Japanese technology and investment giant SoftBank Group is making a monumental push into the generative AI arena, announcing plans to invest approximately 150 billion yen ($960 million) to build one of Japan’s most powerful AI supercomputers. The initiative is aimed at developing a proprietary, world-class large language model (LLM), marking a strategic pivot for the company from primarily an investor to a direct competitor in the AI arms race.
The supercomputer, which will be built at a new data center in Hokkaido, is expected to be fully operational by 2025. It will be powered by Nvidia’s latest and most advanced AI chips, likely from the Blackwell architecture, enabling SoftBank to train a homegrown model with an ambitious target of 1 trillion parameters. This would place SoftBank’s model in the same league as top-tier systems from OpenAI, Google, and Anthropic.
This move is a cornerstone of CEO Masayoshi Son’s vision for an era of “Artificial General Intelligence” (AGI) and his desire to create “AI-iPhone” devices—next-generation hardware deeply integrated with powerful, native AI capabilities. By developing its own foundational model, SoftBank aims to reduce its reliance on American AI providers and establish a leading position in Japan’s burgeoning sovereign AI ecosystem.
The project is being spearheaded by SoftBank’s AI division, which was established last year. The significant capital investment underscores the company’s commitment to not just participate in the AI revolution, but to become a foundational player. This development positions SoftBank to offer a comprehensive suite of AI services, from cloud computing to generative AI applications, intensifying competition on a global scale and signaling Japan’s serious ambition in the AI landscape.
English 
Deutsch