S15.5 · Telecom & Connectivity
Colocation and interconnection facilities where AI training and inference demand has made power and land, not capital, the binding growth constraint.
Data center and interconnection providers own and operate the physical facilities that house compute infrastructure and connect networks to each other. Scope drives the estimates apart — $60-100B for the 2024/2025 base across Grand View Research, Mordor Intelligence and MarketsandMarkets, with MarketsandMarkets projecting $204.4B by 2030 on hyperscaler and AI-driven demand. Whatever the base, this is the segment where AI's effect on telecom and connectivity infrastructure is most direct and least arguable.
The forecast pace is double-digit, and nearly all of it traces to AI training and inference capex rather than traditional enterprise IT outsourcing. Demand concentrates in the US — Northern Virginia and Texas remain the dominant markets — with power-constrained secondary markets emerging as the primary sites fill. Interconnection density lives in the established hubs, Ashburn, Frankfurt and Singapore, anchored by Equinix and Digital Realty. At scale the market is concentrated between those two operators and hyperscaler self-build; beneath them runs a long fragmented tail of regional colocation operators.
The inputs are power, land and real estate, and fiber connectivity; the customers are hyperscalers, enterprises and AI labs. Long-term leases with contracted rent escalators produce EBITDA margins above 45-50% at scale. The binding constraint has moved, and the move is the story: power and land availability now gate growth, not construction capital — a structural shift from a capital-intensity story to a resource-scarcity story. A site with secured, deliverable power commands terms and lease commitments that a comparable but power-constrained site cannot, regardless of how much capital its owner is prepared to spend on construction.
AI dominates demand here more than anywhere else in telecom and connectivity, directly repricing power and interconnection capacity. With power availability having displaced construction capital as the primary constraint on how quickly new capacity comes online, operator strategy sorts into two lines of spend. Securing power — through procurement agreements or on-site generation — is defensive necessity; without it, existing pipelines stall. Expanding into edge and AI-inference sites is the expansionary complement, chasing new demand rather than protecting existing commitments.
The practical consequence is that data center growth is now power-and-land-driven rather than purely a real estate or connectivity story. Operators are being pulled directly into energy markets and industrial real estate — territory that was never previously central to the business — and that pull is already underway, not emerging.