Power, Not Tax Breaks, Is Redrawing the U.S. Data Center Map
Artificial intelligence investment is reshaping the geography of U.S. commercial real estate and infrastructure, with the next wave of data center development likely to favor regions that can deliver ample, affordable power rather than simply the most generous tax incentives, according to the Institute of International Finance.
The IIF’s county-level analysis finds that U.S. data centers are broadly distributed, but the largest facilities are concentrated in less densely populated parts of the Pacific Northwest, Southwest, Southeast and Midwest. Those regions can more readily accommodate the energy-intensive hyperscale campuses needed to support AI computing.
The report identifies power availability, network connectivity, proximity to demand centers, land costs and infrastructure as the key forces shaping location decisions. State incentives matter, but do not by themselves explain the buildout, the IIF said.
Scale Changes the Map
The IIF examined approximately 1,500 existing data centers in the contiguous United States using the IM3 Open Source Data Center Atlas, a project led by the Energy Department’s Pacific Northwest National Laboratory. The sample represents an estimated 30% to 40% of the country’s 4,000 to 5,000 operating data centers.
Facility counts, however, can obscure the distribution of computing capacity. A traditional data center may require 1 to 5 megawatts of power, while a modern hyperscale facility can require 50 to 100 megawatts, according to the report. That difference means a small number of campuses can carry far more significance for power systems and real estate investment than a larger number of smaller facilities.
Virginia’s Data Center Alley, Silicon Valley and the New York-New Jersey-Connecticut tri-state region contain many facilities, but many are older, smaller or colocation sites serving third-party users. The largest hyperscale campuses are more concentrated in Oregon and Washington; Arizona, Colorado, New Mexico and Utah; the Carolinas, Georgia and Tennessee; and Illinois, Iowa, Minnesota and Ohio.
Power Becomes the Constraint
Counties with substantial power-generation capacity tend to host more data centers, particularly in Arizona, California, Illinois, Oregon, Texas, Virginia and Washington, the IIF said. The analysis does not establish causation, however: generation capacity may attract data centers, grow in response to them or reflect other local advantages.
The scale of expected electricity demand illustrates why power has become central to the investment thesis. The Electric Power Research Institute estimates that U.S. data centers consumed roughly 177 to 192 terawatt-hours of electricity in 2024, or about 4% to 5% of national usage. By 2030, consumption could increase to 380 to 790 terawatt-hours, representing 9% to 17% of U.S. electricity demand.
The Energy Information Administration expects total U.S. electricity consumption to reach 4.27 trillion kilowatt-hours in 2026 and 4.349 trillion kilowatt-hours in 2027, both record levels, as data centers and electrification lift demand.
That creates opportunities for utilities, power-equipment providers, transmission developers and real estate owners with properly zoned land and secured interconnection capacity. It also raises execution risks related to grid reliability, transmission constraints, construction schedules, water availability and local political opposition.
Incentives are Not Enough
The IIF found that tax incentives appear to have supported data center development in Iowa, Nevada, North Carolina and Texas. Yet California and Oregon host meaningful capacity despite lower incentive scores, while Virginia’s dominant position is not matched by the highest state-level incentive score.
The National Conference of State Legislatures identifies dedicated data center incentives in 38 states, often subject to investment, employment or other qualifying requirements. But the report cautioned that state-level indexes may not capture local property-tax agreements, utility arrangements or individually negotiated packages.
The result is a more demanding development formula. “The emerging geographic concentration suggests that the next phase of AI investment will favor locations able to combine reliable and cheap power, network connectivity, and proximity to urban centers, not simply the most generous tax treatment,” the IIF wrote.
For investors, the implication is that the most valuable data center asset may not be a completed building. It may be a site with land control, entitlements, fiber access and, above all, a credible path to power.