lobsters score: 1 · tags: ai, vibecoding
This is a bit disingenuous.
First, a majority of non-“AI” datacenters don’t use evaporative cooling. This is in part why the numbers in the past are as low as they are.
Second, not only are “AI” datacenters using an order of magnitude or two more electricity than traditional datacenters, the people building them are cheap and choose to use water over having a longer ROI. There are many ways to cool datacenters that don’t require water, just as there are plenty of ways to choose buildouts that require substantially less cooling. These are choices made by people who simply want to maximize profit, regardless of the negative effects on their neighbors.
Third, datacenter water usage is concentrated in the places that have datacenters. If spread out evenly, it doesn’t seem so bad. Put a multi-dozen-megawatt datacenter next to a town of a few thousand people, and you have a wildly different picture than that same datacenter in or next to a city.
The “AI” datacenter build-out really hadn’t gotten very far by 2023, anyhow. Even the title is misleading because it talks about the past up through 2023, not now, when things are vastly different.
Edit: “magnitude or two more electricity”, not “more data”. I mean sure, more data, but I meant power.
score 4 · canonical

