SoftBank Completes Major AI Computing Infrastructure Expansion with 4,000 NVIDIA GPUs

Tokyo, Japan—SoftBank Corp (9434.T) completed the installation of approximately 4,000 NVIDIA (NVDA.O) Hopper GPUs, expanding its artificial intelligence computing infrastructure to 4.7 exaflops processing capacity.

The installation increases SoftBank's total GPU count to 6,000 units from its previous base of 2,000 NVIDIA Ampere GPUs installed in September 2023.

SoftBank plans to deploy NVIDIA DGX B200 systems in DGX SuperPOD configurations between fiscal 2024 and 2025, targeting 10,000 GPU units.

The expansion represents a JPY 150 billion ($1 billion) investment, with the Japanese government providing up to JPY 42.1 billion in subsidies through its Economic Security Promotion Act certification.

Upon completion, the company's AI computing capacity will reach 25.7 exaflops, a thirty-sevenfold increase from its initial 0.7 exaflops capability.

SB Intuitions, a wholly-owned SoftBank subsidiary, will utilize the infrastructure to develop a 390 billion-parameter multimodal large language model by the end of fiscal 2024.

The infrastructure expansion follows SoftBank's May 2023 announcement of its next-generation digital infrastructure vision, which aims to support Japan's AI development capabilities.

The company plans to offer excess computing capacity to Japanese enterprises and research institutions, entering the growing AI infrastructure service market.

Japanese government data shows domestic AI computing infrastructure investment reached JPY 323 billion in fiscal 2023, up 45% from the previous year.