How Much of a Problem Is AI's Water Use, Really?

"Protect our water." "Water for people not AI." "Don't mess with our water." The signs have appeared from Texas to New Mexico to Arizona as residents organise against the data centres tech companies are building to expand AI capacity.

The anger is about water, energy and pollution together. The water question is the one that turns out to be hardest to answer honestly.

What the water is for

Many data centres use water to absorb heat from air or surfaces, then evaporate it to cool — the same mechanism as sweating. The processors inside can reach internal temperatures of around 176°F, for the reason an overtaxed laptop gets hot.

Why the numbers don't settle it

Energy consumption is not seriously disputed. Water is messier. Claims circulating online range from articles asserting that a chatbot consumes 500ml per query to commentators declaring the issue nonexistent, and the disclosures from large technology companies are often incomplete and inconsistent.

Experts stress that what matters is not the national total but the local one. A facility drawing from an aquifer under stress in Arizona is a different proposition from one on abundant supply, even at identical consumption — and it is the local figure that residents at those protests are actually contesting.

The reporting gap

That is where the argument stalls. Sound answers require site-level withdrawal and consumption data set against local availability, and that combination is rarely published.

The result is a debate in which both the most alarming and the most dismissive claims survive contact with the evidence, because there is not enough evidence to kill either.