The AI compute buildout has triggered a data center construction boom with few historical parallels. The scale is significant enough to reshape regional economies, power grids, and real estate markets — and the race to secure the necessary land, power, and water is already producing visible winners and losers.
The binding constraints are not compute or capital. Both are available in large quantities. The binding constraints are power and water. Data centers require enormous amounts of both: power for compute and cooling, water for thermal management. The locations with the right combination of cheap power (typically from existing grid infrastructure or renewable generation), adequate water rights, and favorable permitting environments are a limited set. Northern Virginia, central Iowa, central Texas, and parts of the Pacific Northwest have been the primary beneficiaries in the U.S. for these reasons.
The winners in the land rush are not primarily the hyperscalers — Microsoft, Google, Amazon, Meta — who are building for their own capacity. They are the real estate investment trusts and colocation operators who locked in land, power agreements, and water rights before the current demand surge made those resources scarce and expensive. Equinix, Digital Realty, and Iron Mountain are sitting on infrastructure that now commands pricing power they did not have two years ago.
The losers are the municipalities and utilities that did not anticipate the scale of demand and are now managing grid strain, water table concerns, and community resistance to industrial facilities that consume resources visibly while providing limited local employment.
The buildout will continue regardless. The grid will have to keep up. In most of the country, it is not currently designed to.