Data center geography points past tax incentives to power and land

The Institute of International Finance’s just-released report, covered by Connect CRE, works through the Pacific Northwest National Laboratory’s IM3 Open Source Data Center Atlas and finds the country’s data centers splitting into two patterns. The largest concentration sits in major population centers, with Virginia’s Data Center Alley and Silicon Valley as the reference points, and facilities there tend to be smaller than centers elsewhere. The larger hyperscaler campuses sit in less-populated stretches of the Pacific Northwest, the Southwest, the Southeast and the Midwest.

For private capital the split marks two different trades. Metro data centers sit close to demand but on land and grid capacity shared with the surrounding economy, which keeps typical footprints down, while campuses in the less-populated regions sit where large parcels and power are easier to assemble, which is what lets them reach scale. The report’s authors caution that the data does not settle which way causation runs—available power may attract the centers, or expected demand may pull generating capacity toward them—and that ambiguity determines whether the scarce asset being purchased is land or power. The report’s honesty about it should be read as an underwriting instruction: as this publication has argued, power and land are the real constraints in this buildout, and the compute is the part capital can buy last.

The tax point is made in one sentence and deserves the emphasis. “The evidence suggests that tax treatment is one element of the location decision, alongside power availability, infrastructure, land costs and proximity to demand,” the authors wrote, adding that further research is needed. The wording refuses to dismiss incentives, but it also refuses to treat them as the lead variable; coming from an analysis built around actual facilities, that is a warning to location coverage that scores projects by abatement size.

For allocation, the lesson is to change the order of the screen: start with the utility map and the land parcel, layer in proximity to users, then apply the tax numbers. A facility with a weak position on power and land is not rescued by a generous abatement, and a counting exercise that tallies existing data centers describes what has been built, not what the next site can support. Read that way, the IIF atlas points to where the next dollar of data-center capital can actually be deployed.