Data Center Real Estate
The buildings, the land, and the power-grid interconnects underneath the racks.
What this layer does
A modern AI data center is a 100–500 MW industrial facility that has to land near transmission capacity, water (for cooling), and skilled construction labor. Permitting and grid interconnection now stretch 4–7 years in tight markets. As a result, the real scarce asset isn’t the building — it’s powered land: a parcel with a signed utility commitment for hundreds of megawatts.
Business models: (1) wholesale colo (lease shell + power to a hyperscaler), (2) build-to-suit hyperscale, (3) retail colo (smaller multi-tenant), and (4) self-build by hyperscalers on their own land. The investable layer is mostly REITs and a wave of bitcoin miners with powered sites being repurposed for AI.
Sub-categories
The two pure-play public DC REITs, plus Iron Mountain’s growing DC business. Long-duration assets with hyperscaler tenants.
Build-to-suit hyperscale platforms. Almost entirely owned by PE / infrastructure funds. Track for IPO or acquisition.
Bitcoin miners with powered substations, converting some/all capacity to AI hosting. Wide range of execution quality.
Most new AI capacity is going up on hyperscalers’ own land, not REIT property. Material driver of construction and electrical demand.
Owners of parcels with utility commitments. Increasingly the bottleneck asset of the AI build-out.
Geographic exposure outside the US grid bottleneck.