SAN FRANCISO—Ever since Nvidia, SoftBank, T-Mobile and other industry heavyweights launched the AI-RAN Alliance at Mobile World Congress 2024 in Barcelona, it’s been a bit of a mystery as to what’s going on behind the scenes.

What exactly does it mean to put artificial intelligence (AI) in the Radio Access Network (RAN)?

Sure, the alliance shared their aspirations to achieve real-time network optimization and improve energy efficiency. The chair of the alliance discussed how three working groups are working diligently to get the ball rolling. But the industry at large was left largely in the dark as to how this is all going to evolve.

That changed significantly when Nvidia CEO Jensen Huang took the stage Wednesday with T-Mobile CEO Mike Sievert at T-Mobile’s Capital Markets Day, where they waxed on about how they’ll change the world with AI in the RAN. Huang remarked how wireless networks are computationally intensive and use “insanely complex” algorithms.

“For the very first time, we’re going to bring artificial intelligence to it,” Huang said, explaining that it requires “a whole new type of computer.”

Nvidia shipped that computer to the lab in Bellevue, Washington, where Nvidia is working with T-Mobile, Ericsson and Nokia to figure out how they’re going to put AI in the RAN and make it work in the real world.

Catching up with CTO John Saw

On the sidelines of T-Mobile’s Capital Markets Day, Fierce asked T-Mobile CTO John Saw for a semi-deep dive into what they’re doing.

T-Mobile has been working closely with Nvidia to build the vision for essentially a different compute platform for networks, he said. The relationship is not exclusive; Nvidia is, of course, free to strike deals with Verizon and AT&T. But T-Mobile is first.

“You heard Jensen today,” Saw said. “There’s this good synergy, good alignment of interests. At T-Mobile, we move fast … We love to work with the Nvidia team, and Ericsson and Nokia as well.”

It bears noting that T-Mobile was first with a nationwide 5G standalone (SA) network, which doesn’t fall back on a 4G LTE core, and it will be first to dabble in 5G Advanced, parts of which President of Technology Ulf Ewaldsson said will come later this year. So the collaboration with Nvidia is another “first” it can boast.

And this isn’t coming entirely out of left field. Everyone understands that wireless networks are becoming “cloudified,” or moving to the cloud. But the question is: What’s the right compute to be using? Is it just central process units (CPUs)? Or do they want to combine CPUs and graphics processing units (GPUs) for a much more powerful combination? 

“The thought is, can we work together to actually, on the same platform that has CPUs and GPUs, run telco workloads side by side, AI workloads together,” he said. “Suddenly you have a multi-purpose network.”

To get started, they might put GPUs in mobile switching offices and slowly include base stations over time, he said.

In the future, “not only can our networks support the traditional telco workloads, including the radio access network, but we also would have sufficient capacity at the edges of our network to support AI workloads for our customers,” he said. “That’s the vision. Essentially, it’s becoming a multi-purpose cloud that does more than just telecom, which is what every other telecom network today is, but we’re thinking differently.”

With a multi-purpose network, the economics are better, and it potentially will create some new economic models to allow T-Mobile to offer AI workloads to its customers. For example, a workload can be devoted to robotics on a factory floor, or consumers could use it to do more things with AI on their phones.

“That’s the vision. You don’t have to build a separate AI network for AI and a separate network for telecom,” he said.

Room for more infra vendors?

Currently, T-Mobile is pursuing all of this with its long-time network infrastructure partners Ericsson and Nokia. Would T-Mobile consider using Samsung network gear?

“This is a project with basically a long horizon,” he said. “We’re not looking to switch our vendors. We like the vendors we have today, Nokia and Ericsson. We like them and they have invested a lot in R&D to make our network the best 5G network. So we want them to take it to the next level with us, and we brought in Nvidia to bring a new computation platform that we can all work on.”

AT&T embraced open RAN seemingly wholeheartedly when it awarded a $14 billion open RAN contract to Ericsson last year. Verizon has deployed thousands of virtualized RAN (vRAN) sites across the U.S., mainly relying on Samsung, and it’s committed to open RAN.

But T-Mobile has been mostly quiet on the open RAN front, the reason being that it doesn’t believe open RAN can match what traditional RAN delivers. 

“We have been a bit more judicious because hey, if we’re going to build a radio access network in the cloud, let’s figure out what’s the best way to do that and can we build a cloud in a way that can also serve us in other ways,” Saw said. “For instance, can you also run AI workloads on the same network as our customers. And that’s why we’re working with Nvidia.”

Will T-Mobile adopt open RAN in its network?

“I think the AI RAN that we talked about today will supercharge O-RAN. There’s a lot of good things happening with open RAN but O-RAN has struggled to meet the same performance as traditional RANs and we think that by bringing in AI RAN that has the Nvidia GPU, we might be able to supercharge it and make it work a lot better. The bonus is it’s the same network that we can also use to run AI workloads simultaneously,” which is not contemplated in an open RAN-based network, he said. 

Original article can be seen at: