SoftBank Launches AI Computing Platform
Featuring Liquid-Cooled NVIDIA GB200 NVL72

SoftBank Corp. ("SoftBank") today announced the launch of an AI computing platform powered by the NVIDIA GB200 NVL72 on December 22, 2025, marking a major milestone in the enhancement of Japan's domestic AI infrastructure. The new platform will be used to provide dedicated accelerated computing resource services powered by NVIDIA Graphics Processing Units (GPUs) for customers and to support the development and commercialization of "Sarashina," a Japan-made large language model (LLM) developed by SB Intuitions Corp.

The NVIDIA GB200 NVL72 is a rack-scale AI computing system that integrates 36 NVIDIA Grace Central Processing Units (CPUs) and 72 NVIDIA Blackwell GPUs in a single rack. SoftBank's newly deployed platform comprises 1,224 NVIDIA Blackwell GPUs and is designed for future expansion to more than 4,000 GPUs. When fully scaled, it is expected to become one of the largest computing systems in Japan, delivering 10.6 ExaFLOPS^*1 of processing performance. To ensure optimal performance in high-density AI workloads, the platform adopts a direct-to-chip liquid-cooling architecture that cools GPUs and CPUs directly. This approach efficiently dissipates heat generated by high-performance AI chips, enabling stable, sustained operation while achieving industry-leading power efficiency.

SoftBank is committed to building different types of social infrastructure to enable a society that co-exists with AI. The construction of SoftBank's AI computing platform was certified for supply assurance by Japan's Ministry of Economy, Trade and Industry (METI) as a "Cloud Program," which designates important materials based on its Economic Security Promotion Act^*2. By providing computational resources that can be utilized by a wide range of Generative AI developers, SoftBank will contribute to the establishment of development infrastructure and strengthen the service provision framework in Japan. In addition to utilizing the AI computing platform among its own group companies, SoftBank plans to provide the infrastructure as a service to companies and research institutions in Japan.

[Notes]

1. *1
   An exaflop is a measure of performance for a supercomputer that can calculate at least 10¹⁸ floating point operations per second.
2. *2
   More details are available in SoftBank Corp.'s press release dated May 10, 2024 (in Japanese).

• SoftBank, the SoftBank name and logo are registered trademarks or trademarks of SoftBank Group Corp. in Japan and other countries.
• Other company, product and service names in this press release are registered trademarks or trademarks of the respective companies.