Japan plans 27,500 Nvidia Rubin chips and $2.4 billion for robot AI project

Japan is planning to buy 27,500 next-generation Rubin chips from Nvidia to build a homegrown foundational AI model for robots. Newly established Noetra, allocated ¥387.3 billion ($2.4 billion) from government coffers through March of next year, will oversee the project and construct a roughly 140-megawatt data center. Noetra’s data center is scheduled to come online in June 2028.

Dozens of companies including Sony Group, SoftBank, Toyota-backed Preferred Networks, NEC and Fujitsu are helping to set up and operate the entity. The venture will draw on engineers from those firms, which have already developed their own AI models such as SoftBank’s Sarashina LLM, Preferred Networks’ PLaMo and NEC’s cotomi.

Noetra was created to bring such fragmented AI efforts together, said Noetra President Hironobu Tamba, who previously led development of SoftBank’s large language model.

The venture plans to release an AI model by March next year, followed by regular updates. The longer-term goal is a model tailored for robotics applications within a few years, Tamba said. The initiative forms part of government efforts to lower reliance on foreign technology and bolster national security.

Minister of Economy, Trade and Industry Ryosei Akazawa said building a robust and vibrant AI ecosystem is important and will serve as a lifeline for resource-poor and disaster-prone Japan. Japan is home to some of the world’s largest industrial robot makers. The government aims to capture more than 30 percent of the estimated ¥60 trillion global robotics market by 2040.

Nvidia co-founder Jensen Huang, who was in Tokyo to promote applications that help robots and machines learn and adapt, said a generation of industrial craftsmen are retiring without passing on their skills and that AI can step in to transfer a master’s craft to the next generation.