As America races deeper into the age of artificial intelligence, the geography of that race is shifting in ways that reveal both opportunity and strain. A new analysis of state-level AI computing capacity places Texas securely at the top of the leaderboard, while a separate body of research warns that the country is building too many data centers in the wrong places. Taken together, the reports highlight a national AI buildout proceeding at extraordinary speed—and running headlong into the limits of America’s power, water, and land resources.
The Texas Royalty Brokers study ranks Texas, Tennessee, and California as the country’s leading AI superpower locations, using a measurement that blends GPU clusters, power availability, chip deployments, and patent activity. Texas stands well ahead of the pack with 17 major AI clusters and more than 6.6 million H100-equivalent processors, a capacity unmatched in the U.S. The state’s 9.17 million kilowatts of power dedicated to AI makes clear just how large the gulf has grown between Texas and the rest of the field.
Yet if Texas represents the high-performance edge of AI infrastructure, Tennessee offers a picture of a rising contender. The state leads the nation in deployed AI chips with a remarkable 1.27 million, and ranks third in total H100 chip equivalents, helping secure its second-place ranking. California, long the nation’s innovation engine, lands in third, buoyed by its deep patent portfolio even as its power capacity trails far behind the emerging central-U.S. hubs.
A Need for Green Fields
But raw compute capacity only tells part of the story. A new report from Cornell University warns that many of the country’s most active AI regions sit precisely where resources are most constrained. Northern Virginia and Silicon Valley, already dense with hyperscale data centers, sit on stressed electrical grids and rely on water supplies strained by climate and population pressures. Meanwhile, a wave of new AI facilities is still being proposed for the American Southwest, where aquifers and river systems are already overburdened.
Cornell’s conclusion is that future data centers should move away from these high-risk hubs and toward regions that pair abundant water with untapped clean-energy potential. The Midwest and Plains states, Nebraska, South Dakota, Montana, and parts of Texas, rank best for long-term sustainability, even though they’re located far from the current centers of cloud computing.
The resource imbalance is a concerning issue. U.S. data centers now consume roughly four percent of America’s total electricity use, a figure expected to more than double by the end of the decade. The International Energy Agency estimates that data centers will draw 426 terawatt-hours by 2030, which is equivalent to the current annual electricity consumption of many medium-sized nations.
AI-heavy data facilities, in particular, use significantly more water and power than traditional data centers. A single generative AI query can burn ten times the energy of a standard web search, and hyperscale training clusters require continuous cooling cycles that pull heavily from local water systems.
With so much capacity pouring into a handful of states, the resource strain is growing. In 2023, data centers consumed 26 percent of Virginia’s total electricity supply, with double-digit shares in North Dakota, Nebraska, Iowa, and Oregon. Utility operators warn that without carefully targeted planning, regional grids could face steep price increases or even infrastructure failures.
Vast Upside?
Despite the challenges, some experts point to enormous economic upside. States that successfully plan and build for AI infrastructure stand to capture billions in investment and thousands of high-wage jobs. Northern Virginia’s experience, now hosting 13 percent of the world’s data center capacity, shows that careful zoning, targeted incentives, and investment in energy efficiency can generate durable regional advantage.
But the broader national picture is more complicated. AI infrastructure is expanding faster than the country’s ability to build the power plants, transmission lines, and water systems required to sustain it. Clearly, it would be beneficial if America’s AI map shifted toward states with the land and resources to support industrial-scale computing. At this point, the question is urgent: can the U.S. data center map align with the realities of natural resources?
