A headline you’ve probably seen
“ChatGPT drinks a 500 ml bottle of water for every 20–50 questions.”
Euronews, 2023 · from a UC Riverside study (Ren et al.)
That’s real. But is AI the water problem?
Where it really goes, the AI myth, and who’s fixing the real problem
UN SDG 6 · Clean Water & Sanitation / Anders Imbrie / [Course] · [Date]
01 · The basics
Water taken from a river, lake or aquifer. Much of it flows back.
Water that never comes back: it evaporates or goes into a product.
Demand close to a region’s renewable supply. Always local.
All the water behind a product, across its whole supply chain.
All US data centers, running for a whole year. US farms use up that much water in about…
Tap = +1 vote · Shift-tap = −1
01 · The basics
118B
GALLONS / DAY62% used up by crops and evaporation
133B
GALLONS / DAYOnly ~3% used up. The rest goes back, warmer
17B
GALLONS / YEAR70–80% evaporates in cooling towers, from one town’s supply
It’s not how much. It’s where and when.
02 · The AI myth
A full year of US data-center water ≈ 5.6 hours of US farm irrigation.
17B gal/yr direct (LBNL, 2023) ÷ 73B gal/day consumed by irrigation (USGS, 2015)
Add the power plants that run them
≈ 3 days. Still small, nationally.
03 · The real problem
71%
of the world’s freshwater withdrawals go to farming
FAO AQUASTAT, 2022 data
126B m³
of treated water lost to leaks every year. Since this slide opened:
0 L
Liemberger & Wyatt, 2019 · ≈ 4M liters per second, ~$39B a year
2.1B
people still lack safely managed drinking water: 1 in 4
WHO/UNICEF JMP, 2025
By 2030, demand is projected to outstrip supply by 40%.
Global Commission on the
Economics of Water, 2023
04 · Who’s fixing it
125M Lsaved / site / year
Closed-loop chip cooling in every new design since 2024
2.5B galfirst disclosure, 2025
Says about two-thirds goes back to local communities
−57%water, IRRI rice trial
Gravity-powered drip irrigation built for small farms
−96%finishing water
Water<Less process; 3B+ liters saved so far
Tech has the worst headlines but moves fastest. Farms need the most help.
05 · What to do
01
Build where water is plentiful, and invest in the local systems you draw from.
02
Price water risk in. Droughts shut plants, and leaks waste $39B a year.
03
Go after the 71%: fund efficient irrigation with farm suppliers, and reuse water.
Water is a local problem with global supply chains.
Would you build your next data center in Arizona? Why or why not?
Process
Claude (Anthropic), Claude desktop app
Anders Imbrie
Lesson: AI turned up conflicting numbers for the same fact, so every figure here was checked against its source.
Sources
01Anthropic. (2026). Claude [Large language model]. claude.ai
02Data Center Knowledge. (2024). Microsoft unveils zero-water data centers.
03Euronews. (2023, Apr. 20). ChatGPT ‘drinks’ a bottle of fresh water for every 20 to 50 questions.
04Latitude Media. (2026, June 11). Amazon finally reports its annual water use at data centers.
05Lawrence Berkeley National Laboratory. (2024). 2024 US Data Center Energy Usage Report.
06Levi Strauss & Co. (n.d.). Water stewardship.
07Liemberger, R., & Wyatt, A. (2019). Quantifying the global non-revenue water problem. Water Supply.
08Rice Today (IRRI). (n.d.). N-Drip gravity drip irrigation trial in rice.
09Smart Water Magazine. (2025). What AQUASTAT 2025 reveals about agriculture.
10The Conversation, via Civil Beat. (2025, Aug.). Data centers consume massive amounts of water.
11The Guardian. (2023, Mar. 17). Global fresh water demand will outstrip supply by 40% by 2030.
12The Register. (2025, Jan. 4). How datacenters use water.
13U.S. Geological Survey. (2018). Estimated use of water in the United States in 2015 (Circ. 1441).
14WHO & UNICEF. (2025). Progress on household drinking water, sanitation and hygiene 2000–2024.
← → move · N notes · P presenter view · F full screen
0:00
Target 7:00