AT&T CEO John Stankey is quietly confident that the carrier has amassed the right combination of network assets to be a significant player in the data center connectivity space, though it remains unseen how far out to the edge that connectivity will be needed.

Stankey told an audience at the recent J.P. Morgan Global Technology, Media, and Communications Conference that AT&T’s fiber and wireless network investment moves have been about building out “great strategic access.”

“That's what our fiber investments in play was about, which is get more fiber to more businesses and more customers, because at the end of the day, for heavy workloads, it is the best, lowest marginal cost technology with the best future proofing that you can invest in,” Stankey said.

AT&T’s fiber expansion included its $5.75 billion purchase of Lumen Technologies’ consumer fiber business and subsequent plans to hit 60 million locations reached by 2030.

Stankey then expanded to the carrier’s wireless spectrum investments, with specific mention of AT&T’s recent EchoStar spectrum purchase. That $23 billion deal, which Stankey had previously noted was a high price to pay, included 30-megahertz of nationwide 3.45 GHz mid-band spectrum and approximately 20-megahertz of nationwide 600 MHz low-band spectrum.

Stankey specifically pointed to that 600 MHz spectrum as allowing AT&T “to get more symmetrical low-band spectrum that will allow us to engineer that in a way that we can have a higher performing network in the upstream than what historically wireless has delivered with our preferred low-band positioning.”

This combination provides “the infrastructure that people want to use, whether it's a LLM [large language model] or somebody's cloud infrastructure,” Stankey said, adding this then plays into supporting hyperscale providers.

“We've been actively in the data center market and actively working with hyperscalers to ensure that we're building shared infrastructure into their access points and have the right relationships through a combination of dark and lit fiber that ensures that our backbone and our aggregation networks can deliver those packets that we get off of our preferred access into their infrastructure, and I have visibility as I hand it into the infrastructure and receive it back and deliver it to the endpoints,” Stankey said. “That's the fundamental approach we're using in how we architect the network.”

Talking the talk

Stankey did add that “we're not maybe as vocal about it as some because we see it as only a part of our strategy to complement our access, where the real value is, and I think part of the reason you should assume that the improved performance that is occurring in our business market segment is related to what we've been able to do to sell on top of that value proposition and that infrastructure.”

That comment seemed tangentially tied to more “vocal” telecom network leaders that have touted the hyperscale opportunity.

Verizon CEO Dan Schulman, for instance, claimed the day before Stankey’s comments that Verizon “might be best positioned to capitalize on AI infrastructure. Obviously, the hyperscalers, alternative cloud providers, large enterprises that are moving rapidly into the AI age, are keen to leverage our assets.”

Schulman had previously stated that Verizon was in “deep discussions right now with hyperscalers, with alternative cloud providers, large enterprises to integrate our fiber, both dark and lit, and our 5G assets to support their AI infrastructure efforts.

“And that can include data center connectivity, ability to help them with their training and inference, and that is the potential for multibillions in revenue,” Schulman added, hinting, “we'll have more specifics on that in the next three to six months.”

Edge opportunity?

Schulman at the J.P. Morgan event also noted that Verizon’s opportunities were tied to the world “moving toward edge computing, toward data connectivity,” two moves that Stankey is not yet convinced of.

Stankey said there’s an evaluation process to determine “in that value chain of delivering that traffic, [does] it make sense to insert some compute capability someplace in where we have advantage points of presence out closer to the edge? And the answer is, it may, don't know. Still watching what those workloads are that are developing.”

“I can certainly come up with some that I think might require that,” Stankey continued. “I think a lot of it will depend on where those data centers for the various hyperscalers and those that own the AI infrastructure will be. My point of view is the large vertical players will probably be present in virtually every metro in the United States, and so then the question is, what is the latency to deliver off that high-performing access network we've built, we're building into that metro and will it require something faster than that extra couple miles to get from the edge of our network into the metro and back? And I think that remains to be seen.”