Texas froze new data-center grid hookups after its own queue hit 474 gigawatts -- and admitted it can no longer tell which requests are real

Texas stopped approving new data centers for connection to its power grid on August 3, after the state's own transmission operator admitted it had lost the ability to tell which of the requests piling up in its interconnection queue were real. Gov. Greg Abbott ordered ERCOT to verify funding, land control, water supply, and true ownership on every pending data-center project before advancing it further through the queue -- the first time a major U.S. grid operator has paused the process entirely, rather than simply trying to study it faster.

The number that triggered it: 474 gigawatts of large-load interconnection requests now sitting in ERCOT's queue, across roughly 1,800 projects, about 90% of them data centers. 474 GW (ERCOT's data-center interconnection queue, September 2026) is more than five times the grid's own record peak demand -- a queue asking, on paper, for more new electricity than the entire state currently draws at its hottest hour of the year.

The number that ate the queue

That figure did not appear overnight. ERCOT's own records put the large-load queue at 48.5 gigawatts in February 2023 -- a baseline from before the current AI buildout really started, when large industrial loads meant steel mills and hydrogen plants more often than server halls. By December 2025, ERCOT vice president of system planning Kristi Hobbs told the grid operator's board the queue had grown almost 300% over the prior year-end total, to 233 gigawatts, over 70% of it data centers. By June 18 of this year, ERCOT's own press release announcing a new review process -- called Batch Zero -- put the number at 438,000 megawatts, 89% data centers. Two and a half months later, it was 474 gigawatts and still climbing.

Hobbs put the problem plainly at that December board meeting: (ERCOT normally reviews large-load requests one at a time. Batch Zero was its own attempt, before the freeze, to study 75-megawatt-and-up projects in coordinated groups instead -- grouping requests so transmission upgrades could be planned around the whole picture rather than approved project by project.) "We have outgrown the process that was established for reviewing these large loads," she said. That was nine months before Abbott's letter, and the queue nearly doubled again in the time it took ERCOT to build a new process for it.

Interconnection queues work the same way at nearly every U.S. grid operator: a developer files a request, pays a comparatively small study fee, and gets a place in line -- with no obligation to ever build, and often no penalty for walking away later. That structure made sense when large industrial loads were rare and mostly serious. It breaks down when a single company can file for the same gigawatt-scale campus at five different sites in three different utility territories, planning to build at exactly one of them once land, power, and financing actually line up, and simply let the other four requests expire or sit unused. Every one of those speculative filings still counts toward the headline queue number until someone checks.

Texas isn't the extreme case, either -- it's just the biggest one to actually act. Outside Texas, ten major U.S. utilities reported roughly 270 gigawatts of their own large-load requests as of this summer, a figure that overlaps only partly with Texas's queue and is itself subject to the same duplicate-filing problem. New York moved earlier and differently, imposing a construction moratorium of its own in July while the ratepayer-cost fight was still working through Congress and five state legislatures -- a step short of ERCOT's interconnection freeze, aimed more at cost allocation than queue integrity, but part of the same summer of state-level intervention.

Verification, not prohibition

Abbott's August 3 letter didn't ask ERCOT to stop building data centers in Texas. It asked ERCOT to check, before advancing any more of them through the queue, whether each project can show it is providing its own power or truly relying on the grid; securing its own water or depending on local supplies; and funding itself independently or leaning on state and local tax abatements and grants. "Any data center project that fails to comply with the verification and audit process to protect the reliability and resilience of the Texas electric grid must be denied," Abbott wrote.

The reliability argument behind that line is straightforward even if the politics around it aren't: a grid operator has to decide, years in advance, how much new transmission capacity and generation to build, and it has to decide based on the demand it expects to actually show up. Build for 474 gigawatts and get 100, and Texas ratepayers have paid for transmission lines nobody needed. Build for a verified 100 and get 300 because the audit under-caught real demand, and the state is back to the rolling capacity shortfalls that made ERCOT a byword after 2021. Abbott's letter frames the freeze as protecting against the first failure mode; critics like BNEF are warning about a version of the second.

ERCOT's response was immediate and procedural: it delayed the Batch Zero classification notices originally due August 7, then filed for a good-cause exception with the Public Utility Commission of Texas ahead of the commission's August 20 open meeting. The PUCT granted it. ERCOT then promised to notify transmission and distribution providers of each project's conditional classification by August 31 -- and, as of a market notice issued that same day, pushed even that deadline back again, citing the time its own "data validation and due diligence" is taking. As of this writing, the audit's actual findings -- which of the 474 GW is real -- still have not been published, more than a month after Abbott's letter.

Batch Zero itself, approved by the PUCT on June 18 and now the process under audit, groups every project of 75 megawatts or larger into a single coordinated study instead of reviewing them one at a time -- a threshold ERCOT set specifically to catch data-center-scale loads while leaving smaller industrial customers on the faster existing track. ERCOT has called itself the first grid operator in the country to use a batch process for large loads at all. The freeze, in other words, landed on a review system barely two months old, built for exactly this problem and immediately overwhelmed by the scale of what it found.

"I think part of what Texas is trying to do is to create some order, and impose some transparency, on an industry." -- Tyson Slocum, director, Public Citizen Energy Program

The data-center industry's own trade group doesn't dispute that some of the queue is inflated -- it disputes how the audit will separate the real from the speculative. "There's a real need to distinguish between speculative projects and serious, committed investors," the Data Center Coalition's Dan Diorio said, without endorsing a grid-wide pause as the way to do it. That's the crux of the disagreement running through every version of this fight, in Texas and everywhere else it's playing out: nobody on record thinks all 474 gigawatts is real, and nobody on record can yet say how much of it is.

Pennsylvania reaches for the same lever

Fifteen days after Abbott's letter, Pennsylvania Gov. Josh Shapiro signed his own Executive Order 2026-05, and the numbers his office released to justify it map almost exactly onto Texas's ghost-demand problem, despite Pennsylvania regulating data centers through state permitting rather than a single grid operator's interconnection queue. Of more than 100 data-center projects the state has tracked through public databases, only 58 had engaged the Department of Environmental Protection about permitting at all. Only 15 had applied for even one DEP permit. Only 5 held every permit needed for a first construction phase.

Shapiro's order requires developers to make a binding commitment to the state's own clean-power and grid standards -- called GRID Requirements -- secure local community approval, cover the cost of any infrastructure their own demand requires, and it bars any nondisclosure agreement between a data-center developer and a state agency reporting to the governor. It also strips every data-center project of eligibility for Pennsylvania's fast-track permitting program, the opposite of the red-carpet treatment such projects were getting a year earlier. "If you can't agree to our strict requirements and get the community where you want to build to say 'yes,' you're not going to have the Commonwealth's support either," Shapiro said at the signing.

The mechanism differs from Texas's -- an executive order tied to state permitting, tax incentives and confidentiality practices, rather than a grid operator freezing interconnection outright -- but the diagnosis is the same: a headline pipeline number built almost entirely from filings that never reach construction, deployed by state and local officials, chambers of commerce, and the projects' own developers as evidence of an economic boom that most of it may never actually deliver.

Is the demand actually fake?

Not all of it, and not evenly. Where utilities have already tightened their own requirements, the results suggest a real -- if smaller -- core of genuine demand sitting underneath the speculative filings. Exelon cut its own tracked data-center demand by roughly 40%, to 11 gigawatts, after imposing stricter collateral requirements on new interconnection customers. AEP Ohio's tally fell by more than half after Ohio regulators required grid-connection study fees of up to $100,000 per project. Neither company eliminated its queue -- both still show gigawatts of demand that survived the tighter bar, which is closer to what a verified, credible pipeline probably looks like than either the original or the post-cut number alone.

"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," said Daniel Farris, an energy attorney at Foley & Lardner. The pattern nationally is the same shape as Texas's and Pennsylvania's: a Bloomberg review found electricity requests topping 700 gigawatts across the Midwest, Mid-Atlantic and South alone -- over ten times independent estimates of the data-center industry's actual current U.S. power use, and a number that necessarily includes the same kind of duplicate, multi-site, never-built filings driving Texas's queue.

Who wins, who loses

A queue built five times bigger than it needs to be doesn't just look bad on paper -- it makes it structurally impossible for a grid operator to plan the transmission lines and generation the real projects will actually need, and it hands every future rate case a padded baseline to argue from. That's the case for the freeze. The case against it is that verification, however necessary, doesn't fall only on the filers who deserve it.

There's a second audience for all of this beyond Texas ratepayers: every company deciding where to put its next data center is reading the same padded queue numbers when it picks a site, and every lender financing that project is underwriting against a grid-capacity assumption nobody has verified. A queue that overstates real demand five-to-one doesn't just mislead regulators -- it can mislead the developers themselves about how contested a given interconnection slot actually is, and mislead lenders about how quickly a financed project will actually get power. Sorting the real 10% to 60% -- Exelon's and AEP Ohio's post-verification ranges -- from the rest is not just a transparency exercise; it's the number every other financial decision in this specific buildout is quietly resting on.

The case the freeze goes too far

BloombergNEF's own modeling, published as the pause dragged past its original timeline, is the sharpest version of that exposure argument. The firm estimates the audit could delay 49.8 gigawatts of data-center load -- nearly a fifth of the entire U.S. development pipeline -- costing developers roughly $8 billion cumulatively by the first quarter of 2027 under a conservative assumption that 60% of the delayed capacity is AI-related, rising to $15 billion if it's closer to 100%. BNEF models a three-month delay pushing capacity originally expected online between the third quarter of 2026 and the first quarter of 2027 back to the second quarter of 2027 instead -- a real cost, in the firm's own framing, that lands regardless of whether any individual project was ever speculative.

None of this is the first time this year a headline electricity number tied to data centers has turned out to measure something other than what it implied. A separate fight, over who actually pays for the capacity that does get built, split Congress and five state legislatures this summer on numbers that looked comparable and weren't. The ghost-demand queue is a related but distinct problem -- not who pays for real capacity, but how much of the capacity everyone is planning around is real in the first place. Texas and Pennsylvania have each now put a number on that gap, using different mechanisms and arriving at strikingly similar shapes: a huge top-line figure, and a real core underneath it that looks like a fraction of the headline. Whether that ratio holds once ERCOT's own audit finally reports back is the thing actually worth watching -- not the 474 gigawatts, but what's left of it once someone checks. Every other state now weighing its own data-center pipeline, and every hyperscaler citing one as evidence of demand, is watching the same audit for the same reason: it is the first real attempt anywhere in the country to turn a queue number into a verified one, and until it publishes, every other state's own unaudited total deserves exactly the same skepticism Texas just applied to its own.