Datacenter buildout tracker: capacity by operator, from SEC filings

The AI buildout is discussed in press releases and measured in nobody's spreadsheet. Gambit builds it from the one source that carries legal consequences for being wrong: the operators' own SEC filings. Each 10-K, 10-Q and 8-K is parsed for site-level disclosures (location, stage, megawatts, land), and every extracted row keeps a link back to the filing. 3,364 candidate disclosures were evaluated to produce the 869 site-filing rows behind this page.

This is the free summary layer. The live version (the map, the per-site rows, the filing links, the capacity-versus-committed history) is on the DATA desk in the terminal.

A site moves announced → land secured → power secured → under construction → operational, and it is disclosed again at each step. The megawatt column is what has been disclosed at that stage; the row count is filings, not sites.

All 22 operators in the dataset, ranked by total disclosed megawatts. % operational is the share of that operator's disclosed capacity currently at the operational stage. It is the difference between a company that owns power and one that owns a press release.

Every row in the tracker carries the document it came from. Here are the twelve largest single-site capacity disclosures, each linked to the filing on SEC EDGAR, so the chain from this table to the original sentence is one click long.

Disclosed capacity is an intention; energized capacity is a fact. The gap between the two is a grid connection. 4,029 projects representing 1,068,260 MW are waiting in interconnection queues, roughly 29 times the entire disclosed datacenter pipeline in this tracker, with MISO the largest single utility. Against that, 14,577 MW of the 36,472 MW disclosed is live today, and the latest announced-to-operational lag is 8.7 months across 86 matched pairs.

Filings only, per site and per stage. The source is SEC 10-K, 10-Q and 8-K documents: 853 of 869 site-filing rows link to a document on EDGAR, the other 16 to the operator's own primary disclosure where the filing points there. Each row is a named place, a stage, a megawatt figure, a date and a filing URL. Nothing comes from an analyst estimate or a trade-press megawatt guess.

Deduplication. Operator totals use the product's deduplicated basis. Summing filing rows per operator double-counts multi-filing sites and overstates capacity, an easy and common error to make with this kind of data.

What it is not. It is not a census of world datacenter capacity. Private operators and the hyperscalers' unfiled capacity are outside it by construction: if it was never disclosed in a filing, it is not here.

This page publishes twelve of the largest disclosures and the operator ranking. The desk holds every site the tracker knows about, placed on a map, each one carrying the filing that disclosed it and the stage it has reached.

Is usable compute scarce or abundant?