Land. Power. Connectivity. These are three of the biggest bottlenecks slowing data center expansions in the U.S. and abroad. And it just so happens that in many cases, telcos are sitting on a hoard of all three.
I’ve said it before and I’ll say it again, telecom operators are sitting on a potential goldmine in the form of old central office (CO) facilities (and, for that matter, cell sites) that could be converted into edge data centers. And as it turns out, the tide is turning ever more in their favor.
There are two current trends driving in telcos’ direction. First, while demand for compute power remains high, the public has started to push back on new data center construction.
A new survey from YouGov of nearly 6,500 adults in the U.S. shows that 47% of citizens believe the construction of new data centers has a somewhat or very negative impact on the country. In fact, “very negative” was the single largest response across all regions, genders, political affiliations and age groups.
Similarly, a Gallup poll of U.S. residents taken in March found that 48% said they “strongly oppose” the construction of a data center in their local area. Lumping the “somewhat oppose” responses in brings the oppositional figure to a total of 71%.
The polls reflect rising anti-data center sentiment that has been spreading across the country and spurring residents into action. Citizens in municipalities across Virginia, Missouri, Wisconsin, Pennsylvania, New Jersey, Ohio, Arizona and many other states have launched efforts to halt existing data center projects and stop new ones from coming to town.
A lot of opposition is centered around land use, rising utility rates and water scarcity. Which makes sense given the scale of the facilities being proposed.
But you know what would probably generate a lot less outrage? Putting edge data centers in existing, underutilized facilities that already have 5-10 megawatts of power allocated to them. Will these be the next Stargate projects? No. That’s not the point.
The point is to serve the second trend working in telcos’ favor, which is the rise of inferencing. Unlike centralized training workloads, inferencing can and should be distributed and preferably run close to where the action is happening. That's especially the case when it comes to various physical AI applications, but still applies to many other inferencing workloads that are less latency-sensitive.
Edge debate
There’s a bit of debate at the moment about the inferencing opportunity for telcos. Many have pitched it as a sweet spot where telcos have a right to play. Personally, I'm not entirely convinced things will play out as the industry hopes, but I can see a path forward for telcos.
Ericsson Americas CTSO Joe Constantine recently argued in an interview with Fierce that physical AI – think humanoid robots in factories and roaming the streets – will require compute at the edge. While he acknowledged there’s room to spar over whether compute needs to be at every cell site or just a smattering of distributed locations, he said these kinds of use cases just won’t work without edge inferencing. Full stop. That's where telcos come in, he said.
The flip side of the coin, however, is that some physical AI and other use cases could be so latency sensitive that inferencing needs to run on-device to ensure the humanoid robots or drones or what-have-you don’t cause harm. And all those other AI workloads? Well, those could just run in those massive data centers because they're not terribly latency sensitive, the thinking goes.
This isn’t telcos’ first edge rodeo. As Fierce has noted before, the current AI moment feels a lot like the mobile edge computing wave from a few years back. There’s a lot of promise without a solid “killer use case” to justify investment. Physical AI could be huge. Or it could go the way of the Metaverse and Google Glass.
That said, it's likely that AI, broadly speaking, isn’t going away (as much as I’m sure we’d all love to not hear about it for a moment). There will be inferencing use cases that need to be served closer to users. There will be sovereign AI demand. And telcos are poised to deliver thanks to their extensive networks of central offices (assuming they haven’t sold these off yet). They just need to execute before the rising mini modular data center players eat their lunch.
Retrofitting may not be easy thanks to liquid cooling requirements. And it may not exactly be cheap. But telcos have a leg up in the sense that they already have land and precious power. And with Google now apparently open to selling its TPUs, there could well be a path for operators to secure highly performant but lower-power chips at substantially less cost than they’d pay for Nvidia GPUs.
The questions is whether telcos want to lead or fade into the background as a utility in the AI era.
Opinion pieces from industry experts, analysts or our editorial staff do not necessarily represent the opinions of Fierce Network.