Scare Gallons vs Real Gallons
AI Data Centers Won't Drink Towns Dry
In July 2026, protesters wheeled a guillotine into a Salem, Oregon hearing on a proposed data center. Their major argument was that the new data center would consume so much water that it would be a de facto death penalty for residents. Or maybe it was making the point that the developers deserved the guillotine for their crimes? Could have been both.
This made me dig into the facts behind the hysteria. This new report, Scare Gallons vs. Real Gallons, uses data from the USGS and a recent LBNL report to put data center water use on the same chart as the rest of U.S. water use. The result was eye opening: thermoelectric power: about 133 billion gallons a day. Irrigation: 118. Public supply: 39. Direct data-center use, per LBNL: about 0.047. The claim that data centers use “nearly a billion gallons a day” is mostly 'indirect' water used to generate the electricity powering the data center, inflating DC water consumption by 12x. This 'indirect' water usage isn't added to any other category like homes, officees, or factories.
The other problem with the anti-data center case is that their consumption assumptions are wrong. They assume that these new AI data centers will use the old 'air cooling' evaporative model which continuously pulls from outside water to cool the entire data center to the low 60s F. This simply won't work in data centers with rack power of more than about 20kW. New AI facilities cool the chip with DLC and use two closed liquid loops. The roof heat rejection equipment will use dry cooling first, adiabatic only on hot hours, and chillers last.
A town can still reject a campus over power, land, or a local conditions. It should not do it based on exaggerated water usage claims attributed to a cooling system that the building will not use.
In the report, I lay out the story using clear and easy to understand terminology - and pictures too!