On July 14, New York became the first U.S. state to bar construction of new hyperscale data centers — an executive order pausing projects of 50 MW or more for up to a year, citing pressure on utility bills and uncertainty for residents (CNBC, Washington Post, the order itself). It did not arrive alone: Data Center Watch, a research project of 10a Labs, counted at least 75 projects worth roughly $130 billion blocked or delayed by local opposition in the first quarter of 2026 (NBC News), and county-level pauses keep appearing week by week.
We take no position on the ban. We publish an index, not opinions on policy. But the ban rests on a question worth stating plainly, because no public record can currently answer it: of everything queued up to be built, how much is real?
Here is the one part of that answer anyone can verify. As of July 27, the seven U.S. ISO generation interconnection queues — the power plants and storage seeking to connect to serve future load growth, of which AI data centers are one driver — list 946.9 GW of requested capacity, 50 MW and up. That total is not deduplicated: a project can appear more than once while it shops utilities or points of interconnection. Of it, at least 145.9 GW is machine-verifiable as signed — meaning the queue’s own published status field marks an executed interconnection agreement (for example, CAISO’s “Interconnection Agreement Status: Executed”), read by a consistent rule rather than by anyone’s judgment. That is 15.4% of the requested total, and it is a floor, not a rate: the signed capacity we can count comes only from the four queues whose status field our reading can use (CAISO, ISO-NE, SPP and PJM, where the signed share runs from 12% to 81%). In two more, ERCOT and MISO, queue design keeps a meaningful share from being computed at all — in ERCOT’s published data, a signed project moves off the active list, so its active-queue share is near zero by construction, not by weakness — and in NYISO our reading finds no usable status field. Their capacity counts as unverified, which is not evidence it is speculative. The true signed share is at least 15.4% — how much higher, we cannot measure. Every figure links to its ISO source and every step of the derivation is public at /verify.
The demand side is where the question goes unanswered — and Texas shows both the scale of it and the shape of a fix. On June 18 the Public Utility Commission of Texas approved ERCOT’s new batch process for connecting large electricity users. In ERCOT’s own account, large users totaling more than 438,000 MW of proposed demand had entered its connection queue by mid-2026, nearly 90% of those requests from data centers, against an all-time system peak of 85,508 MW set on August 10, 2023 (ERCOT, “ERCOT’s New Batch Connection Process for Large Electricity Users,” Trending Topics, June 18, 2026). Set those two figures side by side and the problem states itself: a queue several times larger than everything Texas has ever drawn at once, and no public way to tell which part of it is firm. The comparison is not a forecast — most of that request volume will never energize, and requests are not commitments; it is the size of the question a regulator now has to answer with the records available.
What ERCOT did next is the part worth noticing. Its “Batch Zero” sorts applicants into named tiers — Base Load, Studied Load, Excluded Load — by how far each has actually come through study. That is a graded status for demand, built by a grid operator that sits outside FERC’s interconnection jurisdiction. Inside that jurisdiction, generation projects carry a graded, public status — application, study, executed agreement, in service. Data-center (large-load) requests carry no standardized public equivalent. In the public record, a headline large-load figure and a firm, deliverable figure are the same number. When the record does not distinguish them, communities reach for the bluntest tool available — and a pause stops the real projects along with the speculative ones. Queue history counsels humility about headline requests in any case: of the generation capacity that entered U.S. queues from 2000 to 2020, 13% had reached commercial operation by the end of 2025 and 75% had been withdrawn — for many reasons, not all of them speculative (LBNL, “Queued Up: 2026 Edition,” p. 55).
What would actually shorten the argument — in Albany, in Johnson County, in every hearing room where this now lands — is not a bigger headline number on either side. It is a public, graded status field for large loads, so that anyone can check which requests are firm without trusting anyone’s summary, ours included. Until that exists, we publish the part that can be checked: the floor, dated, recomputable, with its method in the open.
The reading above is live and changes as queues update; cite it with its date. The current figure and a one-click citation are at /the-number. Don’t trust — verify.