The number that ended the Texas grid’s open-door policy is 474 gigawatts.
That is the scale of large-load interconnection requests ERCOT was carrying this summer, more than five times the grid’s record peak demand, with data centers accounting for roughly 90 percent of it. No power system on earth can build five times its peak in a few years. Texas did not treat that queue as fantasy. It treated it as firm demand, first-come and connect-and-go, the same energy-only bargain that made ERCOT the cheapest place in America to plug in a gigawatt of AI compute.
On 3 August, Governor Greg Abbott froze it. He directed the Public Utility Commission and ERCOT to verify and audit every data center moving through interconnection before any new one advances. ERCOT suspended its Batch Zero large-load classification notices, which had been due 7 August. The story looks like a pause. It is not. It is a repricing of who carries the risk when the grid runs short, and the answer is now the data center.
The audit is the smaller half
The audit itself is real and slow. ERCOT is working through about 300 Batch Zero projects, each 75 megawatts or larger, plus crypto facilities of 25 megawatts and up flagged for community-impact review. Requests for information go out from late August through September, with further rounds in October and November, and a verification report is due 10 December. The original Batch Zero study deadline of 9 April 2027 will slip. “Our goal is to head toward a December 10 filing,” ERCOT’s Chad Seely told the commission. “We will not have the study done by April 9, 2027.”
A verification exercise separates the projects that have signed leases, ordered transformers and lined up chips from the speculative placeholders padding the queue. That is useful. A 474-gigawatt number that is 90 percent data centers is not a forecast; it is a list of options, most of which will never be exercised. Auditing it shrinks the phantom.
But the audit does not change what a real, surviving data center is entitled to once it connects. The rule that does sits in a different docket, and it was final months before the pause.
Firm demand became interruptible demand
Senate Bill 6, signed 20 June 2025, defines a large load as 75 megawatts or more at a single site. For any large load interconnecting after 31 December 2025, transmission providers must build curtailment protocols so the load can be shed during a firm load-shed event. Read plainly: the newest and largest customers on the Texas grid are the first the operator can switch off when supply falls short, and they must be built to be switched off.
The obvious dodge is to bring your own power. Put a gas turbine or a wind farm on-site, sit behind the meter, and let the grid curtailment fall on the generator, not the load. Texas closed that door on 24 July. In Docket 59220, the commission approved a co-located arrangement and then held that emergency curtailment of a co-located data center is not capped by the capacity of its paired behind-the-meter generator. The concrete number is the point: two data centers sitting behind a single 265.5-megawatt wind facility were handed a cumulative 525.5-megawatt curtailment obligation. They can be ordered to shed nearly twice what their on-site generator can produce. The co-location does not buy immunity. It buys a bigger curtailment number.
The governing rule, 16 TAC 25.205, was adopted 26 March 2026 and is already final. So the sequence matters. Texas wrote the curtailment regime first, made it final in the spring, set the precedent in July, and only then paused the queue in August. The pause is the headline. The curtailment rule is the change.
Why the capital does not flinch
If large loads are now interruptible and behind-the-meter generation does not shield them, the intuition is that money leaves. It has not. On 6 August, two days after the moratorium, NRG announced a 1.2-gigawatt combined-cycle gas plant for an unnamed hyperscaler, roughly 3.2 billion dollars for the first phase, a minimum 15-year term, delivery targeted for late 2029, with a second identical phase optioned on top.
The structure is the tell. More than 95 percent of the project’s cash flow comes from capacity payments, not from power actually consumed. “We’re paid for the megawatts we build and make available, not for how much the data center runs,” NRG chief executive Robert Gaudette said. That sentence is the new Texas grid in miniature. The hyperscaler is no longer buying firm energy at a wholesale price it assumed would stay cheap. It is buying availability, and paying for firmness as a separate line item, because the state has stopped promising firmness as a default.
This is the quiet death of the energy-only premise for hyperscale AI. ERCOT’s whole design was that you do not pay in advance for capacity; you take the market price and, in the rare emergency, you ride it out. That worked when the largest loads were aluminum smelters and the queue was measured in hundreds of megawatts. It does not work when a single campus wants a gigawatt and the queue is five times the peak. So Texas is bolting a capacity discipline onto an energy-only market through the back door: curtailment rules that make the grid’s obligation conditional, and private contracts that make firmness something the data center funds itself.
The implication other states will copy
The lesson circulating in every other interconnection queue is not that Texas turned hostile to data centers. It is that Texas found a way to keep saying yes without promising the grid can carry them. NRG’s second-quarter Texas margins already fell 131 million dollars year over year as wholesale prices ran near 33 dollars a megawatt-hour against a 52-dollar planning assumption. Cheap power is good for the buyer and brutal for the builder, and no one builds 30 gigawatts of new generation on a spot market that keeps undershooting. Capacity payments and curtailable load solve both problems at once. The buyer gets its megawatts; the builder gets a bankable contract; the grid keeps a switch it can throw before the lights go out on everyone else.
For the AI industry, the number that matters is not 474 gigawatts of requests. It is how many of those gigawatts are firm, and the answer Texas just wrote down is: fewer than anyone in the queue assumed. The buildout continues. The terms changed. Power in the friendliest market in the country now comes interruptible by default, and the cost of making it firm is a line item the hyperscaler pays, not a promise the state makes.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.