America's AI data-center map is being redrawn by power, land, and local opposition

The geography of American AI is still Virginia-shaped. Fifteen of the world's 20 largest hyperscale data-center markets sit in the United States, and Northern Virginia alone holds roughly 12% of global capacity, with about 250 facilities inside Loudoun County, according to a Synergy Research Group tracker reviewed by TechRadar Pro. Northern Virginia plus the Greater Beijing Area together account for 17% of global capacity. But the leader board for the next set of builds is moving.

Texas recorded 71% capacity growth in the most recent year, more than double the 36% global increase. Indiana, Tennessee, and Guangdong entered the top 20 for the first time. Tokyo, Sydney, and South Carolina fell out. The churn shows that hyperscale siting is no longer converging on the original cluster. It is fragmenting, and the cause is a short list of structural drivers that Synergy's chief analyst, John Dinsdale, names plainly: customers, real estate, power, networking infrastructure, incentives, and political stability.

Each driver pulls in a different direction. The first two favored Loudoun County in the 2010s, when AWS and Microsoft wanted to be close to East Coast fiber and federal procurement. Northern Virginia won because of fiber density, Ashburn's Meet-Me rooms, and a customer base that included the federal government and the early enterprise cloud. Those advantages do not expire, but they no longer scale with the size of new AI training campuses, which now arrive at the multi-hundred-megawatt scale and need a different mix of land, power, and water than the original cloud buildout did.

The next two drivers favor anywhere with cheap land and surplus grid capacity. That is why Texas, Iowa, and Ohio have moved up the rankings. Oregon still anchors the Pacific Northwest for hydro availability. Dallas-Fort Worth is now one of the top US markets, sitting between the cheap power of the Permian basin and the fiber backhaul to the East. ERCOT's lower industrial power prices, combined with some of the largest contiguous parcels of undeveloped land in the country, are exactly the inputs a 500-megawatt AI campus needs.

Incentives, the fifth driver, are the most visible lever. Texas, Indiana, and Tennessee have all used tax abatements and sales-tax exemptions on data-center equipment to win anchor tenants. The same playbook has drawn public pushback in Loudoun and Prince William counties, where residents have organized around rising electric rates, water consumption, and the loss of farmland and forest. Local opposition has begun to slow new builds in Northern Virginia. Loudoun County, which for a decade approved projects at near-mechanical pace, is now taking longer to site new substations, and Dominion Energy has warned that transmission buildout in the region will not keep up with demand. Political stability, the sixth driver, increasingly cuts against the US as a whole. Dublin is the only European market in the global top 20, and Asian governments are courting hyperscalers with land grants and dedicated grid interconnects, which is how Guangdong displaced Sydney.

Concentration sits behind all of it. Synergy counts 62% of the world's hyperscale operators as headquartered in the US, and Amazon, Microsoft, and Google together control 57% of worldwide hyperscale data-center capacity, a market structure that means siting decisions are made in a small number of rooms. The new tracker data shows those operators spreading their bets rather than doubling down. New campuses are landing where the local stack works, and the list of what counts as a working local stack is longer than it was three years ago.

That redistribution is the story the 15-of-20 dominance stat hides. The US is still the gravitational center for AI compute, and the top 20 markets worldwide account for 60% of all hyperscale capacity. The next gigawatt, though, is likelier to be built near a wind farm in West Texas or a substation in central Ohio than on a cleared cornfield in Ashburn. Where power, land, and local politics line up, that is where the next data center will go.