Verizon signed a $1 billion dark fiber agreement with Google to connect the hyperscaler's data centers, CEO Dan Schulman disclosed this week. The contract marks Verizon's entry as an infrastructure-layer vendor to AI training clusters, with Schulman flagging additional deals worth multiple billions before December close.
The Google arrangement grants dedicated fiber pathways between Google's compute facilities without routing through Verizon's public network. Dark fiber leases deliver unlit strand capacity, allowing the tenant to deploy their own optical transmission gear and encryption standards. Google controls latency, protocol, and routing—essential when synchronizing distributed training runs across regional availability zones. Verizon provides the physical asset and maintenance access. The $1 billion figure likely represents a multi-year lease with buildout capital folded in, not a one-time sale.
The deal structure matters because it reframes Verizon's AI thesis. The company is not competing in inference or model training. It is positioning as the physical connectivity layer beneath the hyperscalers' AI ambitions, where capital expenditure shifts from Verizon's balance sheet to the customer's operational budget. Schulman's pipeline commentary—deals worth multiple billions by year-end—suggests Microsoft, Amazon, or Meta negotiations are well advanced. Hyperscalers are standardizing on owned dark fiber to eliminate jitter and third-party routing risk as models scale past 1 trillion parameters and training clusters exceed 100,000 GPUs. Verizon holds metro and long-haul fiber spanning 275 markets. The asset was built for consumer broadband and enterprise MPLS. Repurposing it for AI interconnects requires minimal incremental capital but commands premium lease rates because the customer cannot replicate the footprint quickly.
The forward risk is utilization. If Google's AI capital expenditure decelerates or inference workloads consolidate into fewer regions, leased fiber becomes stranded cost. Verizon mitigates this by signing long-term commitments with early termination penalties, but the model depends on sustained hyperscaler spending. The bullish case: AI training is moving from centralized clusters to geographically distributed inference, which requires more connectivity, not less. Edge inference and real-time model updates demand low-latency paths between cloud regions and metro edge nodes. Verizon's fiber grid supports that architecture.
Watch for Verizon's Q4 earnings call in late January, where Schulman will likely quantify the pipeline he referenced. Look for contract length, minimum revenue commitments, and whether deals include buildout clauses that push capital risk back to Verizon. Separately, monitor CapEx guidance for fiscal 2025. If the company holds infrastructure spending flat while revenue from dark fiber climbs, the margin story strengthens. Microsoft and Amazon have both signaled plans to double AI-related infrastructure investment through 2025. If either announces a Verizon partnership before year-end, the narrative shifts from opportunistic deal to structural positioning.
Schulman took the Verizon CEO role in October. The Google deal closed within sixty days of his arrival. That pace suggests the contract was negotiated under prior leadership, but Schulman is using it to telegraph a broader repositioning. The infrastructure is already in the ground. The question is whether hyperscalers will pay Verizon to use it, or build their own.