The South is getting 754 new data centers. The grid plan to feed them is in the Trump administration's crosshairs.

On 30 June 2026, an X post from the markets account Unusual Whales circulated a striking regional arithmetic: of all the data centres planned across the United States, 754 sit in the South — a 62% increase over the region's current installed base, and the largest single regional concentration anywhere in the country. The figure frames a quieter but more consequential collision already underway in Washington. A day earlier, on 29 June, TechCrunch reported that the Trump administration's regulatory posture is now threatening roughly 92 gigawatts of new electricity supply, alongside $121 billion in solar and wind investment — the two technologies that have done the heaviest lifting on new US generation capacity in recent years.

That is the argument this piece is built around. The places racing to host the AI economy are the same places whose grids are about to find out whether federal energy policy is a help or a hindrance. The South is winning the data-centre siting war on cheap land, mild climate, and right-to-work labour law; whether it can also win the power-supply war depends on choices being made right now in Washington, in state capitals from Atlanta to Austin, and in the boardrooms of utilities that have spent two decades burning gas and are being asked, almost overnight, to plan around machine-learning load growth that doubles every few quarters.

What the South actually has on order

The 754-project figure is the headline, but the geography inside it matters. Unusual Whales's tracker aggregates planned facilities across the Census-defined South — a band that runs from Virginia, where the Loudoun County cluster has already produced the world's densest concentration of cloud capacity, down through the Carolinas, Georgia, Tennessee, Alabama, Mississippi, Louisiana, Arkansas, Texas, and Oklahoma. The 62% jump is measured against the region's current operating total, which already includes the legacy hyperscaler campuses around Ashburn, the Dallas–Fort Worth ring, and the newer campuses in central Texas tied to both cloud and crypto-compute workloads.

Three things are worth flagging about that distribution. First, the South's lead is structural, not accidental: cheap land, fast permitting at the county level, a deregulated wholesale market in most of the region (ERCOT in Texas, the Carolinas' partially-deregulated markets, the Tennessee Valley Authority's unique federal-corporation status), and an industrial customer base that was already being courted by utilities trying to replace declining textile and paper-mill load. Data centres slotted in as anchor tenants the way aluminium smelters once did. Second, the demand profile is unlike anything the grid has had to plan around. A single hyperscale campus can pull 300 to 1,000 megawatts continuously, with a power-usage effectiveness that hyperscaler operators have spent a decade driving downward but which still leaves the total draw at multiples of any previous industrial customer. Third, the 62% figure is planned capacity — projects announced or in interconnection queues, not yet under construction. The gap between "planned" and "delivered" is where the next eighteen months of regulatory fights will be fought.

Where the federal lever is now pulling against the buildout

The 29 June TechCrunch story is the second hinge of this argument. According to the outlet's reporting, the Trump administration's moves are on track to jeopardise roughly $121 billion in new solar and wind power, with 92 gigawatts of new electricity supply at risk — a number large enough to cover the entire present-day peak demand of several mid-sized industrialised economies. The mechanism, as TechCrunch describes it, is a combination of permitting friction, transmission-tie-in delays, and what the piece characterises as regulatory "red tape" applied unevenly across generation types. The framing in the piece itself is not neutral: it presents the administration's posture as a brake on the two technologies that have been the biggest contributors to new US capacity in recent years. That framing matters, because the same administration is simultaneously welcoming the data-centre buildout politically and rhetorically.

The dissonance is the story. The political coalition in Washington that talks most loudly about American AI dominance, semiconductor reshoring, and frontier-model leadership is the same one whose energy-policy decisions are slowing the generation mix that those data centres will actually need to run on. Gas can backfill some of it — and will, especially in the short term — but gas turbines have lead times of their own, gas-pricing is now a live political issue (Trump told reporters on 29 June that "oil and gas prices keep falling," per Unusual Whales's transcript), and the LNG export buildout competes for the same molecules that a doubling of data-centre demand would consume.

The political economy underneath the headline number

There is a tempting read of this as a culture-war story — renewables-versus-fossil, Trump-versus-the-green-deal. That read misses the more important structural fact. The data-centre boom is being driven by hyperscaler capex decisions (Microsoft, Google, Amazon, Meta, Oracle, and a tier of newer entrants including CoreWeave and Lambda) that were planned on a five-to-ten-year horizon under the assumption that some form of low-carbon generation would be available at industrial scale and industrial price by the time the buildings came online. That assumption is now in question, not because the technology has stopped working but because the permitting and interconnection apparatus is being asked to do more in less time than any US energy transition has previously demanded.

A second structural point: the South's 754-project pipeline is not uniformly Republican-voting territory. Virginia, North Carolina, Georgia, and Texas all have substantial metropolitan economies with Democratic-voting cities and suburban rings that have actively courted data-centre capex. Atlanta, in particular, has made data-centre recruitment an explicit part of its economic-development strategy for a decade. The politics of welcoming data centres (local jobs, local tax base, prestige projects) and the politics of powering them (transmission lines, gas-pipeline expansions, solar farms, wind installations) are not the same politics. Communities that want the server farms often do not want the high-voltage lines or the wind turbines that would feed them. The Trump administration's posture, read most charitably, is an attempt to force a faster buildout by picking winners among generation types — gas and nuclear over solar and wind. Read less charitably, it is a slowdown on the cheap options without a commensurate acceleration on the expensive ones.

Counter-narrative: this is also exactly what the administration says it wants

A fair reading has to include the White House's own framing. The argument inside the administration — visible in Trump's own remarks on 29 June about oil and gas prices continuing to fall, in the broader tariff-and-energy package being negotiated with European counterparts (Trump told reporters the same day that numerous European countries are discussing imminent implementation of a digital services tax on American companies, per Unusual Whales), and in the policy framework being assembled around "energy dominance" — is that the AI buildout is best served by cheap, abundant, dispatchable generation, and that the previous administration's tilt toward intermittent renewables introduced cost and reliability penalties that the new administration is correcting.

There is internal coherence to that argument. Gas turbines, properly procured, can be permitted and built faster than utility-scale solar-plus-storage in many US jurisdictions. Nuclear, despite longer lead times, offers the baseload profile that hyperscalers have begun signalling they want for next-generation campuses. Hydrocarbon dispatchability gives grid operators a firmer hand than forecasting wind and sun. And the political base that returned Trump to office in 2024 is, on the evidence of polling throughout the cycle, more sceptical of renewable mandates than it is of data-centre construction.

What the argument struggles with is scale. Ninety-two gigawatts is not a marginal adjustment. It is the equivalent of roughly ninety large nuclear reactors, or roughly two hundred modern combined-cycle gas plants, or roughly the entire installed wind fleet of the United States built twice over. The administration's own rhetoric suggests it intends to substitute gas and nuclear for the solar and wind capacity at risk, but the in-service dates on new nuclear capacity in the United States are still measured in decades for greenfield sites, and gas supply chains are now competing with LNG export expansion for the same molecules.

Stakes: who wins and who loses if the trajectory holds

If the federal slowdown on renewables persists and the substitution toward gas and nuclear does not arrive at scale in time, three groups bear the cost. First, the hyperscalers themselves: their announced capex plans were underwritten on the assumption of available, affordable, increasingly-clean power, and any sustained tightening of that supply pushes marginal cost upward and timeline outward. Second, the Southern utilities and grid operators: they sit between the data-centre customers, who have begun demanding renewable-purchase agreements as part of their sustainability commitments, and a federal posture that is making those agreements harder to honour. Third, ratepayers in Southern states, who will absorb the cost of whichever generation mix is built to serve the new load — and who have, in jurisdictions from Virginia to Georgia, already begun seeing industrial-customer rate classes shift to reflect hyperscale demand.

The winners, in the near term, are gas turbine OEMs (GE Vernova, Siemens Energy, Mitsubishi Power), nuclear restarter projects (TVA's Clinch River site, X-energy's small modular reactor programmes, the reopened discussion around Palisades in Michigan), and the engineering, procurement, and construction firms positioned to build whatever generation mix actually gets permitted. In the medium term, the winners could include the Southern utilities that successfully thread the needle — Duke, Dominion, Georgia Power, TVA, NextEra, Vistra — but only if they can execute on a buildout that is, by any historical standard, unusually fast.

What the sources do not tell us

A note of epistemic honesty. The 754-figure and the 62%-increase comparison come from a single X post by Unusual Whales, which is a markets-data account with a track record of surfacing useful aggregations but which is not a primary-source outlet; the underlying methodology — which planned projects are counted, how "current" capacity is defined, what geographic boundaries are applied — is not detailed in the post itself. The 92-gigawatt and $121-billion figures are from TechCrunch's reporting on the Trump administration's posture; the outlet cites unnamed administration sources and frames the piece as critical of the policy. Neither number is independently audited in the source material available for this piece. The administration's counter-argument that gas and nuclear can substitute for the threatened capacity is asserted rhetorically in Trump's public remarks but is not, in the source material, accompanied by a published substitution plan, a permitting timetable, or a capex commitment at the scale that would be required. Readers should treat all four numbers as the best available contemporary estimates, not as settled accounting.

What remains genuinely uncertain is whether the substitution is intended at all, or whether the federal posture is better read as a willingness to accept slower AI-buildout timelines in exchange for a generation mix that the political base prefers. That is a question the source material does not resolve, and one that will be answered, if it is answered at all, by what gets permitted between now and the 2026 midterms.

This piece leans on Unusual Whales as a data aggregator for the regional pipeline figures and on TechCrunch's 29 June reporting on the federal energy posture. Where the two converge — Southern buildout colliding with federal permitting friction — the conclusion is the same: the next eighteen months will be defined less by how many data centres are announced than by which generation mix actually gets built to power them.