On August 3, 2026, Texas stopped letting new data centers advance through the grid-connection process. Governor Greg Abbott directed the PUCT and ERCOT to conduct a comprehensive verification and audit of every data center advancing through ERCOT’s interconnection process, and ordered that no additional data centers be approved until the audit is complete. ERCOT halted further approvals to energize large-load (75 MW+) data center interconnections and virtual currency mining facilities.
Seven weeks later, on September 21, Abbott widened the pause: he directed the Texas Commission on Environmental Quality (TCEQ) to halt all permits sought by data centers until the ERCOT audit finishes, and barred any state agency from moving forward with data-center approvals until the needed information is in hand. The trigger was a number that had stopped behaving like a plan: roughly 474 gigawatts of requests, more than five times the record peak of just over 90 gigawatts, with about 90 percent attributed to data centers. Texas is the fastest-growing data center market in the country. The audit is the moment Texas decided speed without verification was the problem.
The number, taken apart
Start with what the 474 gigawatts actually is, because it is routinely misread as a forecast. It traces to ERCOT’s June 2026 snapshot of tracked large-load requests, presented to the Texas Senate Business and Commerce Committee on July 29, 2026: 474.7 gigawatts of requests, cumulative through 2033, about 90 percent for data centers. It is a list of requests, not a load forecast, and the future-year figures are cumulative requested ramps.
The breakdown by maturity is far more revealing than the headline. Of the total, 284.3 gigawatts had no study submitted to ERCOT, and another 135.5 gigawatts had a study under ERCOT review. Just 9.9 gigawatts had met Section 9.4 requirements only, and 36.0 gigawatts had met Section 9.4 and 9.5 requirements. Only 3.2 gigawatts were approved to energize but not yet operational, and 5.9 gigawatts were observed energized. Only about 9.1 gigawatts had approval to energize.
Two cautions belong on the number. Different counts measure different things: a separate, narrower ERCOT large-load queue stood at about 226 gigawatts in November 2025, up from 63 gigawatts at the end of 2024, with data centers accounting for about 73 percent of that total. The 474-gigawatt figure is the broader tracked-requests tally, not the same queue. And the officials closest to the number were already discounting it: ERCOT president and CEO Pablo Vegas said in April 2026 that the queue overstated likely future load growth, and PUCT chairman Thomas Gleeson said in June that much of the queued demand was speculative and unlikely to materialize.
How the queue got to 474 GW
The queue grew the way queues grow when a reservation is cheap and a position in line is valuable. Developers routinely filed multiple requests for alternative sites for the same project, some purely speculative, to hold a place while deciding whether to proceed. Some requests had no customer behind them. Because each request had to be studied against the loads ahead of it, speculative filings slowed the studies behind them — and because the forecast drew on queue positions, each one also made the forecast look larger.
The process itself was built for a different era. ERCOT used to review interconnection studies individually, under rules designed for roughly 40 to 50 large loads at a time. Then 2025 brought 225 new large-load requests by mid-November alone, and the tracked total climbed from 63 gigawatts at end-2024 to 474 gigawatts by June 2026. ERCOT’s energy-only market design, which carries no capacity-market signal, put no price on reserving capacity or on the transmission upgrades a new load would trigger. None of this means Texas has 474 gigawatts of data centers coming. It means Texas had 474 gigawatts of asks, in a system that had never required the askers to prove they meant it.
Senate Bill 6: the solvency test already on the books
The audit did not arrive in a vacuum. Texas enacted Senate Bill 6 in the summer of 2025, adding Section 37.0561 to the Public Utility Regulatory Act, aimed at protecting retail ratepayers from stranded transmission costs. The law set the framework for large loads of 75 megawatts or more: prospective customers must demonstrate site control, fund interconnection screening studies, disclose substantially similar requests elsewhere in the state, accept ERCOT-directed curtailment during emergencies at facilities interconnected after 2025, and deploy qualifying backup generation on notice.
On September 18, 2026, the PUCT adopted new rule 16 TAC Section 25.194, the SB 6 interconnection standards, effective October 8, 2026. Before a large load can be included in an ERCOT interconnection study, the customer must sign an intermediate agreement, make certain disclosures, and post financial security of $50,000 per megawatt — $5 million for a 100-megawatt project. The commission replaced a proposed non-refundable fee with a flat $100,000 fee plus size-tiered study charges, and gave projects a 24-month grace period before unutilized capacity must be flagged. The security is a deposit against walking away — a solvency test, with the audit as a transparency test layered on top.
Batch Zero: the grouped study the audit interrupted
The audit also interrupted a new grid-planning system that had only just been approved. On June 18, 2026, the PUCT approved ERCOT’s Batch Zero process, its first grouped study of large electricity users. Instead of studying large connection requests one at a time — a project-by-project evaluation ERCOT described as lengthy and repetitive — Batch Zero groups qualifying projects of 75 megawatts or more into a single study, so ERCOT can assess future demand at once, allocate available grid capacity fairly, and identify needed transmission upgrades. ERCOT’s July 29 presentation put preliminary Batch Zero-eligible load at approximately 205 gigawatts: a preliminary eligibility figure, not an approval to connect.
Batch Zero barely got started before the audit stopped it. ERCOT had planned to issue project classifications by August 7. In an August 3 market notice, ERCOT said it would not issue the classifications by August 7 and would seek a good-cause exception at the PUCT’s August 20 open meeting. The audit then folded into the Batch Zero machinery: ERCOT issued a Batch Zero Eligibility request for information on September 9 and a State and Community Impact request for information on September 14, with responses due October 12 with a notarized attestation. ERCOT expects to publish a report on its review by December 10, 2026.
What the audit examines
The August 3 directive set the audit’s scope, and the follow-ups widened it. The original directive ordered a comprehensive verification and audit of all data centers in the interconnection process for compliance with existing ERCOT and PUCT requirements, plus state and community impact topics: state financial assistance, own power versus grid reliance, water sources and uses, community-impact mitigation, and ownership. Any data center that fails to comply may be denied interconnection to the ERCOT grid.
The September 14 follow-up brought water to the center. Abbott directed the Texas Water Development Board (TWDB) to partner with ERCOT on the audit and to compel data centers to respond to surveys on water consumption, supply sources, and water-efficient technologies — with civil and criminal liability for non-response and denial of TCEQ permits. TWDB was expected to report on its enforcement activities to the governor by October 14. The September 21 directive then closed the loop: TCEQ must halt all permits sought by data centers until the ERCOT audit completes, and report compliance to the governor’s office by October 19. The pause applies to all data center projects, including fully islanded facilities on behind-the-meter generation — a project cannot dodge the audit by building its own power plant and skipping the grid.
Note what the directives do not contain: a deadline. Abbott’s August 3 letter set none, though ERCOT aims to complete its review by December 2026. The governor has also pledged to pursue legislation in the session convening in January 2027 that would codify cost obligations and could require new data centers to add to Texas electric capacity rather than only to demand, mandate water-efficient cooling, require annual usage reporting, repeal sales-tax exemptions, and require setbacks and noise limits. Those are proposals, not settled law.
The water question is the second grid
Water deserves its own section because it is the issue that turned a grid-management exercise into a statewide development pause. A data center’s electricity demand is visible on the grid; its water demand is drawn from local supplies, and Texas communities had begun asking why facilities straining local water should be approved on grid paperwork alone. The audit treats water as a gating item, not a footnote: sources, consumption, cooling technology, and efficiency measures are all in scope.
The early evidence on disclosure was not encouraging. An early-October analysis by Andreessen Horowitz’s American Dynamism team reported that, as of June, only 28 of 377 companies had answered a state survey on resource use — a concrete illustration of why the governor moved from requesting information to compelling it. The same analysis described developers filing requests across several sites with no customer attached — the audit’s premise: the state was being asked to plan a grid and allocate water around projects whose basic facts it could not verify.
The delay math, carefully attributed
The most quoted estimate of what the pause costs comes from BloombergNEF. In an August 5 report, BNEF said Abbott’s August 3 order could delay 49.8 gigawatts of new data center electricity demand from advancing on the ERCOT system — nearly 20 percent of the 253-gigawatt U.S. development pipeline. The firm modeled a three-month delay on additions forecast for 3Q 2026 through 1Q 2027, slipping to 2Q 2027, and estimated cumulative revenue at risk of just over $8 billion if 60 percent of delayed capacity is AI-related, rising to roughly $15 billion at a 100 percent AI-compute mix.
Treat that as what it is: a scenario built on assumptions, not a measured loss. The $15 billion figure is the top of a range that moves with the AI share of delayed capacity, and it assumes a uniform three-month slip that the audit’s actual timeline may or may not produce. More telling: BNEF forecast only 1.2 gigawatts of ERCOT data center additions between 2Q 2026 and 1Q 2027 — against 474 gigawatts of requests. The queue was never a construction schedule. The pause, for its part, does not shut down operating facilities, and contracts for already-approved capacity are unaffected; its cost falls on real projects facing months of delay, financing uncertainty, and contracts to renegotiate.
What it means for developers
For developers, the pause reprices every stage of a project. The first is diligence: a queue position is no longer an asset in itself. What matters now is the file behind it — site control, water sourcing, ownership and incentive disclosure, on-site generation plans, community mitigation. A project with a thin file is not merely delayed; under the directive, it can be denied interconnection outright. The second is capital: financial security of $50,000 per megawatt under the PUCT rule, plus carrying a project through a months-long audit with no certain end date, means speculative filings now carry real costs. The third is strategic: projects that can demonstrate maturity stand to gain from the thinning of the queue, because every speculative project removed sharpens the signal for the ones still standing.
What it means for other states
Other states are watching Texas, the test of whether verification scales. The national turn is already underway: Virginia imposed a $0.011/kWh electricity consumption tax on data centers effective July 1, 2026; Ohio, Maine, and Illinois each paused or curtailed data center tax incentives in 2026; TVA’s new data center tariff, effective October 1, charges new and expanded loads above five megawatts roughly $1.5 million per megawatt to reserve capacity; PJM’s new Large Load Registry, effective October 12, will track actual project status for loads of 50 megawatts or more. Different mechanisms, same logic: make the reservation costly or verifiable enough that only serious projects file for it. Texas matters most because it is the biggest test and the most unlikely convert.
What to watch
Four dates and one question will determine whether the pause becomes a precedent or a cautionary tale. The first is October 19, when TCEQ must report compliance with the permit halt. The second is December 10, when ERCOT expects to publish its report on the State and Community Impact RFI responses — the document that will show how much of the queue survived verification. The third is the January 2027 legislative session, where the governor’s proposed package faces a legislature historically sympathetic to development. The fourth is the restart of Batch Zero itself.
Underneath all of them is whether the industry builds a planning system that measures commitments instead of counting reservations. Texas did not pause because data centers are unwelcome; it paused because its planning system could not tell a real project from a placeholder. The audit is a filter, not a verdict. Its success will be measured in the gap between the load developers request and the load they actually build.