An investigation spanning several major U.S. grid regions has laid out a set of numbers that simply don't add up. Utilities across the country have received interconnection requests from very large power users — chiefly data centers — that now total more than 700 gigawatts. Actual power consumption by U.S. data centers today is estimated at roughly 70 gigawatts. The requests outnumber real usage by a factor of ten.
The gap has a name in the U.S. power industry: phantom demand. It refers to interconnection requests that get filed and queued but never materialize — the product of the same project being submitted to multiple utilities at once, and of companies with neither the capital nor the ability to build reserving a spot in line just in case.
Texas: from 48GW to 474GW
Broken down by region, the numbers get even more extreme. Texas's data center interconnection requests jumped from 48 gigawatts in 2023 to 474 gigawatts now — nearly a tenfold increase in three years. The Midwest, Mid-Atlantic and South combined have received roughly 270 gigawatts of requests. For comparison, tech companies' announced AI data center investment plans add up to more than $700 billion — and even if every dollar of that were spent on schedule, it still wouldn't buy enough capacity to match what's been requested.
The root of the distortion is that queuing itself costs almost nothing. The marginal expense of filing an interconnection request is close to zero, so developers submit the same project to several utilities simultaneously, build wherever power comes through first, and keep the other slots as backups. For any single developer that's a rational move. For grid planners, it means receiving a pile of letters of intent with no way to tell which are real.
Thomas Gleeson, chair of the Texas Public Utility Commission, put the dilemma bluntly:
“When you don't know what real demand is, you don't actually know how to build infrastructure for it.”
Once the toll gate went up, demand shrank fast
States have converged on the same response: put a price on the queue so that reserving a spot is no longer free.
Exelon began requiring applicants to post performance guarantees, and its demand forecast immediately dropped about 40%, down to 11 gigawatts. Ohio raised the cost of a grid interconnection study to as much as $100,000 per request, and AEP Ohio's pipeline of demand reservations shrank by more than half. Pennsylvania tells the same story: of more than 100 data center proposals, only 20 have actually advanced to filing for a permit.
Beyond financial thresholds, several states now require applicants to disclose their ultimate beneficial owner, screening out middlemen and shell entities. The tools aren't sophisticated, but the effect is immediate: the moment real money is on the line, demand forecasts shrink by 40% — which is itself an answer to how much of that 700 gigawatts was ever real.
Who ends up paying for the inflated forecasts
The cost isn't abstract. In the PJM grid region alone, inflated forecasts have added $29.4 billion to capacity costs — money that eventually flows through electricity rates onto residential and commercial bills. Build to the inflated forecast, and someone has to pay for the extra transmission lines, substations and capacity reserves. Build to real demand instead, and the grid risks falling short the moment one large project actually materializes.
Attorney Daniel Farris points to the other side of the story. He says operators who rushed in after spotting what looked like a gold rush are now learning the hard way just how difficult it is to actually build and energize these projects. Reserving a place in line costs nothing; breaking ground does — and that gap is starting to weed out a cohort of players.
The data center power narrative just split in two
Over the past year, U.S. policy has pushed almost entirely toward speed: shortening interconnection approval timelines, letting developers build their own power plants, treating AI data halls as priority projects. All of that assumed the demand behind it was real. Regulators are now getting the other half of the signal — that the demand figures are heavily padded, and the water needs wringing out before anyone talks about going faster.
A rough calculation shows why this still matters: even after cutting the duplicate filings and pure placeholders in half, the remaining 300-plus gigawatts still dwarf what the U.S. grid can add over the next five years. Real demand is genuinely rising fast — nobody can say exactly how fast. The takeaway for readers watching compute expansion collide with power constraints elsewhere: screen the demand first, then argue about where the electricity comes from.
Sources: Reuters investigation, ITHome, CocoLoop, Texas Public Utility Commission public statements; the 700GW request volume versus 70GW actual usage figures, Texas's 48-to-474 gigawatt increase, Exelon's 40% downward revision to 11GW, PJM's $29.4 billion capacity cost increase, and Pennsylvania's 20-of-more-than-100 permit filings have all been checked against public materials.