3 SEP 2026 — Governor Greg Abbott has told the Public Utility Commission of Texas and the grid operator ERCOT to stop signing new large-load interconnection agreements and to audit every request already in the queue. The queue has grown from roughly 48 gigawatts in 2023 to more than 474 gigawatts. That figure is not a demand forecast. It is a register of free options, and in the three places where somebody has recently charged for one, most of the queue disappeared.
What is known
The directive was issued in late August. It halts new interconnection agreements above a review threshold while the audit runs, and it asks developers already in the queue to disclose ultimate ownership rather than the shell entity that filed, any state or local incentives claimed or expected, water use, and plans for on-site generation. The audit is aimed at completion by December, and its results are expected to feed the next long-term load forecast.
ERCOT's all-time record peak was 91.089 gigawatts, set on 22 July 2026, which itself broke the 85.5 gigawatts recorded in August 2023. The queue is therefore more than five times anything the Texas grid has ever had to serve at once.
Thomas Gleeson, who chairs the Texas commission, put it plainly: "When you don't know what is real, you really don't know how to build the infrastructure."
A queue counts filings, not buildings
An interconnection request is cheap to submit and valuable to hold. It reserves a position in a line that is served in order, so the rational move for a developer who might build is to file early, file wide and decide later.
Filing wide means the same building can appear in several queues at once while its owner compares terms, and nothing in the process deduplicates them. It also means a request can be lodged by a company that has neither the capital nor the customer to finish, because the cost of being wrong is a withdrawn application.
The industry term for the resulting inflation is ghost demand. It is a useful phrase as long as it is not read as fraud. Most of these filings are ordinary optionality behaving exactly as the rules reward.
Three places where the option was priced
For evidence that the queue is padded, skip the audit; it has not happened yet. Look instead at three utilities and states that recently made a request cost something.
AEP Ohio began charging roughly US$100,000 for a connection study. Its data-centre pipeline fell by more than half. Pennsylvania's Governor Josh Shapiro signed an executive order on 18 August imposing stricter permitting on projects of 25 megawatts or more; of more than 100 proposed data centres in the state, 20 applied. Exelon revised its own data-centre demand down by 40 per cent, to 11 gigawatts, on 30 July.
A fee, a permit and an internal revision produced reductions of a similar order. When three different mechanisms converge like that, the pattern is worth more trust than any single figure.
What the freeze does not settle
The queue can be inflated while real growth continues, and the audit will not settle one by measuring the other. United States data centres draw somewhere around 60 to 70 gigawatts today, and the capital being committed to build more is enormous: the largest technology companies are spending on the order of US$700 billion this year.
Treat the 474 figure as close to meaningless as a forecast. The underlying load is rising and is hard to plan for, and a planner who discounts the queue by 80 per cent still has to build something.
The cost of getting this wrong is asymmetric, and that explains the freeze. Build for phantom load and ratepayers are left holding transmission they do not need. Build for real load that was hidden inside a discounted queue and the grid comes up short during a heat event. Texas has been on the wrong side of the second failure before.
The same ratio, on a much smaller grid
Southeast Asia has the identical gap between commitment and consumption, at a scale its grids can absorb far less easily. Tenaga Nasional has reportedly signed electricity supply agreements covering about 7,100 megawatts across 49 data-centre projects in Malaysia, against actual load of roughly 850 megawatts as of October 2025.
That ratio is not evidence of speculation in the Texan sense, because a signed supply agreement is a firmer commitment than a queue position and a facility fills its racks over years rather than at once. The more telling comparison is to the grid itself. Peninsular Malaysia's entire installed capacity is on the order of 27,000 megawatts, making the contracted pipeline alone more than a quarter of it. Texas is auditing a number five times its peak. Malaysia is contracting a number that is a meaningful fraction of everything it has.
We covered the first organised community objection to a Southeast Asian data centre and the health cost of the gas turbines being built to serve these loads. The question the Texas audit raises for the region is simpler than either: nobody publishes an equivalent of the ERCOT queue for ASEAN, so there is no number to discount.
The figure to watch after December
The audit's useful output is a definition, not a smaller headline number. No standard exists for what a utility reports as data-centre demand. Some publish only signed contracts, others publish every inquiry, and the same country's total can differ by a factor of five depending on the compiler.
A published, ownership-verified count of requests with a named end user and committed capital behind them becomes a template other regulators can copy. A smaller unexplained number means the exercise moved the problem rather than solving it.
View author profile → · Editorial policy
