THE LOADSTAR NUMBER

The ratio is the number; the floor is the honesty. Where our reading finds no machine-readable status field (NYISO, 24.9 GW, about 3% of the queue), every gigawatt counts as unverified — unread, never speculative.

Where U.S. ISO generation interconnection queues publish a machine-readable interconnection-agreement status, the signed share our reading finds varies widely by queue — from 11% to 81% across the 4 queues whose status field our reading can use (CAISO, ISONE, SPP and PJM); the figure for each queue, with the rule we read it by, is at /verify. Those shares are not a maturity ranking and should not be read against each other: each one reflects its own queue’s disclosure practice and market design. In 2 further queues (ERCOT and MISO) our reading cannot produce a meaningful signed share at all, for reasons specific to each and unrelated to one another. In one, signing moves a project off the active list altogether, so signed projects are not in the population we read. In the other, the status field is filled in on only a sliver of rows, so the rest of that queue carries no readable status — silence in the field, which is not evidence that an agreement is missing. We report no share for either and draw no conclusion from either; they are not comparable with the shares above. In the other 1 queue (NYISO), no usable machine-readable status field appears in our reading — and we have not resolved whether the queue publishes no such field or our reading fails to find it. Their 24.9 GW of requested capacity therefore count as unverified, with no claim about how much of it is signed. Across all 7 queues (50 MW and up), 958.2 GW is requested — a total that is not deduplicated, since one project can appear more than once while it shops utilities or points of interconnection — and at least 145.2 GW of it is machine-verifiable as signed. That is 15.1% of the requested total, but read it as a floor and not as a rate: the signed capacity we can count comes only from the queues with a readable status field, while the total covers all 7. We do not estimate how far above that floor the real share sits. A queue position is an early-stage signal in any case: of the 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that entered U.S. queues from 2000 to 2020, 13% had reached commercial operation by the end of 2025 and 75% had been withdrawn (LBNL, ‘Queued Up: 2026 Edition’, p.55).

This is the GENERATION interconnection queue — power plants and storage seeking to connect to serve future load growth, of which AI data centers are one driver — not data-center (large-load) requests, which most ISOs do not publish. ‘Verified’ means OUR consistent reading of each queue’s own published status field (e.g. CAISO’s ‘Interconnection Agreement Status: Executed’, MISO’s post-GIA stage), including projects already under construction — which require an executed agreement; it is not a certified or audited figure. Status taxonomies differ by ISO, so per-queue shares are not directly comparable, and a near-zero share can be queue design rather than weakness: ERCOT runs a connect-and-manage regime outside FERC’s interconnection jurisdiction, and in its published data a project whose agreement is signed moves out of the active queue (it is listed as Completed) — so ERCOT’s active-queue signed share is near zero by construction, a queue-lifecycle artifact, not a read on project maturity. MISO’s near-zero share has a different cause, not a shared one: its post-generator-interconnection-agreement stage field is populated on only a small slice of its rows, so most of its queue carries no readable agreement status at all and cannot count as signed under our rule — an absence of data, not evidence of speculation. Where our reading finds no usable machine-readable field (currently NYISO), that says nothing about those ISOs’ disclosure — it may be our parsing limit — and those GW are counted as unverified, never as phantom. An unverified request is NOT a phantom: queues are unsigned by nature and status is graded (FERC Order No. 2023 replaced serial first-come with first-ready cluster studies). The raw queue is not deduplicated; the same project can appear more than once (part of what FERC’s RM26-4 show-cause targets). Historically most queued generation never reaches operation: of the capacity that requested interconnection from 2000 to 2020, 13% reached commercial operation by the end of 2025, 75% was withdrawn and 10% was still active (LBNL, ‘Queued Up: 2026 Edition’, p.55; by project count rather than capacity the completion figure is 19%, p.4). That cohort ran under the serial regime Order 2023 has since replaced, so it benchmarks history rather than this queue — which is why we report what carries a signed agreement today instead of forecasting what will be built. Descriptive of today, not a forecast.