded AI factory. The facility combines 27,500 Rubin graphics processors and 13,750 Vera processors with a total power output of 140 megawatts. Tokyo is investing up to six billion dollars over five years in models for industrial robots.
Noetra builds data center with thousands of Rubin chips
The operator of the facility is the newly established consortium Noetra, which, according to Nvidia’s announcement, is backed by the companies SoftBank, Sony, NEC, and Honda, supported by 44 other companies and organizations. The Japanese Ministry of Economy, Trade and Industry (METI) is financing the project through the state funding agency NEDO, scientifically accompanied by the research institute AIST. For the current fiscal year, approximately 2.4 billion dollars are allocated, with up to six billion dollars over the entire duration until 2030 from so-called GX transformation bonds of the state.
Technically, the factory is based on 382 racks of the Vera Rubin NVL72 type, each of which combines 72 graphics processors and 36 main processors. In total, this results in 27,500 Rubin graphics processors and 13,750 Vera processors, interconnected via Nvidia’s Spectrum-X Ethernet technology and an NVLink connection with 260 terabytes of data throughput per second per rack. The contract for the project was awarded as early as the end of June 2026, with the construction phase expected to last until 2030. The racks are based on Nvidia’s reference architecture DSX, which is designed for data centers with particularly high power density and has primarily been used in the company for large cloud customers.
Japan aims for market leadership in AI robots
The new computing infrastructure is intended to train multimodal foundational models for industrial robots, digital twins - virtual representations of real factory facilities - and autonomous manufacturing systems. It is part of the state FRONTia program, which, according to the ministry, aims to promote cross-industry AI applications in manufacturing, logistics, healthcare, and telecommunications. Nvidia CEO Jensen Huang stated that Japan invented modern manufacturing and is now building the AI factories for the next industrial revolution. METI Minister Ryosei Akazawa emphasized that the initiative aims to contribute to solving global societal challenges.
The background is Japan’s robotics strategy presented in March: The country aims to capture a 30 percent share of the global AI robotics market, which is estimated to be around 133 billion dollars by 2040. Ten million AI-supported robots are expected to be in use across eighteen industries by then. The Japanese government also justifies the initiative demographically: By 2040, the working population is expected to shrink by twelve million people. According to the FRONTia program, initial pilot applications are to be tested in individual factories before the end of the construction phase.
Analysts estimate costs of billion-dollar hardware
As Tom’s Hardware reports, citing market data, the 382 server racks alone cost between 1.9 and 2.7 billion dollars, based on list prices of five to seven million dollars per rack. According to the report, the investment bank Morgan Stanley estimates the GPU silicon area alone at around 1.5 billion dollars at a unit price of about 55,000 dollars per graphics processor - unverified independently. These calculations come from external analysts and not from Nvidia or Noetra themselves, who have not published their own figures on hardware costs.
According to Tech Times, the initiative differs from previous Japanese AI initiatives in that the state has issued the infrastructure for the first time itself, rather than leaving it to corporations or universities. Whether the Rubin chips will be available as planned also depends on Nvidia’s mass production, which, according to company statements, will not start until the second half of 2026. Delays in chip manufacturing would directly affect the Japanese timeline until 2030 and impact the economic viability of the billion-dollar investment.
It will be crucial whether Japanese industrial companies actually share their production data for the FRONTia program. Factory operators traditionally treat sensor and operational data as trade secrets, but physical AI models can only learn from exactly such data. If access remains restricted, the expensive Rubin clusters risk missing their actual goal: independent Japanese foundational models for robotics instead of mere hardware supply.
