T-Mobile emerging as Nvidia's big AI cheerleader

T-Mobile officials are getting louder – and more specific – about the benefits they see at the intersection of AI and wireless. And AI chipset vendor Nvidia appears to be a linchpin in T-Mobile's unfolding strategy.

"We intend to be a company that drives the future on 6G. ... And that's why we've struck our unique partnership with Ericsson, Nokia and Nvidia to help invent AI RAN and bring the future about in a way that disproportionately benefits T-Mobile customers," said T-Mobile CEO Mike Sievert on the company's recent quarterly conference call. "I think the future bodes very nicely for 6G."

The AI-RAN Alliance launched at the MWC Barcelona trade show earlier this year to integrate AI technology with radio access networks (RANs). Alongside Nvidia (the world's second-most valuable company by market cap), the alliance's founding members include industry heavyweights such as AWS, Arm, Ericsson, Microsoft, Nokia and Samsung. But the group counts only two network operators as members: Japan's SoftBank and T-Mobile in the US.

Since its introduction earlier this year, the AI-RAN Alliance's efforts have been crystallizing. During the summer the group gained a top-tier leader: Alex Jinsung Choi, a SoftBank fellow and former Deutsche Telekom executive who has also held leadership positions at the Telecom Infra Project (TIP) and the O-RAN Alliance. The alliance also launched a handful of working groups.

The T-Mobile embrace

But it was in September when T-Mobile made the AI-RAN Alliance a cornerstone of its Capital Markets Day. Nvidia CEO Jensen Huang joined T-Mobile CEO Sievert onstage for an in-depth discussion on the future of AI in telecommunications. Huang also highlighted Nvidia's new AI Aerial platform, which aims to optimize network performance and efficiency by integrating AI with RAN technology, in part by leveraging Nvidia's new Blackwell generation of GPUs paired with its Grace CPUs.

Nvidia's AI Aerial platform represents the company's efforts to expand its role in the telecom industry on the back of its successes selling GPUs into data centers. If Nvidia is successful, that could have major implications for big network equipment companies like Ericsson and Nokia, which may have to build their future products around Nvidia's chips for customers like T-Mobile. It could also sideline companies like Marvell and Intel that currently supply chips for wireless networking equipment.

Also at its Capital Markets Day, T-Mobile announced that its two main wireless equipment suppliers, Ericsson and Nokia, would team with Nvidia to establish an AI-RAN Innovation Center in T-Mobile's Bellevue, Washington, headquarters.

"There's a potential big cycle coming," Sievert explained last month, in discussing 6G. "But ... it may be a cycle that's fundamentally more efficient to roll out than prior cycles. And the promises of open RAN that have been made for years may finally be realized at scale in the AI RAN era. And this allows you to offset potential future cost of replacement of future technologies with fundamental efficiency, and so we're optimistic."

Broadly, Sievert explained that T-Mobile hopes to direct the creation of 6G through its work with Nvidia and its other partners. "We wanted to put down markers about our vision for what comes next with AI RAN because it's so promising on the ability to do more," he said.

Going commercial

Just as Nvidia's CEO made an appearance at a T-Mobile event, so too did T-Mobile's John Saw appear at Nvidia's recent AI Summit in Washington, DC.

Saw arrived at T-Mobile through that company's purchase of Sprint in 2020. As the networking chief of Sprint, Saw was a champion of the midband 2.5GHz spectrum that eventually found its way into the heart of T-Mobile's 5G story. Now he's T-Mobile's CTO, working under T-Mobile's networking chief Ulf Ewaldsson.

At Nvidia's AI Summit, Saw explained that 6G hopefuls see AI as a critical piece of technology. However, "not many people have given a lot of detail as to how they are going to get there."

He said the AI-RAN Association, and T-Mobile's work with Nvidia and others, is a way to do that.

"We really want to discover a path to commercialization. This is more than a science project," he said, adding that the AI-RAN Alliance may ultimately start working with the 3GPP standards body that primarily develops cellular technologies. "But having said that, there's a lot of work to get done."

Saw said the AI-RAN Alliance's efforts might be commercially ready in time for the rollout of 6G, expected around 2030.

"Or, if we're lucky, maybe earlier," he added.

Like Sievert, Huang and others, T-Mobile's Saw said there are a number of reasons to combine wireless and AI technologies. For one, putting Nvidia's GPUs into T-Mobile's cell sites could allow the network operator to sell speedy AI services at the edge of its network. That's an idea that other network operators like Verizon have also championed.

A RAN with AI

But Saw also said Nvidia-powered AI capabilities could supercharge the performance of T-Mobile's RAN network. Specifically, Saw said there are computational limitations in the CPUs that T-Mobile currently uses inside its network, such that they can't handle the reproducing kernel Hilbert space (RKHS) channel estimator algorithm. That's one of the many complex calculations that can be used to manage wireless connections across a cell site.

"Guess what, a GPU might be able to do it," Saw said. "Those are the things that we're trying to investigate."

Indeed, Nvidia research outlines the significant performance gains network operators can get by using its GPUs and the RKHS approach to channel estimations.

It's worth noting the benefits of an AI-driven network have floated all the way to the top of T-Mobile's leadership team.

"When the trees' leaves come out in the spring, it changes the radio propagation and we have to make manual changes to how we operate the network in the spring and in the fall when the leaves come and go," Sievert, the T-Mobile CEO, told Fortune recently, in response to a question about how AI could help the operator's network. Sievert said AI could allow T-Mobile to better route its signals around physical obstacles, like the leaves on trees.

"An AI-enabled network will understand that that's happening," he said. "These are the kinds of things that sort of represent that, in a world that's constantly in motion and customers that are constantly in motion, a smarter and smarter network will be able to get you the massive capacity that you need."

The rest of the market

Of course, T-Mobile is by no means the only company that has eyed the benefits of high-performance, GPU-based AI processing inside a radio access network.

For example, radio vendor Fujitsu showed off radio technology this year that used Nvidia's GPUs to clean up wireless signal processing. The company recorded a dramatic improvement in upload speeds, from 252 Mbit/s to 366 Mbit/s.

Similarly, Korea's SK Telecom, Japan's Docomo and vendor Nokia have demonstrated what they're calling a 6G air interface that uses AI for better performance and energy efficiency.

And the US government continues to toy with the idea of using AI technologies to support spectrum sharing.

"The potential scale of these impacts could be staggering, ranging from network planning, spectrum management, enhanced user experiences via content optimization, virtualization, network slicing, as well as predictive maintenance and security. Truly, there does not seem like any area of the wireless cellular network that might not be impacted by these tremendous technologies," wrote Chris Pearson, president of the 5G Americas trade association, in a recent post to the group's website outlining the many ways that AI technology might be able to improve wireless communications.

"It's an amazing time," Pearson wrote.

But questions remain. Many continue to wonder whether the benefits derived from AI-powered network operations will be big enough to offset the high cost of AI-friendly GPU chipsets like those from Nvidia. That's partly why Nvidia has positioned its AI Aerial platform not only as a way to improve wireless network operations but also as a way for wireless network operators to make money from the sale of AI services to other companies.

However, that approach assumes that use cases – from robots to augmented reality glasses – will eventually emerge that demand AI services with speedy connections and low latency. That hasn't happened yet.

Nonetheless, AI could "suddenly open up new economic opportunities for the operator," argued Saw, the T-Mobile executive.