S15.4 · Telecom & Connectivity
Wholesale long-haul and regional fiber networks experiencing the sector's clearest AI-driven repricing as hyperscalers pre-purchase decades of capacity.
Fiber and middle-mile infrastructure providers own and operate the wholesale long-haul and regional networks that carry traffic between cities, data centers and carrier networks. Distrust any single market figure here: competing scope definitions put dark and wholesale fiber revenue anywhere from $8-15B in 2024 (Grand View Research, Transparency Market Research, Mordor Intelligence), a spread wide enough that the estimates are not really measuring the same thing. Physical build-out is the better proxy — US long-haul route-miles have roughly doubled over the past decade, overwhelmingly on hyperscaler-anchored demand (Zayo, Lumen, Uniti public disclosures).
Growth is accelerating sharply, and the buyer driving it is the hyperscaler: datacenter interconnection demand for AI workloads is producing multi-year capacity pre-sales and new long-haul construction (Zayo, Crusoe, Lumen 2024-2025 disclosures). Capacity concentrates in the US, with new AI-driven corridors emerging in Texas and the Ohio Valley alongside established transatlantic and transpacific subsea routes. Wholesale supply sits with a handful of carriers — Zayo, Lumen, Uniti and GTT among them — next to a growing pool of fiber that hyperscalers own and operate privately rather than lease.
Upstream, the business depends on construction and field services and on securing conduit and right-of-way; downstream, the natural buyers are data center operators, mobile carriers needing backhaul, and enterprises buying wide-area network capacity. The contracts are indefeasible-right-of-use (IRU) and lit-service agreements carrying rent escalators — high fixed cost, low marginal cost per additional unit of capacity sold, infrastructure economics in the same family as towers and data centers. Once a route exists, lighting additional wavelengths or fiber strands costs a fraction of the original construction. That is why incumbency on an established long-haul route beats new capital trying to build competing capacity into the same corridor.
No segment in telecom shows the AI repricing more cleanly. Training and inference traffic is directly repricing long-haul and subsea fiber capacity, and hyperscalers have shifted from buying capacity as needed to pre-purchasing multi-decade IRUs — locking in supply for themselves and contracted revenue for the route owners years ahead of use. The new dollars sit in hyperscaler-anchored builds, corridors that exist only because AI demand called them into being. Legacy long-haul overbuild for refresh and redundancy is maintenance spend by comparison, defending existing routes rather than adding demand.
The growth of hyperscaler-owned private fiber is also blurring the old line between wholesale telecom infrastructure and the internal infrastructure large technology companies build and operate for themselves. That is not a forecast; it is already underway.