0

Verizon is positioning long-haul fiber and legacy central offices for two phases of the AI infrastructure buildout: connecting distributed data centers today, then supporting inference closer to network endpoints

The first phase of the AI infrastructure boom has created enormous demand for compute, power and optical connectivity inside data centers. As compute clusters scale beyond individual facilities, the networking problem is expanding outward; hyperscalers need high-capacity fiber connecting data centers across metropolitan and long-haul routes — a need operators seem poised to meet. 

Verizon sees that transition creating a new growth business on top of its core connectivity proposition. But the more interesting part of its strategy extends beyond data center interconnect. As AI moves from centralized training toward inference, Verizon also sees an opportunity to repurpose thousands of legacy central offices as locations for edge compute.

Taken together, the strategy gives Verizon two ways to participate in the AI infrastructure cycle. Fiber connects concentrations of compute today; powered, permitted and fiber-connected central offices could put inference infrastructure closer to enterprises, users and physical AI applications tomorrow. In both cases, Verizon is attempting to turn assets accumulated during earlier generations of telecommunications into infrastructure for the AI economy.

The scale-across play

Speaking at the Goldman Sachs Communacopia + Technology Conference, Verizon CEO Dan Schulman was bullish about the scale of the opportunity. “We are in the right place at the right time to be a part of one of the greatest capital expenditure things around AI infrastructure that we’ve ever seen in our generation,” Schulman said.

The evolution he described is straightforward. AI infrastructure initially required enormous amounts of connectivity within individual data centers as increasingly large clusters tied accelerators and racks together. The next problem is connecting separate facilities so hyperscalers can effectively aggregate compute across a much larger geographic footprint. In this case, “There is absolutely the need for dark fiber connectivity,” he said.

That plays directly into Verizon’s existing asset base. The company has extensive metro and long-haul fiber, and its AI Connect business is now packaging that infrastructure specifically around hyperscaler requirements.

At the Citi Global TMT Conference, Verizon CFO Tony Skiadas described AI Connect as “a new revenue stream above our core business.” Verizon has already signed a long-term Google agreement worth more than $1 billion, has additional deals in its pipeline and sees the category becoming a multibillion-dollar opportunity. Skiadas said Verizon expects AI Connect to become meaningful to its results in 2027.

Fiber supply becomes strategic

Verizon’s new agreement with Corning provides some indication of the scale at which it intends to pursue the opportunity. The multi-billion-dollar agreement covers more than 80 million miles of high-density optical fiber and connectivity solutions between 2027 and 2032. The fiber will support Verizon’s broadband expansion as well as long-haul routes connecting AI data centers. Verizon says cell sites, enterprise data centers and residential neighborhoods will ultimately connect into the same high-capacity fiber architecture.

The headline number is striking, but supply assurance — a lynchpin of managing AI infrastructure-related constraints — is a key element. “Fiber is in short supply as well,” Schulman said at Goldman Sachs, adding that the Corning deal gives Verizon confidence it can execute its planned construction without getting thrown by material availability.

Power is the obvious example, but fiber availability, conduit space and established rights of way can also constrain how quickly new compute facilities can be connected. Verizon is effectively securing one of the basic physical inputs required to participate in that buildout through the remainder of the decade.

The fiber also has multiple potential uses. Verizon can serve hyperscale customers, support broadband expansion, connect wireless infrastructure and consume capacity internally. This makes the AI both a standalone infrastructure bet and an incremental monetization layer on top of an asset the company already needs.

From copper-era cost centers to inference infrastructure

The more interesting and novel piece of Verizon’s strategy comes as AI moves out of the data center. Schulman highlighted what he called “a big demand for inference computing” at Goldman Sachs, pointing to low-latency applications including robotics, autonomous driving and remote surgery.

Verizon potentially has another legacy asset positioned for that transition in its central offices. As the company retires copper, thousands of those facilities lose their traditional purpose. But the buildings are already connected to telecommunications infrastructure. Many have power, existing permits and physical locations distributed through metropolitan areas.

“As we decommission our central offices, take copper out of them … they’re power-ready, already permitted, and we have thousands of them,” Schulman said at Goldman Sachs. Verizon is already signing deals around converting some of those locations into inference edge-computing facilities, he added.

The economics are potentially compelling because network modernization works in both directions. Retiring copper lowers legacy operating costs. Skiadas told investors at Citi that removing legacy network elements, replacing copper with fiber and reducing access costs are among the largest components of Verizon’s cost transformation.

But instead of just abandoning every location associated with that old architecture, Verizon may be able to turn selected central offices into infrastructure for a new computing architecture. The transition is essentially copper-era network node to inference-era compute node. If distributed inference develops as Verizon expects, a physical asset that once represented legacy cost could become a new source of revenue.

AI infrastructure can compound fiber economics

Verizon is also framing AI Connect in a way that addresses one of the biggest risks associated with the current infrastructure boom: deploying too much capital ahead of durable demand. Skiadas said at Citi that Verizon expects AI Connect construction to use “success-based capital,” with investment tied to specific opportunities rather than speculative network expansion.

The economics can also improve as routes attract more customers. “If we have a tenant on the fiber route, if we like that margin, it’s a really good margin,” Skiadas said at Citi. “If we get two or three tenants on that route, it’s very attractive for us.”

Verizon can then use the same fiber for its own transport requirements, creating another source of value through avoided network cost. That creates a potentially powerful infrastructure model where securing an anchor customer justifies construction, adding tenants improves utilization and returns, and the network can also be used internally.

The longer-term architecture could extend even further. Verizon’s recent 6G trials combine edge AI with sensing, including experiments using cellular infrastructure to detect crowd density and track drones and vehicles. An AI Sports Companion prototype placed a locally optimized AI workload on Verizon’s edge network to support AI glasses, while Verizon CTO Yago Tenorio described the broader objective as an AI-native network designed around future AI workloads and wearables.

Verizon’s near-term opportunity is concrete because hyperscalers need fiber, Verizon owns routes they want, and contracted AI Connect revenue is beginning to materialize. The more consequential long-term bet is that the same infrastructure transition can extend into inference.

Verizon’s participation in the AI economy, then, is about (of course) carrying more traffic, but doing so while turning two assets inherited from earlier generations of telecom — long-haul fiber and central offices — into the connective tissue and distributed compute locations of the AI economy. If it can do that while retiring the costs associated with the legacy network, AI infrastructure could materially change the value of infrastructure Verizon already owns.

For a broader look at the role of fiber in the AI infrastructure supercycle, download this free RCRTech Trends piece, “The AI traffic effect: A new investment cycle for fiber networks.”