⚡︎ Independent, reader-supported & ad-free · Follow the megawatts in all 50 states ⚡︎ Support Us
Watch · Our Take

Process Engineer Explains: The Math Behind "Water-Efficient AI Data Centres" Is Laughably Wrong

Mide · 2 months ago

The short version: A process engineer's math shows 'water-efficient' AI data centers still guzzle millions of gallons, and efficiency gains only fuel more buildout via Jevons paradox.

Our Take

Mide, a process engineer, does something most PR decks won't: runs the actual thermodynamics. Heat rejection is heat rejection — you can move it from open-loop cooling towers to liquid cooling, but the physics of dumping server-generated heat into the environment doesn't just vanish because a company slaps 'water-efficient' on a press release. The breakdown of the heat load math is the kind of homework tech companies clearly hope nobody does.

The Jevons paradox point is the one that should sting communities getting pitched these facilities as environmentally responsible. Efficiency gains per server don't shrink the water bill — they justify building more servers, more racks, more campuses. 'Green' becomes a marketing layer over an industry whose total draw keeps climbing. And that draw isn't abstract: it shows up in local water tables, in utility bills, in the fine print of permits county boards approve with little scrutiny.

We say it every time because it's true every time — the tradeoffs are real, the accounting is murky, and the people signing the deals aren't the ones paying the water bill or watching the aquifer drop. Check our facility map to see what's proposed or already running in your area, and if there's a project near you, our take-action page has next steps for making your voice part of the record before the ink dries.

This is GridWatch the USA’s original commentary. The video above is the work of Mide, published on YouTube — full credit to the creator.