06/03/2026
"AI is draining our rivers" is one of the stickiest tech narratives of the last two years. It's also a misallocation of public concern at a scale I've rarely seen in 25 years of building infrastructure.
The facts:
■ All US data centers consumed ~17.4 billion gallons of water in 2023 (about 0.3% of the public water supply).
■ US irrigation pours ~43 trillion gallons a year onto cropland. That's a factor of more than 2,000 to 1.
■ In the Colorado River basin, alfalfa alone — cattle feed, much of it exported to dairies in China and Saudi Arabia — consumes more water than every city and industry in the basin combined.
But here's where I argue against my own side: the critics are wrong about scale, and right about shape. Data center water isn't a volume problem. It's a concentration problem — too much demand landing on too few pipes, often in the driest places. And a concentration problem has a name for its solution: distribution. The people sounding the alarm have, without realizing it, written the specification for decentralized, behind-the-meter, closed-loop infrastructure — the exact opposite of the drop-a-gigawatt-anywhere model we're racing to build.
My new essay, "The Almond and the Algorithm," is about learning to worry about water accurately — and why that discernment is the skill the age of AI will demand most.
Link:
The internet says AI data centers are draining our rivers. The numbers say we're worried about the wrong water.
