The Water Footprint of AI

MASNeo 45 points 70 comments August 11, 2026
doi.org · View on Hacker News

Discussion Highlights (9 comments)

throwfaraway135

I agree that water is an important resource but the wording of "consumption" always rubbed me the wrong way. Is the water contaminated with chemicals? Maybe heat? Can people or wildlife consume it? Can you use rainwater?

pcl

”AI's global water footprint could reach 4.2–6.6 billion cubic meters by 2027” This is about as much as the annual consumption of the California almond farms. (An acre-foot is approximately 1200 cubic meters.) https://www.c-win.org/cwin-water-blog/2022/7/11/california-a...

exizt88

Water consumption of AI is a non-issue. Why even spend research and policymaking resources on this?

shaky-carrousel

The issue with AI datacenters was never about water. Water is an easily dismissable strawman that sloppers love to bring on. The real issue with AI datacenters is the huge amounts of wasted energy, after decades of energy efficiency efforts.

cannonpr

Did we somehow mass build pure lossy evaporatively cooled data-centres in the AI era while I was asleep ? Most of them im familiar with lose at best 3-4% of their water per day due to recovery systems even when they are evaporative or open loop to a lake. Yet nearly everyone I hear talking about this seems to treat it as if they lose 100% of their water loop contents per day as if it’s a matter-antimatter destruction system…

nmeofthestate

0.16% (high estimate) of global water usage by the way.

TSiege

Whether or not you think water use is bad for AI, another aspect hard to ignore with it is that the closed loop systems are mixing PFAs into the water for closed loop systems https://www.theguardian.com/environment/2025/oct/04/pfas-pol...

felooboolooomba

Can someone explain why the data centers are using this insane amount of water. What's the ratio of closed system, evaporation?, and straight-down-the-drain water?

sceptic123

Who could have guessed this would end up with many polarised "massive problem" / "not a problem" comments

Semantic search powered by Rivestack pgvector
4,128 stories · 37,281 chunks indexed