AI data centers request 700 GW of U.S. power, but much of the demand may be an illusion

America’s AI data center boom has produced a staggering new number: more than 700 gigawatts of electricity requests across parts of the United States, roughly enough to supply every home in the country. There’s just one problem. Nobody knows how much of that demand is real.

Texas is now trying to find out.

The state recently halted new grid connection approvals for data centers and other large electricity users as officials audit a queue that has ballooned from roughly 48 gigawatts in 2023 to more than 474 gigawatts today, according to a Reuters review of utility and grid data.

“Data centers have requested roughly as much electricity across the middle swath of the United States as it takes to power every home in the country, but much of that demand may be an illusion,” Reuters reported.

The surge is so extreme that regulators are questioning whether many proposed facilities will ever get built.

“When you don’t know what is real, you really don’t know how to build the infrastructure for it,” Texas Public Utility Commission Chairman Thomas Gleeson said at an industry conference in March.

The problem stretches far beyond Texas. Reuters found another roughly 270 gigawatts of data center electricity requests across 10 major utilities in the Midwest, Mid-Atlantic and South.

Taken together, requests from very large electricity users, most of them data centers, exceed 700 gigawatts. That is more than 10 times industry estimates for current U.S. data center electricity consumption.

The AI boom has created a ‘ghost demand’ problem

AI companies are spending at a scale rarely seen in the technology industry. Planned data center spending by Big Tech is expected to surpass $700 billion this year, fueling a race among developers and landowners to secure grid connections before competitors do.

That race has created a strange side effect. Developers can request enormous amounts of electricity years before a facility is built, sometimes before financing, customers, or final construction plans are in place. Similar projects may appear in multiple utility queues as developers search for the best location.

The result is a grid planning nightmare.

Utilities need years to build transmission lines, substations, and new generation. Build too little and regions could face electricity shortages. Build billions of dollars of infrastructure for speculative projects that disappear, and households and businesses could be left paying for assets that were never needed.

OpenAI Data Center (Image credit: OpenAI)

Evidence suggests the problem is real.

Chicago-based Exelon cut its estimate of high-probability data center demand by about 40% to 11 gigawatts after introducing stricter collateral requirements.

AEP Ohio saw its data center pipeline fall by more than half after state regulators adopted new rules that included grid connection study fees of up to $100,000.

“The entities that rushed into the space, because there was a sort of pot of gold, are maybe now learning the hard way just how difficult some of this is to actually construct and bring online,” Daniel Farris, an attorney at Foley & Lardner who advises data center developers and hyperscalers, told Reuters.

Yet stripping speculative projects from the queue does not make the underlying problem disappear. PJM Interconnection, which operates the largest U.S. electricity grid, is already dealing with major demand growth from data centers. Existing and forecast data center consumption helped drive a $29.4 billion increase in capacity costs across roughly the past four PJM auctions, according to grid monitor Monitoring Analytics.

“The reality is that the load is showing up, and generation is not at the pace we need it to,” PJM spokesman Jeff Shields said.

Texas starts asking who is really behind the AI data centers

Texas Governor Greg Abbott has ordered a broad audit to separate credible data center projects from speculative requests.

Developers will have to disclose who ultimately owns proposed facilities, closing a gap that previously allowed projects to identify an affiliate rather than the actual company behind them. Officials are seeking information on taxpayer incentives, water consumption and plans for on-site electricity generation.

“The PUCT and ERCOT cannot make decisions to guarantee grid stability and reliability based on substantially incomplete information,” Abbott said.

Texas is not alone in pushing back on the data center buildout. In July, TechStartups reported that New York had moved to halt new AI data centers amid concerns over rising electricity costs and environmental impacts, another sign that states are starting to scrutinize the infrastructure demands behind the AI boom.

Pennsylvania is moving in a similar direction.

Governor Josh Shapiro signed an executive order on August 18 imposing stricter permitting requirements and greater disclosure for projects of 25 megawatts or more. More than 100 data centers have been proposed across Pennsylvania, yet only 20 have applied for permits needed to move forward, a member of the governor’s office told Reuters.

Many have yet to secure an electricity source or a customer that can finance the project.

That gap may be the defining question hanging over America’s AI infrastructure race. The country clearly needs far more electricity for data centers, and credible projects alone could strain grids already struggling to add generation.

But 700 gigawatts of requests does not mean 700 gigawatts of data centers are coming.

Texas is betting that before America spends billions building infrastructure for the AI boom, it should first figure out which data centers actually exist beyond a spreadsheet.