Nvidia, SoftBank plot big rewards from AI in the RAN

  • SoftBank is partnering with Nvidia to transform base stations from cost centers to revenue-producing AI assets
  • SoftBank revealed it’s using Nvidia AI technology to supercharge its 5G network
  • Nvidia and SoftBank estimate telco operators can potentially unlock billions in AI revenue opportunities 

As if Nvidia didn’t already have enough to celebrate, CEO Jensen Huang took the stage at the Nvidia AI Summit Japan this week to herald a series of firsts with SoftBank – including the successful pilot of a combined AI and 5G telecom network.

This isn’t the first time the two have come together on AI in the 5G Radio Access Network (RAN). Along with a few other industry heavyweights, they led the formation of the AI-RAN Alliance, unveiled at Mobile World Congress Barcelona in February.

But now they're sharing some of their most ambitious hopes and dreams for AI in the RAN. The big one: The ability for operators to turn their base stations from cost centers into AI revenue-producing assets.

To set the stage, SoftBank conducted an outdoor trial in the Kanagawa prefecture, where it demonstrated that its Nvidia-accelerated AI-RAN solution achieved carrier-grade 5G performance and was able to do so while using the network’s excess capacity to run AI interference workloads concurrently.

Traditional telco networks are built to handle peak loads and usually use only about one-third of their capacity. With AI-RAN, they can expect to monetize the remaining two-thirds of the capacity for AI inference services, according to Nvidia. 

Nvidia and SoftBank estimate that operators can earn roughly $5 in AI inference revenue from every $1 capex they invest in new AI-RAN infrastructure. That’s not peanuts. Based on SoftBank estimates, that translates to a return of up to 219% for every AI-RAN server it adds to its infrastructure.

During a call with media, Nvidia SVP for Telecom Ronnie Vasishta confirmed that SoftBank will be the first to receive the Nvidia DGX B200 systems, which will serve as building blocks for its new Nvidia DGX SuperPOD supercomputer.

SoftBank isn’t disclosing how much it’s investing in the AI RAN network. Plans call for commercial operations to being in 2026, a spokesperson told Fierce. 

Before this week’s summit in Tokyo, Nvidia already had one notable Japanese telco customer for its Aerial product. NTT DoCoMo was the first-ever telco in the world to deploy a high-performance 5G virtual RAN (vRAN) from Fujitsu built on the Nvidia Aerial vRAN stack and Nvidia Converged Accelerators. DoCoMo expects the solution to reduce total costs by up to 30% and lower the power consumption at base stations by up to 50%.

In the U.S., Huang was a surprise guest at T-Mobile’s Capital Markets Day, where he joined T-Mobile CEO Mike Sievert to talk all things AI and wireless. T-Mobile, Nvidia, Ericsson and Nokia are all founding members of the AI-RAN Alliance. They’re working at a Bellevue, Washington lab to figure out how to bring AI and RAN together for better revenue-generating network experiences.

T-Mobile CTO John Saw told Fierce at the time that T-Mobile has been working closely with Nvidia to envision a different compute platform for networks. The vision is to build a multi-purpose network that creates new revenue opportunities for T-Mobile.

Scope of AI in the RAN is evolving

Using intelligence in the RAN is not a new idea. However, the scope of AI in the RAN is evolving as operators learn more about the tradeoffs with the broader open RAN movement, said analyst Stefan Pongratz of the Dell’Oro Group.

“Although the open RAN train is slowing somewhat this year, our position here remains unchanged — operators will over time incorporate more openness, virtualization, intelligence and automation into their RAN roadmaps,” he told Fierce.

Though it's still early days, Pongratz sees progress in AI-for-RAN as operators incorporate intelligence to improve network performance and spectral/energy efficiencies.  

“As the 5G coverage improves, operators can slow the expansion plans and figure out what they can do to broaden the role of AI in infrastructure and overall network management and ultimately enhance the economics with existing sites,” he said. 

Update 11/13/2024 9:20 a.m. ET: This story was updated with comment from SoftBank.