About
What this is
US Data Center Watch is an open-source map of US data centers with modeled community-impact data: annual electricity consumption, water withdrawal, land footprint, capital investment, and where public — public subsidies and jobs. Every number on the site traces to a public source with a date-verified citation.
Why it exists
Hyperscale operators publish global aggregates (Meta's fleet used 14,975 GWh in 2024; Google used 30.8 million MWh) but rarely break those numbers out per site. Communities hosting a campus see the substations and water lines but not the operator's own annual disclosures, the subsidies that got the deal done, or how their site compares to the next one over. We connect those public threads to a map so the local picture is visible alongside the national one.
We do not run any servers ourselves, run no ads, and accept no operator funding. The site is a static export from a YAML-source-of-truth — anyone can audit the pipeline, the model, or open a correction against a single facility.
What's in the data right now
Coverage
| Facilities | 338 |
| Operators | 173 |
| States | 41 |
Status mix
| operational | 331 |
| under construction | 6 |
| announced | 1 |
| decommissioned | 0 |
Sourcing confidence
| high | 8 |
| medium | 9 |
| low | 321 |
high = primary operator disclosure or
government filing;
medium = derived from disclosed annual MWh
or backed by trade press;
low = OSM-imported stubs awaiting
human curation. See the
methodology page for the modeling pipeline.
How we source data
The source-of-truth is hand-written YAML in
data/facilities/{state}/{slug}.yaml in our public repo. Every
field traces to a URL in a sources[] array whose
supports[] list names which fields each URL backs. The full
sourcing hierarchy and the omitted-field discipline ("don't fabricate; explain
why we don't have it") are documented in the
curation guide.
We default to strict primary sourcing: operator-published material first,
local government / EDC records second, named trade journalism third, aggregators
last. Where an operator publishes annual MWh but not nameplate capacity (Meta,
Apple do this), we back-derive IT load using the modeling pipeline's
published assumptions and flag the field as
confidence: medium.
How to contribute
Two paths, both via GitHub:
- Spot something wrong, missing, or stale? Use the facility correction issue template — pre-filled with the fields a curator needs. On mobile, every facility tooltip has a "Report data issue ↗" link that pre-fills the slug for you.
- Want to curate a facility? Read the curation guide and open a PR. Hand-curated facilities flow to the live map on the next deploy.
Tech credits
- Astro for the static-site framework
- MapLibre GL JS + deck.gl for the map renderer
- Carto Dark Matter basemap (© OpenStreetMap, © Carto)
- OpenStreetMap for the secondary-tier facility import (low-confidence stubs)
- Cloudflare Pages hosting + R2 for the static dataset
Open source, MIT-licensed. Source: github.com/zzulanas/usdatacenterwatch.