Coverage & gaps.

Corporate ownership

No U.S. state requires a data center to disclose the ultimate corporate parent behind the LLC that builds it. Piercing that shell is the point. We resolve operators to their real parent through CC0/public-domain sources — GLEIF's who-owns-whom registry, SEC Exhibit-21 subsidiary lists, and Wikidata — and we name the operators that still withhold theirs.

The frontier is the bottom of the tree: single-purpose LLCs that buy the land and appear in no 10-K. We've started cracking them automatically — 147 land-buying shells unmasked from open .gov business records by matching each LLC's formation address to a corporate HQ (e.g. Google's Questa/Design/Arti, Meta's Siculus). Expanding state by state.

Hearings & campaigns

We turn the dataset into action: automatically tracking the local hearings where data centers are approved or stopped, and organizing supporters around them. Hearings are polled from public meeting agendas (Legistar & Granicus, including staff-report PDFs), AI-extracted from news for jurisdictions on no platform, and cross-linked to authoritative county application records. Every hearing is tagged with its source and confidence.

localview

182

Legistar API

26

Granicus + staff-report PDF

14

official site

1

escribe

1

Browse them on the campaigns directory or learn how to act in the organizing toolkit.

Pipeline freshness

The last run of every ingest source, straight from the ingest_runs log.

Global coverage

DataCentersExposed is expanding from the United States to the 32 countries driving the AI data-center buildout. We add a country only when we can source it openly — empty country pages stay out of the search index until they carry real, sourced data. Here is where each country stands.

31 of 32 countries are live. Browse them on the countries directory. International boundaries come from geoBoundaries (CC BY 4.0).

Known gaps

• Realized subsidy dollars are tracked at the state-program level — what each state's own tax-expenditure / auditor report says its data-center tax break costs. GASB 77 makes naming the recipient voluntary, so most jurisdictions disclose only program totals; per-facility dollars are reachable only where a county ACFR itemizes a named company, or via a Good Jobs First data-sharing agreement (they are an ally, not a source we scrape).
• Texas and Wisconsin publish which data centers get the sales-tax break but withhold the dollar amounts by state law — we show the program, flagged as withheld, not an invented figure.
• Virginia DEQ's data-center air-permit list sits behind Akamai bot-protection; per-facility generator nameplate MW is pending a browser/FOIA path.
• Most facilities are located from OpenStreetMap (a single source). They are labeled "mapped", not "operating", until a second source corroborates them.
• MW is disclosed for a minority of sites; the rest await ISO interconnection-queue matching (gridstatus) and air-permit nameplate data.
• Water figures cover self-supplied withdrawals reported to state agencies. Most data centers buy municipal water, which no state publishes — those gallons are reachable only by public-records requests, facility by facility. We publish records, never estimates.
• Automated hearing tracking covers jurisdictions on Legistar and Granicus (whose agenda items, including code-based ones, we read down to the staff-report PDF), plus AI-extracted news for everywhere else. Boards on PrimeGov, CivicClerk, and the newer Granicus engagement portal hide per-item agenda text behind JavaScript or PDFs, so those are currently covered by the news layer only — not item-level polling.
• Structured land-use application records are wired for Loudoun County (the LOLA registry — 15 active data-center projects with parcels and case numbers). Other Data Center Alley counties publish similar ArcGIS application layers that aren't ingested yet; until then their pipeline shows only what reaches an agenda or the news.
• Automated shell-unmasking (resolving the LLCs that buy data-center land to their real parent by matching formation address to a corporate HQ) currently covers Iowa, whose Secretary-of-State registry is free, geocoded, and openly licensed. Texas (which also exposes officers, enabling individual-officer clustering) and Ohio are next; the county-deed buyer cross-reference is not yet wired.

See the methodology for how each source is verified, and the corrections log for changes after publication.