In an AI race currently dominated by companies from the United States, like OpenAI and Anthropic, and China’s Alibaba Group, Japan appears to be laying the groundwork to build and run its own large-scale AI systems locally.
A new report claims that Japanese investment group SoftBank has set up a new company with partners including NEC and Honda Motor to “build a foundation model for ‘physical AI,’ which autonomously controls robots and machines, through a public-private partnership.”
Sony Group, the three megabanks, MUFG Bank, Sumitomo Mitsui Banking Corp., Mizuho Bank, as well as Nippon Steel and Kobe Steel, have reportedly invested in the venture. Tokyo-based AI developer Preferred Networks is also expected to collaborate on building the model.
David Adubiina
A report also claims the Japanese government is considering backing the company. Through a program run by the New Energy and Industrial Technology Development Organisation under the economy ministry, up to ¥1 trillion (about $6.2 billion) is being earmarked over five years to support domestic AI development, and this new company is expected to apply.
The motivation is partly competitive and partly strategic. As AI begins to handle sensitive industrial and operational data, depending on foreign foundation models is starting to look like a risk.
That thinking is now materialising into action. SoftBank Corp. has reportedly set up a new company dedicated to developing what’s being described as a domestic foundation model for “physical AI.”
The company aims to develop a roughly one-trillion-parameter AI model by the end of the decade, capable of processing not just text, but images, video, and audio, a multimodal system that can sit at the heart of factories, robots, and vehicles.
By fiscal 2030, the goal is for this AI to actively enable collaboration between humans and machines on manufacturing floors across Japan.
To support this, SoftBank is reportedly converting a former Sharp LCD factory in Sakai, Osaka, into a domestic AI data centre packed with advanced GPUs. The idea is to ensure that data used to train and run the model never has to leave Japan, reducing the risk of sensitive industrial information leaking overseas. That data sovereignty angle is as important as the AI itself.
Damilare Odedina
