SoftBank Group Corp. will soon be the first to build a supercomputer powered by Nvidia’s Blackwell GPU, in a bid to meet increasing demand in Japan, a country that is looking to get a foothold in the artificial intelligence race.
SoftBank’s telecom division plans to build the most powerful AI supercomputer to support numerous local services within Japan, according to both companies’ statements on Wednesday, the machine will be powered by Nvidia’s DGX B200 unified AI platform, which combines CPU with AI accelerator chips. The next supercomputer would also feature Nvidia’s more advanced Grace Blackwell chips.
Nvidia’s AI acceleration chips have become the most sought-after commodity among global tech giants that use the chips to develop and operate their AI models.
The announcement has pointed out that SoftBank may have secured a front-row seat for Nvidia’s new chips, as the new Blackwell lineup has only been announced earlier this year, but manufacturing has been lackluster. Nvidia CEO Jensen Huang said that the supplies should be more readily available once the production has gone full steam.
At Nvidia’s Tokyo AI summit, Huang said that the company has been going back and forth around the planet to promote what it called the new industrial revolution. The event aims to promote the utilization of AI systems in various nations and reduce Nvidia’s reliance on a few customers in the U.S.
Softbank’s effort would create an AI grid that ran across Japan, Huang remarked.
AI radio access networks or AI-RANs, will be a superior option for remote robotics, autonomous vehicle support, and powering other services, it would also consume less electricity, Huang added.
Softbank’s telecom unit, which operates Japan’s third largest mobile network, will begin testing the new network with its partners Fujitsu Ltd and IBM Corps.’s Red Hat.
