Water consumption: AI vs animal agriculture

With the changes to web development ushered in by AI, there seems to be less point in posting the kind of ‘how-to’ technical articles that I’ve been writing here thus far. I thought I’d try blogging about other things…

⁂

There has been plenty gnashing of teeth online over AI water consumption. LLMs and the like do indeed use a lot of water, which is a concern, especially in localities hosting data centres. However, it’s worth maintaining some perspective.

AI water consumption is predicted to be between 0.38 and 0.60 km³ in 2027. Animal agriculture, by comparison, is much, much worse, consuming 228 km³ annually. Breeding animals for humans to kill and eat consumes roughly 380× to 600× more water than AI does. Another way to think of it: in 2027 AI is predicted to consume between a twentieth and a twelfth of the volume of water in Loch Ness (Scotland’s largest loch by volume, and the largest body of water by volume in Britain), whereas animal farming already consumes just over 30 entire Loch Nesses annually!

(‘Consumption’ here means water that’s evaporated and lost, as opposed to ‘withdrawal’, which is largely returned to the system. Water withdrawal for AI in 2027 is predicted to be between 4.2 and 6.6 km³. Animal agriculture, in contrast, withdraws 566 km³ per year: 513 km³ for feed, 40 km³ for drinking water and 13 km³ for ‘service’ [non-potable] water.)

There are of course other environmental impacts of both AI and animal agriculture, not to mention the serious ethical violations inherent in the latter.

But if water consumption and the environment in general concern you, you’re much better off adopting a plant-based diet (and advocating for it to others, including institutions and governments) than worrying too much about AI.