SoftBank will build the most powerful AI supercomputer in Japan using Nvidia DGX B200 accelerators from the Blackwell family

This week, Nvidia held an event with speeches from its founder Jensen Huang and SoftBank CEO Masayoshi Son. The latter corporation will be among the first to use Nvidia DGX B200 generation Blackwell computing accelerators in Japan’s most powerful supercomputer focused on solving artificial intelligence problems. At the same time, SoftBank will launch an AI-RAN generation network in Japan to control autopilot and robots.

Image source: NVIDIA

As Bloomberg explains, the Japanese SoftBank will be the first company to have the opportunity to build a supercomputer based on the latest Nvidia DGX B200 accelerators, which combine 8 Blackwell generation chips and 2 Intel Xeon Platinum 8570 processors in one case. In the future, SoftBank will switch to using Grace Blackwell – more modern accelerators implying the presence of Arm-compatible Grace processors of Nvidia’s own design. SoftBank Corp., part of the SoftBank Group, is the third largest cellular operator in Japan.

The close relationship between Nvidia and SoftBank was established back in the last decade, as until early 2019 the latter owned 4.9% of the former’s shares. SoftBank founder Masayoshi Son said he is ready to make serious bets in the artificial intelligence segment. Nvidia solutions will also be used by telecom operator SoftBank Corp. when creating a new generation network (AI-RAN), which will be able to transmit information using the 5G standard and simultaneously manage client devices using artificial intelligence. They can be vehicles with autopilot and various kinds of robots. The pilot project will eventually allow similar networks to be scaled throughout Japan. Next-generation networks are not only better suited for managing complex devices, but also consume less power. Fujitsu and Red Hat will be SoftBank’s AI-RAN testing partners.