- AT&T, T-Mobile and Verizon are all evaluating AI-RAN, but they’re not marching toward the same GPU-heavy future
- T-Mobile is the most visibly aligned with the Nokia/Nvidia AI-RAN camp, with prototype radios expected from Nokia later this year
- One broad takeaway: AI-RAN may be a single buzzword, but the Big 3 are pursuing it in different ways
It’s no secret that Nokia and Nvidia are dominating the headlines when it comes to their AI-RAN portfolio. They kickstarted the discussion around GPUs in the RAN and they’re working closely with T-Mobile at its AI-RAN Innovation Center in Bellevue, Washington.
So, it’s easy to see who’s aligned with whom. That’s why it was so interesting to get Verizon CTO Yago Tenorio’s take on the situation.
Here's Verizon CTO Yago Tenorio’s take on AI-RAN
“I’m a great admirer of GPUs in general,” he told Fierce. However, “I think in the industry there’s been some conflation about what you need on the radio and what GPUs are for.”
He described how Verizon worked with its vendors on a pair of prototype smart glasses that will augment the fan experience at sporting events. The glasses use cameras and microphones to enhance the action on the field with stats about the players and more.
They used a set of GPUs at the network edge and “it works amazing,” he said. “It works really fast.”
However, “do we need to put them in the radio? Absolutely not. Would the experience be better if we put those GPUs on the network? No, it’d be the same at most. Therefore, I think we’re talking about two different things.”
One is using GPUs for applications purposes, like they did for the smart glasses, which he refers to as “talking glasses.” The other one is using GPUs for processing the radio workload at the base station.
“Do you need a GPU for doing just RAN? No, today, you don’t. You can do that with a CPU. And can you embed AI into the radio and do some inferencing? Not for the end use case, but for just getting better at the radio performance … you can embed that onto a CPU. You don’t need a GPU for that. That’s true today and it’s probably true two years from now, three years from now, maybe five years from now, or a decade from now.”
Basically, using a GPU to reprogram the entire radio stack is unnecessarily complicated, he said.
This is T-Mobile's AI-RAN plan
Contrast that with where T-Mobile is coming from. Remember, T-Mobile is heavily entrenched with Nvidia and Nokia in the AI-RAN laboratory. Later this year, T-Mobile expects to get AI-RAN radio prototypes from Nokia and to really start putting them to the test. (Note: T-Mobile is the only U.S. carrier of the Big 3 publicly devoted to using Nokia RAN infrastructure in its mobile network.)
T-Mobile Network CTO Ankur Kapoor acknowledged Nokia’s claims about achieving 50% spectral efficiency with AI-RAN by the end of 2027 and 100% by the end of 2028.
“We’re in the early stages of this, but actually, what I’m seeing today is very promising,” he said during Fierce’s AI-RAN virtual summit earlier this month. “It’s actually starting to show up in the network,” with some early demonstrations showing 30% or more spectral improvements.
For T-Mobile, more spectral efficiencies will translate into more capacity for fixed wireless access (FWA) home broadband customers, for example. That also means higher speeds for these customers.
That said, there’s a lot of overhang attached to GPUs: high power consumption and high cost, to name a couple biggies.
“We think of it very differently,” he said. “It’s not about how much power these infrastructures are consuming. It’s what are the customer benefits that we can deliver … That’s what we’re going to make decisions on."
There will be places where a non-GPU is enough to carry the workload. “I don’t think it’s going to be a binary yes or no. I think it’s going to really kind of depend on what are the needs, what are you trying to deliver as a customer experience, what are the services you’re trying to do and more importantly, how much demand is out there," Kapoor said.
Where is AT&T with AI-RAN?
If Verizon is at one extreme and T-Mobile is at another, AT&T might be considered to be somewhere in the middle. Maybe.
AT&T’s main vendor is Ericsson, which won AT&T’s open RAN business several years ago and is in the process of replacing the gear in markets where AT&T had deployed Nokia. Ericsson is also is a founding member of the AI-RAN Alliance, but it didn’t get a $1 billion infusion from Nvidia like Nokia did.
Nonetheless, AT&T is keeping an open mind about GPUs in the network, according to Rob Soni, vice president of RAN Technology at AT&T.
AT&T considers AI RAN to be anything that allows it to bring AI technology into RAN operations. These could include AI applications that help with troubleshooting and inline network optimization.
Does he anticipate putting GPUs in the network? In base stations? Everywhere?
The short answer: Yes and no.
“We work quite closely with our partners on evaluating where they are overall with the Nvidia platform. We're also interested in other GPU platforms specifically for this purpose because tailoring it for a RAN application requires a certain power/price/performance position,” he said during the Fierce AI-RAN summit.
They’re tracking the evolution “quite closely,” he added.
“We’re very interested in continue to track this evolution quite closely. We're not interested in particularly running third-party workloads at cell sites yet, unless there's a business case,” he said.
They’ll also be watching closely because even though it doesn’t appear that GPUs will be necessary in Integrated Sensing and Communication (ISAC) use cases, there’s potential for GPUs to help. So far, AT&T has worked with Ericsson to demonstrate ISAC using a CPU.
Regarding Nvidia, it’s a company with “great assets” and a lot of innovative talent, including some folks who are former AT&T colleagues. “I think they’re working quite hard to try to break into this business and we always appreciate that effort from non-traditional players to try to help us continue to innovate in our market.”
But AT&T doesn’t care about the GPU or CPU labels so much. “We have an ability to consume any silicon. If a GPU shows up and it fits the network and it makes sense from a business perspective, we’ll do it. But we won't just do it just for the sake of doing it and trying to establish a playground for third parties,” he said.
Read more about AI-RAN and the Big 3 wireless carriers on Fierce Network
Verizon adds more firepower to its 6G Innovation Forum
Nokia’s AI-RAN pitch gets reality check in new study