Hyperscale is single-tenant build-to-suit at 50 to 200+ MW per phase. Colocation is multi-tenant facilities at 5 to 50 MW per site. Edge is small distributed deployments at 1 to 10 MW for low-latency workloads. AI compute overlays onto hyperscale at higher rack density and liquid cooling. Each finances differently.
Four buckets: four financing profiles
Hyperscale data centers are built-to-suit for a single investment-grade cloud or AI tenant under a 10- to 20-year triple-net lease, typically 50 to 200+ MW IT load per phase. Bank-led syndicates and dedicated debt funds finance construction. CTL life-co perm dominates stabilized takeout.
Colocation data centers serve multiple enterprise tenants under shorter (3 to 7 year) leases with the operator providing power, cooling, security, and cross-connect services. Revenue per kW is higher than hyperscale on a per-MW basis, but operating costs and tenant churn are also higher. Lenders use operating-business underwriting rather than CTL underwriting.
Edge data centers (1 to 10 MW typical) sit in metro areas to support low-latency workloads. Regional banks, SBA 504 for owner-operators, and a thin bench of specialty debt funds finance edge. AI compute is not a separate building type but an overlay, AI training workloads typically deploy in hyperscale-class facilities at 30 to 100 kW per rack with liquid cooling, while AI inference can deploy across hyperscale, colocation, or edge.
Sources
- 2024 United States Data Center Energy Usage Report · Lawrence Berkeley National Laboratory (LBNL) · 2024
- Powering Intelligence: Analyzing Artificial Intelligence and Data Center Use of Electricity · Electric Power Research Institute (EPRI) · current