A retail data center uses 6 million gallons of water a year, about the same as a golf course. A hyperscale data center, the type that runs AWS or GCP uses 200+ million gallons a year. An “AI” data center uses an order of magnitude more. Those numbers are all on the low range. The higher estimates are 1.8 BILLION gallons annually.

Those are just direct water usage. Water is also used to cool the power plants supplying the power to these data centers. Even if they use new forms of cooling, like closed loop systems, those systems trade the water usage on site for energy usage. That increased energy usage uses more water at the power plants.

This water isn’t reused, like on golf courses, it’s not collected and recycled. It’s pulled from aquifers and rivers and evaporated straight into the sky. It’s not only destroying the downstream environments, but it’s destroying underground aquifers that humans survive on.

There is literally no way to look at AI data centers and claim they are good for the environment. And if you come in and claim that agriculture uses more, you’re just an idiotic troll. Food keeps us alive, LLMs do not.

According to scientists at the University of California, Riverside, each 100-word AI prompt is estimated to use roughly one bottle of water (or 519 milliliters).

https://www.eesi.org/articles/view/data-centers-and-water-consumption

Pssh. You bought your home server instead of just assembling it from a box of scraps you had lying around?

Every damn time.

I think rsync is lowercase, and maybe zip. The rest I think are uppercase. Complete guess though.

God fucking dammit.

I’m in a location I’ve never been before so I truly believed it.

That sounds like a major concussion. Wishing you the best

It’s not a typo, just a meme. https://knowyourmeme.com/memes/commit-sudoku

I commented on that video the day it came out and I’m still having arguments with people on it.

There seems to be two common threads:

• claiming that data centers use as much water as a golf course and no more energy than any other big building “because they use new more efficient designs”
• comparing it to things that humans use to survive, like agriculture, refrigeration, heating and cooling

Both of these seem to be some right wing influencer or news casters talking points and they’re both insane.

The actual facts:

1. Retail data centers use about 6 million gallons of water a year. Yes that is almost the exact same as a golf course.
2. Hyperscale data centers (the data centers that AWS, GCP, Azure, etc run on — the ones built, owned and operated by Amazon, Google, and Microsoft — use 200+ million gallons a year. This is a fucking order of magnitude larger than a retail data center and IT’S STILL NOT HOW MUCH AI DATA CENTERS USE.

We don’t have numbers for AI data centers because they’re hiding them, but just from energy usage alone the water usage will be more, because that’s what the water is used for.

3. Yes, new designs exist…but they don’t solve anything. Those new designs trade off water for electricity. Either they use more electricity or more water. Unless the design is completely renewable energy they are using just as much water (guess what power plants use to cool their systems?!?!?)
4. The majority of new AI data centers are not using these new systems, so it’s literally just a red herring to stop your argument about water usage.
5. Agriculture is used to keep people alive. If you deleted LLMs from the planet right now, nothing would happen. People would go back to their daily lives. You can’t delete food from the planet.
6. I had a sixth but I can’t remember it. If I remember it I’ll put it here.

If y’all want my sources I can post them, just wanted to jot all this down quickly between meetings.

Outer wilds. I’m literally

Huh? Google doesn’t own ChatGPT. And no, your personal commute to work isn’t a millionth of a fraction of what is used to train these models.

AI datacenters account for less than 10% global co2 emissions. Of course ten percent more global warming is bad, but i personally have several things that contribute more than using AI ten hours a week.

No, no you don’t.

Then they still used just as many resources? All these local models are distillations of the big models.

Bad flags:

• Baltimore
• California
• Mississippi
• Los Angeles

Mediocre flags:

• St. Louis
• Kansas City
• Kingman
• Lincoln

Good:

• Colorado
• Denver
• Frederick County
• Madison
• New Mexico
• Portland
• Tulsa

Huh? External sound does not mean the object produces sound.