Data-Centre Capacity as a Real-Estate Category: Power, Land, and Interconnect as the Actual Constraints

Data-centre real estate has been the strongest performing US real-estate category for three consecutive years, and the trade press coverage of the category has been unusually good. What is less clearly said, even in that coverage, is that the category has stopped being a real-estate story in any meaningful sense. The binding constraints on the specific developments actually being built in 2026 are power availability, land that sits next to power availability, and interconnect capacity between the two. The building is the least constrained input, and treating the category as a building-supply problem produces the wrong strategic conclusions.

The first observation is that utility interconnect timelines have become the practical development timeline. The specific US utilities serving the primary data-centre corridors, Northern Virginia, the Dallas-Fort Worth area, Phoenix, and the specific Ohio and Iowa markets that have absorbed hyperscaler demand, are quoting interconnect timelines that are meaningfully longer than the physical construction timeline for the buildings themselves. Developers who secured land and permits in the last two years but did not lock in interconnect capacity are, in the current environment, holding sites that cannot be built to their planned load and are being repriced accordingly.

The second observation is that transmission-level constraints are now propagating back to individual utilities in ways they were not three years ago. A utility that could historically add load to serve a large customer through routine local upgrades is, in several of the most active corridors, discovering that the additional load requires upstream transmission commitments the regional operator cannot promise on the timeline the customer requires. The result is a slowdown that shows up first as pushed timelines and eventually as a rationing conversation the industry has not had before.

The third observation is that behind-the-meter generation, on-site combined-cycle capacity, and specific power-purchase structures with independent generators have moved from marginal to central in the deal structure for the largest new developments. The specific hyperscalers driving demand have, in the last eighteen months, become materially more involved in the power side of their own development pipeline than they were in the previous cycle, and the deals actually announced show the shift clearly. Real-estate investment trusts and traditional data-centre developers competing for the same tenants without a comparable power-side capability are being outbid on the specific opportunities where power certainty is the deciding factor.

The fourth observation is that the water-and-cooling conversation, which has largely been treated as an environmental impact story, is becoming an operational availability story in the specific arid-climate markets that have absorbed significant capacity. Municipal water allocations for new development are being restricted in ways that constrain the specific cooling architectures currently deployed at scale, and the shift to liquid-cooling and dry-cooling designs is being driven as much by supply constraints as by efficiency preferences. Developers who have not modeled water availability as a binding constraint in Arizona, Texas, and specific California markets are, in current planning conditions, working from an incomplete map.

The fifth observation is that AI training and inference workloads have changed the specific building specification enough that the retrofit of existing capacity for the new use is a meaningful capital project rather than a straightforward reconfiguration. Power density per rack, cooling requirements, and network fabric architecture are all materially different from the earlier cloud-generation specification, and the specific existing buildings that can be retrofitted at reasonable cost are a smaller share of the standing stock than the initial commentary implied.

For a real-estate investor with data-centre exposure, the working question is whether the current portfolio has been evaluated against the actual binding constraint, which is the certainty of power delivery on the required timeline, rather than against traditional real-estate criteria of land quality, market fundamentals, and building specification. Portfolios that treat the category as a real-estate category continue to underprice the specific assets with locked-in power and to overprice the specific assets that are still awaiting utility clarity.

For a utility, the parallel question is whether the interconnect queue and the transmission planning are being resourced against the actual demand pipeline or against the historical load-growth assumptions that shaped the previous decade of planning. Utilities that continue to run against the old assumptions are, in the specific markets under most demand pressure, becoming the effective bottleneck on economic activity that the trade press has begun to describe them as.