Asmongold told his audience that AI doesn’t use a lot of water and described the focus on its water footprint as propaganda pushed by people who are scared the technology will take their jobs or careers.
AI itself doesn’t drink anything. The water goes into data centers, which train models, run them for users, and need constant cooling.
Chips turn electricity into heat. Many facilities dump that heat using cooling towers, which evaporate water. Others use closed-loop liquid cooling, direct-to-chip systems, or plain air cooling, which uses less water but more electricity.
There’s also indirect use. Power plants supplying the grid consume water too, so a site with a low on-site number can still carry a footprint elsewhere.
Asmongold’s framing didn’t separate any of that: withdrawal versus consumption, potable versus reclaimed water, training versus everyday use, or national totals versus one town’s supply.
He has a point on one thing. Viral “one prompt equals a bottle of water” posts are usually modeled averages, not measurements. The widely shared 500 ml figure came from academic work covering a conversation of multiple prompts, tied to specific models and specific infrastructure at a specific time. Google, meanwhile, published a 2025 estimate putting a median Gemini text prompt at around 0.26 milliliters. Numbers move fast.
Nationally, data centers are also small compared to agriculture. Lawrence Berkeley National Laboratory estimated US data centers directly consumed roughly 66 billion liters of water in 2023, a rounding error next to farming.
The part that gets messy locally
A gallon in a rainy state isn’t the same as a gallon in a drought-hit basin. Individual campuses can pull hundreds of thousands of gallons on a hot day, often from municipal systems, sometimes with tax breaks and publicly funded infrastructure upgrades attached.
That’s why residents keep showing up at council meetings, and why Microsoft, Google, Meta, and Amazon all now publish water figures and “water positive by 2030” pledges. Those pledges are annual and global, which doesn’t stop a specific town running short in August.
As for the propaganda claim, there’s no evidence in the clip that job fears are driving environmental criticism. Both arguments show up in the AI debate, but they’re separate issues, and neither one proves the other wrong.
