Indoor fixtures show an efficient → wasteful band with the U.S. average (◆); food shows the water behind a single serving; an AI prompt is so small it's a hairline here.
Indoor: measured per-use averages from the Water Research Foundation's Residential End Uses of Water (2016), with efficient/standard rates from EPA WaterSense & ENERGY STAR (REU2016 measured indoor use at 52.1 gal/person/day). Food: total green + blue + grey “virtual water” per typical serving, from the Water Footprint Network and Mekonnen & Hoekstra (2011, 2012) — global averages that vary widely by region and method. AI: water evaporated to cool data centers plus the water used to generate their electricity, per query. Operators (OpenAI, Google) disclose on-site cooling only (~0.3 mL); independent estimates add power-plant water (~3–5 mL), and “thinking” models run far higher — from OpenAI (2025), Google (2025), Li et al. 2023 and Jegham et al. 2025. Figures swing widely with each data center's location and grid.