Softbank to Invest Heavily in Supercomputing for GenAI Push

LinkedIn Google+ Pinterest Tumblr

Softbank, the Japanese telecommunications giant, is gearing up to bolster its supercomputing capabilities in a bid to support its ambitious generative AI (GenAI) strategy. As reported by Nikkei, Softbank plans to allocate JPY150 billion ($960 million) over the next two years towards upgrading its computing infrastructure, a significant leap from the JPY20 billion invested last year.

This substantial investment will largely involve acquiring graphics processing units (GPUs) from Nvidia, a key player in the AI hardware market. These GPUs will not only power Softbank’s in-house large language models (LLMs) but will also be made available to corporate clients in need of supercomputing resources.

The move underscores Softbank’s commitment to seizing opportunities in the rapidly expanding AI landscape, particularly the demand for AI capable of conversing in Japanese. Last year, Softbank established a dedicated division focused on LLM development and GenAI services, subsequently launching it as SB Intuitions, a fully-fledged subsidiary, in August.

SB Intuitions has made significant strides, having deployed its computing platform by October and currently working on its inaugural homegrown LLM. This model, slated for completion this year, will boast an impressive 350 billion parameters, surpassing OpenAI’s GPT-3 in scale.

Interestingly, Softbank’s reliance on Nvidia for its AI computing platform aligns with their established partnership. Nvidia’s hardware, including the Nvidia DGX SuperPOD AI supercomputer equipped with over 2,000 Nvidia Tensor Core GPUs, forms the backbone of Softbank’s infrastructure.

This strategic investment comes amidst heightened competition in the AI sector. OpenAI’s recent expansion into Japan with GPT-4, capable of Japanese language processing, intensifies the pressure on Softbank to provide a compelling alternative.

Nvidia, in pursuit of dominance in the GenAI market, continues to innovate. Its upcoming DGX GH200 supercomputer, featuring the powerful Grace Hopper GPUs tailored for GenAI tasks, is poised to attract companies looking to develop their own LLMs.

While specifics on pricing remain undisclosed, the considerable investments by Softbank and others underscore the fierce competition and potential rewards in the burgeoning GenAI landscape. This move signifies Softbank’s determination to carve out a substantial presence in this rapidly evolving technological frontier.

Write A Comment