About

Summary

Individual AI data centers are growing in scale, consuming power that rivals that of states. For example, the campus proposed for Pike County, Ohio comes with a dedicated 9.2 gigawatt natural gas power plant — described by regulators as capable of “powering more than half of Ohio’s typical electrical load.”

Numbers like “9.2 GW” are hard to hold in your head. This site tries to make them tangible by putting them next to something everyone already has an intuition for: states.

How it works

  1. Pick one or more data centers from a curated list of large, publicly-reported U.S. projects.
  2. The tool sums their announced power capacity.
  3. It then highlights the U.S. states whose combined average electricity draw adds up to roughly the same amount — using one of two matching strategies you can toggle between:
    • Fewest whole states (default) — accumulate states from smallest to largest average power draw until the target is met. This answers “how many entire states does this equal?”
    • Closest numeric match — a subset-sum search for the combination of states whose total lands closest to the target. Numerically precise, but it typically returns only one or two arbitrary states.
    Alongside the matched set, the tool reports where the selection would rank if it were a state and its share of the U.S. total. A treemap below the map sizes every state by its average draw, with the selected data centers dropped in as siblings for direct comparison.

Why “average power,” not peak demand?

A data center’s announced capacity (in MW or GW) is a continuous load — it runs close to that draw around the clock. To compare fairly, each state’s annual electricity consumption is converted into an equivalent average power draw: annual GWh ÷ 8,760 hours in a year. This is the constant power a state would need if its yearly electricity use were spread evenly across every hour — directly comparable to a data center’s steady-state GW figure.

As a sanity check: Ohio’s 2024 electricity sales imply an average draw of about 17.6 GW — consistent with public reporting that the proposed 9.2 GW Pike County plant would represent roughly half of the state’s typical load.

Data sources

Every dataset used here is documented in full, with retrieval dates and direct source links, in data/README.md in the project repository. In short:

  • State electricity consumption: U.S. Energy Information Administration (EIA), Form EIA-861 annual retail sales by state (public domain).
  • State boundaries: U.S. Census Bureau TIGER/Line shapefiles via the R tigris package, repositioned so Alaska and Hawaii appear as insets.
  • Data centers: hand-curated from public reporting, company announcements, and utility/regulatory filings — every project links to its source. No free, comprehensive, machine-readable dataset of U.S. data center power capacity currently exists; the most complete one we found (Cleanview’s behind-the-meter data center report) is paywalled, so this project builds its own small, fully-sourced list instead.

Caveats

  • Announced capacity isn’t built capacity. Figures reflect the best public estimates at time of writing and will change as projects are built out, delayed, or revised.
  • This is a curated sample, not a census. The data center list covers ~10-15 marquee projects chosen for how well-documented and how large they are — it does not attempt to catalog every U.S. data center.
  • Average power, not instantaneous demand. Real electricity use fluctuates hour to hour and season to season for both data centers and states; averaging smooths that out in exchange for a fair, simple comparison.

Code

All data processing and cartography run in R; the interactive map is built with Observable JS and D3. Source: github.com/jhelvy/datacenter-states.