The problem, in the open
Everyone downstream pays for a queue nobody verifies.
Announced gigawatts get treated as real: planned around, priced in, built for. Then a bill, a moratorium, or a write-down proves otherwise. No grid operator publishes which requests intend to build. The most important number in the AI buildout has no referee.
—
requested vs. signed capacity in live U.S. generation queues (live figure loads from the index)
~13%
of 2000–2020 queued capacity had reached operation by end-2025 (LBNL, Queued Up 2026)
nobody
owns the layer that verifies which requests are real
For whom
One wound, felt six different ways.
A regulator calls it a filing. A utility calls it capex at risk. An investor calls it diligence. A journalist calls it a citation. Underneath, everyone is asking the same thing: is this real, or is it phantom? LOADSTAR answers that question for each of them — same evidence, same methodology, read through the angle that decides their next move.
You need a defensible cut for a filing.
Every score in LOADSTAR links to its sources — official filings and press, labeled which — is sealed by hash, and can be recomputed by anyone — including the parties in your docket. That's the standard a citation in an order needs to survive scrutiny. In June 2026, FERC ordered six grid operators to filter speculative large-load requests out of their interconnection queues — the first time a federal regulator formally named the problem this index measures. We didn't cause that order; we started measuring the same problem before it had a name.
You need to know which queue requests are real before you spend capex.
Interconnection queues run inflated by duplicate filings and speculative requests — in ERCOT alone, queued generation and storage requests (50 MW and up) run roughly 300 GW against a system peak of around 85 GW. LOADSTAR cross-references independent signals — permits, energy contracts, hiring, federal registration — to flag which requests have real corroboration and which look like duplicates of the same project under different filings, before you commit capital to infrastructure a project may never need.
You need to know if the demand behind an asset is real before allocating capital.
Historically, most capacity that enters U.S. interconnection queues is never built (LBNL's Queued Up series tracks this), so announced demand and built demand are rarely the same number. The regional Compute Formation Index reads that gap today — how much of what's been announced in a market has independent corroboration versus how much is still just a filing — as a descriptive measure of present conditions. Sealed calls begin resolving around December 2026, and we publish the outcome either way. What you get before committing capital: a corroboration signal on the demand behind a specific market, sourced and dated.
You need to know which regions have real buildout underway.
A site-selection commitment puts real capital behind a bet on a specific market — usually based on the same public announcements anyone else can read. The regional score cross-references that announced pipeline against independent signals — permits, energy contracts, hiring — to separate corroborated construction from filings that are still speculative; it's descriptive of today's market. Sealed calls begin resolving around December 2026, published either way. What you get before you commit: evidence, sourced and dated, to check against your own thesis.
You need to see where competitors are really building — and where they're only announcing.
Buildout planning means reading what competitors are actually building, not just what they've announced. For scale, one historical benchmark. Of the capacity that requested interconnection in the U.S. between 2000 and 2020, 13% of that capacity had reached commercial operation by the end of 2025, and 75% of that capacity had been withdrawn (LBNL, Queued Up 2026). That cohort predates the AI buildout and is mostly wind, solar and gas, so it describes how queues have behaved over time — not a forecast for today's queue, and not a claim about any particular company. LOADSTAR reads the queue at a granular level and cross-references it against independent signals — permits, energy contracts, federal registration — to separate corroborated buildout from filings that never move; it's a signal, not a verdict on any one competitor's plans. What you get: granular corroboration data, by region and by queue entry, sourced and dated, as input to your own buildout planning.
You need a sourced, dated figure you can cite.
The public reality index and the methodology behind it are free to cite: every score links to its sources — official filings and press, labeled which — is sealed by hash and dated, and can be recomputed by anyone, including you, before you publish it. We're not asking you to buy anything. We're giving you a number you don't have to take our word for.