Compared to what?

The question

AI's water and electricity use is real, and it is growing. Most coverage hands you a single number — so many gallons, so many terawatt-hours — with nothing to set it beside. A number with nothing beside it can be made to sound like almost anything.

This atlas puts AI, and the data centers it runs in, next to things people already have a feel for: homes, cars, golf courses, cattle, aviation, Bitcoin, streaming, gaming, cement, steel, and the electric grid itself. The point isn't to argue that AI is large, or that it is small. It is to supply the scale, and let you draw the line.

What you'll find

The homepage is an atlas of comparison plates. Each plate states its figures, draws the comparison as an engraved chart, names the boundary it uses, and carries a confidence note, its sources, and the date it was last checked. Open a plate and the full working is one click away on the methods page, where every figure has a derivation, an assumption list, and a citation.

Why the numbers are ranges

Most figures here are ranges rather than single points. Partly the sources disagree on the point. More often the answer depends on where you draw the boundary. Count only the water that runs through a data center's cooling loop and the figure is one thing; add the water used to generate the electricity that data center runs on and it is an order of magnitude larger. Neither framing is a trick — they are answers to different questions. A range is not a hedge. It is usually the honest shape of the answer.

The editorial standard

Every figure is sourced. Every boundary is stated. Every plate carries the date it was last checked against its sources, and those sources are re-verified on a schedule — when one changes, the figure is reviewed rather than quietly left to drift. Where a number rests on thin data, the plate says so plainly instead of borrowing false precision.

Corrections

If a figure looks wrong, or a source has moved or been updated, the fastest way to fix it is to open an issue on GitHub. The whole site — data, sources, and code — is open. Corrections are welcome, and credited.