The joint effort, showcased during the keynote at Nvidia’s AI Summit in Japan, aims to create what the companies described as a “new breed of infrastructure” that would enable operators to turn their traditional base stations into “AI revenue-producing assets.”
Subscribe today for free
Trials of a jointly developed AI-RAN solution in Kanagawa, Japan, saw SoftBank use the AI-RAN solution to achieve carrier-grade 5G performance while using the network’s excess capacity to run AI inference workloads concurrently.
The companies contend that traditional telecom networks only use one-third of their capacity and that operators could potentially monetise the remaining two-thirds for AI inference services.
Nvidia and SoftBank estimates suggest operators could earn roughly $5 in AI inference revenue from every $1 of capex it invests in new AI-RAN infrastructure.
The Japanese giant is working on its own AI-RAN solution dubbed AITRAS, which will power AI workloads on Nvidia’s upcoming GH200 Superchip while also supporting traditional RAN workloads, with Nvidia’s AI Aerial software providing stability on the stack. SoftBank claimed it can return up to 219% for every AI-RAN server it adds to its infrastructure.
“With our extremely powerful AI infrastructure and our new, distributed AI-RAN solution AITRAS that reinvents 5G networks for AI, we will accelerate innovation across the country and throughout the world,” said Junichi Miyakawa, president and CEO of SoftBank.
SoftBank said it plans to offer AITRAS reference kits to operators from 2025, before expanding the solution to make it available to operators globally from 2026.
“Shifting from single-purpose to multi-purpose AI-RAN networks can mean five times the revenue for every dollar of capex invested,” said Ronnie Vasishta, SVP of telecom at Nvidia. “SoftBank’s live field trial marks a huge step toward AI-RAN commercialisation with the validation of technology feasibility, performance and economics.”
SoftBank, Nvidia to build AI supercomputer
Beyond its AI-RAN efforts, it was also announced that SoftBank is working with Nvidia to build Japan’s most powerful AI supercomputer.
The Japanese firm will be among the first companies in the world with access to Nvidia’s Blackwell systems to use as the foundation for its upcoming supercomputer.
SoftBank plans to use the Nvidia-powered supercomputer to power generative AI development and business efforts.
Upon completion, the SuperPOD-based system is expected to be Japan’s most powerful supercomputing unit to date, ultimately surpassing Fujitsu’s Fugaku, which was the world’s most powerful supercomputer from June 2020 to June 2022, before losing its Top500 crown to Frontier.
RELATED STORIES
Blackwell GPU flaw '100% Nvidia's fault,' says CEO Jensen Huang
Ericsson, SoftBank to jointly develop AI RAN solutions
Nvidia unveils AI Aerial: Simulation telco tools to design AI-enabled RAN tech