Verizon, AT&T, and SK Telecom are betting big on AI infrastructure. But technology advisory group Omdia says the headline numbers are ceilings, and warns revenue will trail capacity growth for years.
In sum – what to know:
Unproven scale – Verizon’s $1bn Google DCI agreement is signed and live; Schulman’s claim of a multi-billion-dollar “growth vector” from 2027 is still a forecast.
Small contracts – Hyperscalers spec deals at their eventual ceiling (sometimes 400 waves of capacity) but sign carriers up for a fraction at the start (as few as six waves).
Traffic ≠ revenue – Capacity demand from AI is “guaranteed” to rise, Omdia says, but revenue per bit keeps falling, highlighting a structural difference in the two growth curves.
Verizon chief Dan Schulman recently told investors that AI infrastructure deals will “stack up” in 2027. The carrier’s $1 billion dark-fiber deal with Google, signed in late July, is, he believes, just the start. Other deals worth “multiple billions” will follow by year-end, forming what he calls “long-duration, high-quality contracted revenue streams” that stack up into a new growth vector from 2027, before expanding substantially in the next five to ten years.
“This is a very different revenue growth profile than Verizon has had in a very long time,” he said.
As discussed this week, telcos are increasingly positioning themselves as infrastructure providers for the AI economy. The Google deal is evidence of that shift. Technology advisory group Omdia’s Telco B2B AI Monetization Index puts total telco AI revenue at $4 billion globally in 2025, growing at a CAGR of 65 percent to 2030. But how fast, and how far, that shift scales is still an open question, and the numbers suggest it’s earlier days than the framing implies.
Brian Washburn, chief analyst for telco B2B solutions at Omdia, has spent three years modelling AI network traffic. He has recognised a pattern behind headline deals like these: hyperscalers spec out contracts at their eventual ceiling, but sign carriers up at the beginning for a fraction of it. “A lot of these multi-billion-dollar contracts have potential, and the growth curve is very promising,” Washburn said. “But it all starts kind of small.”
A “wave”, short for wavelength, is a single channel of light on a fiber-optic cable, each one capable of carrying a set amount of data; carriers light up more waves as demand grows, without laying new cable. A hyperscaler might spec a deal at 400 waves of 400 gigabits each, framing that as the ceiling for a 20-year agreement. The initial signed contract, however, might cover six waves, which is around 1.5 percent of the stated capacity.
Fiber-glut parallel
Washburn draws a direct parallel between the current overbuild and the fiber glut in the 1990s, where too much investment capital was chasing after too few commercial revenue dollars, the dynamic that led to fiber routes being turned off, and eventually led to consolidation of CLECs like Level 3 and Zayo. While he expects something similar in AI infrastructure, given there are “too many players and too much fragmentation,” he sees a crucial difference for carriers. “We are the secondary beneficiary,” he said.
If the AI infrastructure layer consolidation becomes a mess, the underlying demand for connectivity (supporting GPUs as a service and AI platform requirements, for example) doesn’t disappear, it just gets served by whoever is left.
Fortunately, telcos are being careful about separating out opportunities with neoclouds and hyperscalers that are real and cogent, and those that are good ideas on paper and chasing investment. That discipline to work on smaller and more pragmatic deals, rather than bigger speculative ones, Washburn said, is partly forced, as telcos don’t have access to the near-unlimited capital of AI players. One example of this approach is the pre-purchase of optical transport equipment to provide wavelengths with short provisioning windows; a calculated, incremental commitment rather than a speculative land-grab.
AT&T’s network redesign is a case in point. The carrier is deliberately re-engineering for upstream capacity rather than the downstream speeds carriers have marketed since 4G, betting that agentic AI traffic (such as drones, autonomous vehicles, robotics) will push large volumes of latency-sensitive data from the edge back to the network. Like the dark fiber buildout, it shows how telcos are building out infrastructure ahead of proven demand.
SK Telecom is making a similar bet. The carrier has set up a dedicated subsidiary, SK Hyper, to build up to 15GW of AI data centre capacity, backed by up to $540 million in committed capital through to 2030. But even SK Telecom is hedging the timeline, with the first 5GW phase not due until 2029, and the full 15GW target not due until 2035, “subject to market demand and other factors.”
An upside caveat
Telcos may not need the AI infrastructure story to play out perfectly in order to benefit. Capacity demand from AI traffic is “guaranteed” to keep rising, he said. Revenue is a different matter: “The amount of revenues that one can extract from that will continue to rise, but at a far slower pace than the demand for capacity,” he said, because as the technology keeps advancing, “the price per bit keeps going down.” Growing traffic and growing revenue, in other words, are not on the same curve, and the gap between them is structural.
Washburn points to Level 3’s national network, where conduit laid in the 1990s is now more than 25 years old. The operator is now putting fresh fiber into empty space in that same conduit, having run low on capacity in its original fiber pairs, to extend the asset’s working life by another 25 years. A 50-year planning horizon on a single piece of infrastructure, he said, is the kind of timescale most industries cannot imagine. By comparison, Schulman’s AI infrastructure narrative involves a 2027 horizon, which is a fraction of the timescale telcos are actually building for.
This mismatch may be less a contradiction than the key differentiator. Whether “multiple billions” from a handful of deals compounds into the decade-long growth or settles into something steadier and smaller remains to be seen. Either way, carriers are betting the pipes remain a useful asset, regardless of which version of the story turns out to be true.