Is there any reason the water can’t be safely consumed later? It’s not toxic or nuclear is it? The cooling water didn’t just up and disappear did it?
Edit: Links provided in the comments…
- https://arxiv.org/pdf/2304.03271
- https://www.youtube.com/watch?v=H_c6MWk7PQc
- https://www.youtube.com/watch?v=H_c6MWk7PQc&t=1264
- https://pivot-to-ai.com/2026/03/06/how-much-water-do-the-data-centres-use-its-a-secret/
- https://ui.adsabs.harvard.edu/abs/2025EcInd.17012986J/abstract
- https://en.wikipedia.org/wiki/List_of_tallest_cooling_towers
- https://en.wikipedia.org/wiki/Niederaussem_Power_Station
Notable comments:
- https://lemmy.world/comment/23672269
- https://sh.itjust.works/comment/25288634
- https://lemmy.cafe/comment/16350045
- https://sh.itjust.works/comment/25294655
Edit addendum: I’d like to thank everyone that’s participated in this question thread, sorry if I missed any good relevant links in the comments.
To be clear, I still loathe the whole AI datacenter era, it really is heavily wasteful of resources, notably energy, but I wanted to better understand the water usage situation.
And all that is to… Produce steamy furry porn
If we’re lucky it’s furry porn. I’m more worried about the non-consensual porn of real people, including simulated CSAM.
The most funny thing is, that once trained, the model can run and make furry porn on your local machine. So they don’t even make any money on this.
I haven’t tried furry porn models in particular, but all local image generation I’ve tried locally was really bad. It was with a 3070, so nothing really meant for this.
Mainline stable diffusion is pretty horrible. Try using SDXL community fine tunes (if they fit into your VRAM). They’re worse than cutting edge cloud models, but they punch way above their league