About
About Know Your Data Center
A public tracker of the largest AI data centers in the United States: where they are, how big they are, what they cost, and what local governments decided. Built from public data and documents, with sources linked.
Who's behind this
Built and maintained by one person. Not funded by or affiliated with any company, government, political party or campaign.
Sources & licenses
We combine the open datasets below with records we compile from public documents.
Epoch AI, “AI data centers”
epoch.ai/data/ai-data-centers. Satellite-based estimates of power use, compute, chips, capital cost, water and construction timelines for the largest AI sites. Retrieved Nov 15, 2026. License: CC BY 4.0. Modified: US sites only, numbers rounded. Not endorsed by Epoch AI.
PNNL, “IM3 Open Source Data Center Atlas”
Pacific Northwest National Laboratory (U.S. Department of Energy), doi.org/10.57931/2550666, version v2026.11.02. Locations of existing US data centers, derived from OpenStreetMap (© OpenStreetMap contributors). License: Open Database License (ODbL) 1.0. Our reformatted copy is free to download under the same license: existing-data-centers.geojson.
U.S. Energy Information Administration
Electric Power Monthly, table 5.6.A: average retail electricity price by state (August 2026 vs August 2025). U.S. government data, public domain. Retrieved Nov 15, 2026.
GDELT Project
gdeltproject.org. Recent news headlines. We show the headline, outlet and date, linked to the original article.
Deals, public-record entries and local facts are compiled for this site from primary documents: company announcements, SEC filings, utility commission records, county and city records, court filings and local reporting. Every entry links to its source.
Map tiles © OpenStreetMap contributors via OpenFreeMap. ZIP code locations: U.S. Census Bureau Gazetteer files (public domain). Company logos belong to their owners and are shown only to identify the companies.
What the numbers mean
- Power (MW / GW)
- Electricity used by the computers (servers and AI chips). The whole site uses roughly 10–30% more for cooling and other losses.
- Planned
- The power a site reaches once every planned phase is built: a satellite-based estimate, or the developer's stated figure where we have one.
- Built
- The share of planned power that is already running.
- H100-equivalents
- Total computing power expressed as a number of NVIDIA H100 chips, so different chips can be compared.
- Investment
- Estimated capital cost of chips, buildings and power equipment, in 2025 dollars.
- Water (MGD)
- Million gallons of water per day, modeled from cooling equipment rather than metered.
- Status
- Operating: running ("expanding" means more is planned). Under construction: nothing running yet.
Limits
- Most figures are estimates. Epoch AI reports about 80% fall within a factor of 1.4 of the true value.
- Coverage is the largest, frontier-scale AI sites in the United States, not every data center.
- Operators rarely confirm chip counts or water use.
- Deal values marked “Reported” were not confirmed by the companies.
- The public record is a starting set and is not complete.
Neutrality
This site takes no position for or against any company, project or policy. It reports what was announced, filed or decided, and by whom, with links so anyone can check. If the project ever accepts sponsorship, sponsors will be named on this page.
Company logos are trademarks of their owners and are shown only to identify the companies.
Corrections
Spotted an error, or know of a vote, permit or lawsuit that belongs here? Open an issue on the project’s repository with a link to the source.