Softbank to spend $960m on computing infrastructure to develop an AI services business

Softbank says it will increase its spending on computing infrastructure from last year’s $130 million to  $960 million during 2024-25, to give the company a leading capability in AI, reports the Nikkei.

Last year SoftBank set up a new division, now promoted to be a subsidiary, called SB Intuitions.

SB Intuitions is said to be developing a large language model (LLM) with 390 billion parameters for delivery in FY 2024. Open-AI’s GPT-3 has  175 billion parameters.

Parameters means variables i.e. the factors influencing an inference from the LLM. The more variables put into the mix, the greater the accuracy of the inference – or that’s the theory.

Next year SB Intuitions is said to be starting an LLM with 1 trillion parameters  – OpenAI’s GPT-4 is said to have 100 trillion parameters.

In addition, SoftBank intends to build datacentres around Japan which will be used to deliver services based on in-house developed LLMs to customers