Hacker News
'The Water Footprint of AI' Lands in Water Research as Per-Query Estimates Still Span Three Orders of Magnitude
A new paper in Elsevier's Water Research (doi 10.1016/j.watres.[redacted]) hit HN at 41 points and 50 comments within five hours. The surrounding literature it enters remains wildly unsettled: UC Riverside's figure of ~519 ml for a 100-word response sits against Google's ~0.26 ml per median Gemini text query and OpenAI's ~0.3 ml, with a UN University report putting 2025 datacenter consumption at 1.8 million Olympic pools and a high-AI scenario reaching 9.3 trillion liters by 2030. The spread is a scope argument, not a measurement dispute — Scope 1 on-site cooling versus Scope 2 electricity-generation water is what separates the numbers.
↳ Follow the thread