UN: AI's Water Use Will Match the Needs of 1.3 Billion People by 2030
AI Is Threatening Natural Resources for Billions
A new UN University report warns that the environmental cost of AI is being mismeasured: focusing on carbon alone hides the water and land footprints of data centers. By 2030, AI data centers could consume 945 TWh of electricity, with a water footprint equal to the basic needs of 1.3 billion people in Sub-Saharan Africa and a land footprint twice the size of Jakarta. The report highlights that inference, not training, drives most energy use, and that low-carbon choices can worsen water and land impacts.
If we keep judging AI sustainability by carbon alone, we might think that renewables make AI infrastructure clean but that is solving one problem while creating other problems, often in places that didn’t ask for it.
- VCFundedGenYer
Really sad and telling to see this flagged. HN doesn't care about the environment.
- maelito
Well, extremely large US models trained all-year will way too much money, yes. Not models like Mistral, they consume way less resources.
- onion2k
Now do Netflix and YouTube! How about golf courses, almond farming, and Disneyworld?
Everything has a downside attached. Most things are terrible for the environment if you look at the raw numbers, and if you're not interested in them or they're things you don't use then it's easy to just dismiss them as pointless nonsense that we should stop doing.
However, if you're only looking at the downside then you're being dishonest. The point is never "Let's stop this because it's bad for the planet", but "Is the net benefit worth the cost?"
That could still be a no for AI, but there are things we ought to stop doing long before we come for AI. Notably golf. It's a terrible waste of land and water, and boring.
- api
I challenge people to look at some numbers here.
Is it possibly excessive? Yes. But this is insane hyperbole.
Of course that's how you get attention, I guess. Rational discourse is ignored. To get people to pay attention you have to make everything extreme and stupid.
- throwthrowuknow
The title is incredibly misleading. AI is not using anything that would go towards the “billions” presented in the article. So unless they are proposing moving the billions to North America in lieu of building datacenters then I don’t see a valid argument.