AI's Water Usage Is Growing - Measuring it is top priority
New reporting confirms AI data centers' water use is growing, but so are the tools to measure and cut it. Here's why standardized water metrics matter.
Fresh reporting this week put a number on a question that keeps coming up: how thirsty is AI, really? Data center cooling drew tens of billions of liters of water in 2023 — still under 1% of total U.S. water consumption — but researchers warn that figure could roughly triple by 2030 as AI buildout accelerates, approaching the annual drinking-water supply of a city the size of New York. At the same time, Google says efficiency gains have cut its own per-query water cost to a handful of drops for a typical chatbot exchange.
That gap — "basically nothing" versus "a major city's water supply" — isn't really a contradiction. It's a measurement problem. Per-query estimates, facility-level liters, and fleet-wide projections are all answering different questions, and right now companies mostly get to pick which one they publish.
That's exactly the gap AI for Ecology exists to close. We push for open specifications and independently run benchmarks so operators, regulators, and the communities living next to these facilities can compare water and energy performance on equal terms — not take a company's self-reported headline number on faith. A credible certification path for data center water and energy performance would let buyers, investors, and neighbors see, concretely, which facilities are actually improving and which are just better at picking favorable metrics.
Source: Genetic Literacy Project
