you are viewing a single comment's thread
view the rest of the comments
[–] 14 points 1 day ago (3 children)

I know people LOOOOOOVE to use how many bajillion gallons (wtf is a gallon? is that a bucket? learn to use litres) AI data centers use, without looking into how much that is as a percentage of total water use:

Basis: 322 billion US gallons per day, equivalent to 117.53 trillion gallons per year. Percentages are calculated against total US water withdrawals in 2015. (USGS)

Category Annual volume, trillion US gallons Share of total US water withdrawals Source
Total US water withdrawals 117.530 100.000% USGS, 2015 (USGS)
Thermoelectric power 48.545 41.304% USGS: 133 billion gal/day (USGS)
Irrigation 43.070 36.646% USGS: 118 billion gal/day (USGS)
Public supply 14.235 12.112% USGS: 39.0 billion gal/day (USGS)
Industrial, self-supplied 5.402 4.596% USGS: 14.8 billion gal/day (USGS)
Aquaculture 2.756 2.345% USGS: 7.55 billion gal/day (USGS)
Mining 1.460 1.242% USGS: 4.00 billion gal/day (USGS)
Domestic, self-supplied 1.190 1.012% USGS: 3.26 billion gal/day (USGS)
Livestock 0.730 0.621% USGS: 2.00 billion gal/day (USGS)
All data centers: direct water consumption, 2023 0.0174 0.0148% LBNL: 66 billion liters (LBNL)
All data centers: direct plus electricity-related consumption, 2023 0.2288 0.1947% LBNL: 66 billion liters direct plus nearly 800 billion liters indirect (LBNL)
AI servers: annual scenario estimate, 2024–2030 0.193–0.297 0.164–0.253% Nature Sustainability: 731–1,125 million m³ per year (Nature Sustainability)

The first eight categories add up to 99.879%. The small difference from 100% is caused by rounding in the USGS daily figures.

The data-center and AI rows represent water-consumption footprints, while the USGS sector rows represent water withdrawals. They are included for scale comparison and should not be added to the sector totals.

  • source
  • hideshow 6 child comments
  • [–] 5 points 1 day ago

    Thanks for sharing stats. If we could convert 5-10% more of total grid to renewable energy it would more than make up for additional data center water consumption growth nationally.

    We’re obviously in the Fuck AI community but at least there’s one path to limiting the external damages.

  • source
  • parent