I know a lot of people are concerned about the environmental impact of AI, so I keep an eye out for detailed information about it. I came across two articles today (Slow Boring references Andy Masley):

• Slow Boring: There’s plenty of water for data centers
• Andy Masley: Why using ChatGPT is not bad for the environment (this is a very long summary of an even longer post – I have skimmed both of them)

Short version:

• Water complaints from data center construction seem to be more about water quality issues from construction (e.g. well water sediment), which would be caused by constructing anything of that scale – distribution centers, warehouses, shopping centers, etc
• Water use is much less significant than energy use
• AI computation uses much less water and energy than other data center uses e.g. Zoom, streaming video
• Water use by AI is vanishingly small compared to things like raising cattle or leaking pipes
• ChatGPT queries just don’t add up to much compared to everyday energy usage tasks (e.g. 1 query = vacuum cleaner for 10 seconds, use laptop for 3 minutes) or water consumption (200,000 queries = one hamburger)

My take is that while AI energy usage will be a significant new industrial consumer, it’s not significant in either water or energy usage compared to other uses, and the impacts on capital and labor markets will be orders of magnitude more impactful than environmental impact.