Small Bottle, Big Pipe challenges AI water myths

The headline claim that “AI uses less water than the public thinks” is too cute by half; this paper says the issue is less about splashy per-query numbers and more about concentrated infrastructure demand that local water systems may not be able to absorb.

• –The big takeaway is capacity, not just usage: even if intensity falls, peak withdrawals can still strain municipal systems on the hottest days
• –The burden is geographically uneven, landing hardest on communities that host data centers rather than on AI users in the abstract
• –The paper’s policy angle is practical: report peak water use, coordinate water-power planning, and push “Water Capacity Neutral” commitments
• –It also highlights an operational tradeoff: when water gets scarce, operators may fall back to dry cooling, which can raise power demand and shift the problem elsewhere
• –For AI builders, this is a reminder that infrastructure optics matter; water accounting is becoming part of data-center siting, permitting, and community relations