Press & Citations
Free, citable data — where AI power is real
THE LOADSTAR NUMBER — the generation-queue reality gap · live
~15%
of the queue is machine-verifiable as signed · 145.2 of 958.2 GW · a floor, not a national rate · CAISO 81% · ISONE 68% · SPP 40% · PJM 11%
Verified = an executed interconnection agreement read from each queue’s own published status. In 1 of 7 queues our reading finds no usable machine-readable field — those GW count as unverified, not phantom. Caveats travel with the number. what this means → · verify the derivation · source · method · governance
Numbers you can drop into a filing or a story in five seconds — each linked to a primary source. Our index verdicts are sealed in a dated, externally timestamped ledger; queue statistics are descriptive reads of live official data. Built for regulators and journalists — explicit about the method and its limits.
Updated Aug 29, 2026 · automatic
How to cite us
LOADSTAR, the reality index for AI power (CIIE), accessed Aug 29, 2026. Link to the specific number's primary source and to our open methodology.
Embed the live number on your page:
<iframe src="https://index.loadstar.energy/embed/number" width="340" height="150" frameborder="0"></iframe>Citable numbers (live)
~15%
Where U.S. ISO generation queues publish a machine-readable interconnection-agreement status, the signed share varies widely by queue: CAISO 81% · ISONE 68% · SPP 40% · PJM 11%. In the other 1 queues no usable machine-readable field appears in our reading — those GW count as unverified, not as speculative. Across all 7 queues (50 MW and up): 958.2 GW requested, 145.2 GW machine-verifiable as signed (about 15%) — an inventory floor, not a national rate.
Descriptive, from live official-queue data. This is the generation queue (the supply pipeline for all load growth), not data-center load requests; 'verified' is our reading of each queue's own published status field, not a certified figure. Status taxonomies differ by ISO. An unverified request is not a confirmed phantom.
148.8 GW
PJM's generation interconnection queue (50 MW and up) is 148.8 GW, but only 16.6 GW (~11%) shows a signed agreement or construction status across 651 requests.
PJM is the grid serving Ohio and 12 other states plus DC. Descriptive; the queue is a graded status, not binary. Behind-the-meter load bypasses the queue.
5,290 MW
NYISO codes its large-load interconnection requests by end use, including a dedicated "AI Datacenter" class (DAT-AI). In the July 31, 2026 edition that class carries 5,290 MW across 12 requests — 39% of the 13,568.9 MW NYISO publishes as classified load.
One jurisdiction, the operator's own file, edition-dated. In the file, none of the AI-class requests has reached an executed agreement — expected for a cohort filed mostly in 2025. Class maturity measures queue age, not credibility.
496
LOADSTAR has sealed 496 dated, hash-sealed predictions of which queued projects will withdraw (the speculative dying), from a pool of 976 at-risk requests.
Prospective, dated and hash-chained BEFORE the outcome (not hindsight); the chain root is folded into our externally timestamped anchor. A risk prior, not a validated hit rate — resolves over time (~Dec 2026).
Where AI is forming — top regions (CFI 0–100)
| West Texas / Abilene | 86 |
| Seattle Metro | 85 |
| Phoenix Metro | 80 |
| Northern Virginia | 80 |
| Silicon Valley | 79 |
| Kansas | 77 |
Full ranking and per-region verdicts on the map.
Verify everything
Don't trust — verify. Open methodology · the sealed Journal · externally timestamped root · the Reality Report newsletter. The method changes when the evidence changes.
🌾 The Harvest — our judgment days, pre-announced: December 15, 2026 – February 19, 2028 (first in 107 days) and 2027-05-24+
Nobody in this niche pre-announces their own judgment days. December 15, 2026 – February 19, 2028:
the sealed index predictions mature and their verdicts publish automatically (hit, miss, or unresolved — all of
them — the ledger is append-only, so we can't hide the misses). 2027-05-24: all
496 sealed withdrawal bets reach final accounting — and any predicted project that withdraws
earlier resolves early, week by week, in public. Come back on those dates and judge us — or subscribe
and we'll bring the scoreboard to you.
FAQ — “If no region is phantom, where did the phantom go?” Into the queue.+
Phantom is a project-level phenomenon living inside real regions — regions are
aggregates, and almost no region is 100% hype. The pain sits one level down: where queues publish machine-readable agreement status the signed share varies widely by queue (CAISO 81% · ISONE 68% · SPP 40% · PJM 11%); across all 7 queues, about 15% of requested GW is machine-verifiable as signed — an inventory floor, not a national rate (in 1 of 7 queues our reading finds no usable field). Unverified ≠ phantom; that verification gap is the point.
Northern Virginia is our most-real region and carries gigawatts of speculative queue at the same
time — both things are true. Utilities size substations per project, not per region; regulators approve
tariffs against load forecasts summed project by project. That's why our detection runs per request:
committed-vs-speculative tier per queue entry, duplicate-request flags (the same project shopping several
utilities), per-announcement credibility scores, and 496 sealed per-project withdrawal predictions.
The regional index is the citable headline; the project layer is where the phantom actually lives — and dies.
Media assets
🏭 Watch the index being built — live — 95 autonomous agents ran in the last 72 hours, collecting and sealing. The only index you can watch working.
Citing us? Here is the mark — free to use with attribution. The blue star is the index; the red ghost is what we hunt.
On deadline?
Write to contact@loadstar.energy — answered in writing within 24 hours, usually much sooner. We work in writing only: it keeps every answer verifiable and citable.
Corrections to anything we publish follow the policy at /governance: dated, on the affected page, never a silent edit.
Boilerplate — copy and paste
LOADSTAR is an independent index, published by CIIE, that measures how much of the announced U.S. AI data-center buildout is corroborated by public records — interconnection queues, regulatory filings and permits — with every figure linked to its primary source in a dated, hash-sealed ledger. It is descriptive of today's corroboration, not a forecast; its sealed predictive track record begins resolving in December 2026.
Need an Ohio- or region-specific pull?
We'll pull the numbers for your story or filing, free — you can cite them, and check them.