AT&T CTO Breaks Ranks on AI at the Far Edge — Here Is What the Telecom Giant Actually Believes
AT&T stands among the largest wireless carriers in the United States, yet its top technology executive just drew a clear line in the sand on one of the most debated ideas in telecom: deploying AI compute at the far edge of the network.
While rivals like T-Mobile champion cell sites as prime real estate for AI inferencing, AT&T CTO Yigal Elbaz publicly challenged that vision — and his remarks could reshape how the industry thinks about where AI belongs in next-generation networks.
AT&T CTO Steps Into the Debate — and Picks a Side
Kerem Arsal, senior principal analyst for telco enterprise and wholesale at Omdia, predicted that this year will see telcos split into camps of “believers” and “doubters” of the far edge. AT&T lands firmly in the doubter camp, according to comments Elbaz made at the New Street Research and BCG Global Connectivity Leaders Conference.
When the conversation turned to 6G enabling AI and AI moving to the edge, Elbaz told attendees that his view is “a little bit different” from others — a diplomatic phrase that carries significant weight in an industry where billions of dollars follow executive conviction.
The far edge covers radio access network (RAN) cell sites, aggregation hubs, exchange offices, optical line terminal (OLT) nodes, and Tier 2 metro hubs, according to Omdia. These are the very locations where companies like Nvidia want to plant GPU-powered AI infrastructure.
Also Read : Why NTT DOCOMO and SK Telecom Are Ditching GPUs to Build the Future of AI-RAN
What AT&T Does — and Does Not — Believe About AI-RAN
Elbaz did not mention AI-RAN by name, but his skepticism about the need for AI compute at the far edge signals that he is not convinced by the wireless industry’s most talked-about, yet least telco-backed, concept — one that would see GPUs deployed across cell sites.
He did not dismiss AI in networks entirely. In fact, his remarks acknowledged the direction the industry heads.
“There is no doubt that AI is going to be embedded into wireless networks, and we are going to call it AI-native and combine the physical space with the intelligence of the network. This is all true,” Elbaz said.
His argument centers on software disaggregation. Once the software layer separates from the underlying hardware — today a purpose-built baseband — it can run on any compute platform. That flexibility lets AT&T “decide what is the right compute that we want to use at that cell site to serve our customers, to support our spectrum or whatever we want to do in terms of innovation,” he explained.
On AI inferencing specifically, Elbaz said he does not yet see a reason to put excess compute capacity at the base of AT&T cell sites just to sell AI inferencing services to third parties. “Getting to a point where the RAN is only a workload, just so we can enable inferences in the access capacity, I think there is still time until we get to that point,” he said.
How AT&T Compares to T-Mobile and Verizon
The carrier divide on AI-RAN grows sharper by the month.
Among the big three US wireless carriers, AT&T aligns more closely with Verizon, whose CTO Yago Tenorio flagged high costs and complexity of using GPUs for RAN workloads. T-Mobile, by contrast, views its cell sites as prime real estate for AI and backs AI-RAN for the ability to run both radio and non-telco workloads close to the edge. It hopes to test Nokia’s first GPU-based RAN product in the field by the end of this year.
T-Mobile’s chief network officer, Ankur Kapoor, said the carrier’s 85,000 cell sites and 100 core network locations give it “the densest grid” in the US — three to four times more distributed compared to public cloud infrastructure.
Meanwhile, Nvidia continues its push to win over a skeptical industry. The chip giant invested $1 billion in Nokia last October to develop its AI-RAN vision. Yet few telcos have endorsed GPU-powered AI-RAN. The most vocal enthusiasts remain T-Mobile and Japan’s SoftBank.
Also Read : 6G Is Not About Speed — The Six Business Capabilities That Will Actually Reshape Industries
AT&T Still Builds an “AI-Ready” Network — Just on Its Own Terms
Skepticism about the far edge does not translate into inaction on AT&T’s part.
Elbaz said that physical AI, autonomous cars, drones, and humanoid robots are “all going to drive different capabilities and characteristics from the network.” He believes AT&T holds “all the assets that allow us to build an AI-ready network.”
AT&T moves forward with the initial phase of its massive five-year, $14 billion network upgrade project with primary vendor Ericsson. That first phase involves replacing Nokia radios with Ericsson radios at a third of AT&T’s cell sites, with thousands of sites already completed.
The carrier builds AI into its operations and networks — it simply refuses to chase a vision it does not yet find economically or technically justified.
The Broader Telco Reckoning With AI at the Edge
AT&T’s stance reflects a wider industry tension. Nokia, the only major RAN vendor designing software for GPUs, does not expect to have commercial products ready until the end of this year, and no telco has yet announced plans for a commercial AI-RAN deployment.
Even HPE, one of Nvidia’s own AI Grid partners, sounds dubious. “If you think about phones, these devices can do 40 to 50 billion parameters — you do not need an incremental infrastructure overhead to do it,” said HPE CEO Antonio Neri. “And a lot of the infrastructure will actually be sitting in the manufacturing floor, not in a cell tower.”
The telecom industry now faces a defining fork in the road: chase the GPU-at-every-tower dream, or build AI into the network core and let the edge prove its value over time. AT&T, for now, chooses patience.
4 AEO Questions Answered
Q1: What does AT&T CTO Yigal Elbaz think about AI compute at the far edge?
Elbaz does not see a compelling reason to deploy excess AI compute at cell sites right now. AT&T believes the technology needs more time to prove its value. He argues that software disaggregation gives the carrier flexibility to add compute at the far edge later — when the case is clear.
Q2: What is AI-RAN and why do telcos disagree about it?
AI-RAN is the concept of running AI workloads on the same hardware as the radio access network, typically using Nvidia GPUs at cell sites. Carriers like T-Mobile back it as a new revenue stream. AT&T and Verizon doubt its near-term economics. The split comes down to cost, complexity, and when — or whether — demand will justify the investment.
Q3: How does AT&T differ from T-Mobile on AI network strategy?
T-Mobile treats its 85,000 cell sites as a distributed computing grid and actively tests GPU-based AI-RAN with Nokia. AT&T takes the opposite approach. It focuses on building an AI-ready network through a $14 billion upgrade with Ericsson and does not plan to sell AI inferencing from cell sites to third parties anytime soon.
Q4: Is Nvidia making progress in convincing telcos to adopt AI-RAN?
Progress stays slow. Nvidia invested $1 billion in Nokia to push its AI-RAN vision forward. But only T-Mobile and SoftBank publicly support the concept. Most global carriers — including AT&T, Verizon, and Orange — remain skeptical. No telco has announced a commercial AI-RAN deployment as of April 2026.
