Verizon Sells Google $1 Billion of Unused Fiber to Link AI Data Centers

Verizon will sell Google more than $1 billion of dark fiber to link the search giant's AI data centers, the phone company disclosed on its Q2 2026 earnings call. The asset is not new. The fiber is cable Verizon laid during past buildouts and never lit, and the deal is the first signed hyperscaler contract under a new business line the carrier calls "AI Connect." On the same call, Verizon's new CEO, Dan Schulman, told investors that more such deals are expected by year-end, worth what he described as "multiple billions of dollars in revenue over the next several years."

The fiber is the easy part to explain. Dark fiber is fiber-optic cable that was installed but never turned on. Carriers laid extra glass during the 2000s and 2010s because adding more fiber to a trench was cheap relative to the cost of returning later, and a lot of it has been sitting unused. The Google deal monetizes a slice of that stranded capacity by dedicating it to data-center-to-data-center traffic. Verizon is also adding to that footprint through its Frontier Communications acquisition, which closed on January 20, 2026, putting more long-haul fiber under one roof.

The harder piece is what Verizon is doing with its old central offices. A central office is the local building that used to house the telephone switch for a neighborhood, the place where copper lines from homes and businesses met the rest of the network. As Verizon retires copper in favor of fiber to the home, those buildings have been emptying out. Under the AI Connect plan, hundreds of them are being repurposed as small data centers for AI inference, which is the part of AI where a trained model produces an answer in real time, rather than the longer training step where the model is built.

The buildings matter because they are already where Verizon's fiber terminates, and because they sit close to where people and machines actually use AI. Inference workloads such as autonomous driving, remote robotics, and remote surgery need response times measured in milliseconds, and the speed of light sets a hard ceiling on how far a signal can travel in that window. The bet is that there is a market for small, distributed compute near the user, alongside the giant hyperscale campuses that train the models in the first place. Schulman cast the Google deal as a complement to that shift. Hyperscalers still build the giant campuses, but they need fiber between them, and the workloads that have to run close to the customer need the small buildings.

The number to watch is the one Schulman has not yet signed. The Google agreement is a real contract with a real counterparty, but the "multiple billions" figure is a forward-looking projection that depends on additional deals the CEO says are coming by the end of 2026. The execution risk is the workloads themselves. Autonomous fleets, factory robots, and remote-surgery systems all generate traffic that fits the inference-edge thesis, but none of them operate at consumer scale today. If those workloads stay niche, the buildings and the fiber still have value, just not at the price the CEO is currently describing to investors.