Four hundred and seventy-four gigawatts. That is how much electricity data centers and other large users have asked to draw from the Texas grid, according to documents from grid operator ERCOT and Governor Greg Abbott’s office. In 2023 the figure was about 48 gigawatts. Nobody in Austin believes the new number is real, and the state has stopped pretending otherwise. Texas has become the first major data center hub in the United States to freeze new grid connections for the facilities while it audits what is actually behind them.
The industry has a name for the gap between requests and reality. It calls it ghost demand.
How the queue filled with phantoms
Connecting to a grid has historically been cheap to ask for and slow to receive. With Big Tech’s planned AI data center spending topping 700 billion dollars this year, according to Reuters, developers and landowners with access to power have rushed to file interconnection requests, often several for the same site, to hold a place in line. Many have no signed customer, no financing and no engineering plan. They are betting that a hyperscaler will want the location later.
A Reuters review of utility and grid data found that requests from very large power users across parts of the Midwest, Mid-Atlantic and South now exceed 700 gigawatts, more than ten times industry estimates of current US data center consumption. Ten of the largest utilities in those regions, including AEP Ohio, Southern Company and PPL, report roughly 270 gigawatts of data center demand requests between them. There is no common standard for how those figures are compiled, so some utilities count only signed contracts while others tally every inquiry.
What happens when someone asks for a deposit
The clearest evidence that much of this demand is illusory comes from utilities that have started charging for a place in line. Exelon, based in Chicago, cut its estimate of high-probability data center demand by about 40 percent to 11 gigawatts after imposing stricter collateral requirements, the company said in a July 30 investor presentation. In Ohio, AEP Ohio’s data center pipeline dropped by more than half after state rules introduced grid connection study fees of up to 100,000 dollars.
Texas is going further. Abbott’s order requires every project in the ERCOT queue to reveal who ultimately owns it, replacing earlier rules that allowed only an affiliate company to be named. Applicants must also detail whether they rely on taxpayer-funded incentives, how much water they intend to use and whether they plan on-site generation. The Public Utility Commission of Texas and ERCOT will run the audit, and any project that fails to comply is to be denied a connection.
Pennsylvania followed on August 18 with an executive order from Governor Josh Shapiro that tightens permitting for projects of 25 megawatts or more and demands disclosure of plans and end users. Of more than 100 data centers proposed in the state, only 20 have applied for the permits they need, an official in the governor’s office told Reuters. Most have neither a power source nor a customer.
The climate cost of guessing wrong
This is where the grid problem becomes a climate problem. Utilities plan generation years ahead. If they build for demand that never arrives, they overbuild, often with gas plants, and ordinary customers pick up the bill. If they discount the requests too aggressively and the real load does show up, the result is shortages and emergency measures that tend to favour whatever can be switched on fastest, which is rarely the cleanest option.
The costs are already visible. In PJM Interconnection, the largest US grid covering 13 states including Virginia’s Data Center Alley, growth in existing and forecast data center demand drove a 29.4 billion dollar increase in capacity costs for homes and businesses over roughly the past four auctions, according to grid monitor Monitoring Analytics. PJM’s own spokesman put the tension plainly to Reuters: the load is showing up, and generation is not arriving at the pace needed.
Real demand is still enormous
Strip out the phantoms and the picture is still daunting. Even after utilities slash their forecasts, they are left with enough substantiated data center requests to overwhelm grids that are already short on supply. Ghost demand does not mean AI’s appetite for electricity is exaggerated. It means the current queue is a terrible instrument for measuring it.
The Texas audit will be the first serious attempt to separate signal from noise at scale, and its findings will shape how other states, and the utilities that serve them, decide what to build over the next decade. Watch for the first numbers out of ERCOT. If the 474 gigawatt figure collapses the way Exelon’s and AEP Ohio’s did, the AI power crisis will look very different from the one currently being priced into electricity bills. For more coverage of AI’s impact on energy and climate, visit Mylistingo.
