[cross-posted on Substack and the EA Forum]

I estimated how much water the AI industry (e.g., data centers) will use in comparison to other industries (factory farming, non-livestock agriculture, and other uses). You can view the app here and the prompting here.

AI water consumption is not noticeable unless there is a continued explosion in AI demand up until 2050. Continued means roughly the same growth we are seeing today. 

If I were to input a custom growth rate starting at the standard high growth path (700 Billion Litres in 2025, 40% compounding growth) and then taper to 30% and 20% for the next two decades, AI would place as the second highest water usage on the Earth by 2050, behind agriculture. I recommend using the custom growth rate to see whether or not AI would destroy our environment — but even for official low and high estimates by published researchers, AI water use is high but (very) negligible compared to other water uses on Earth. 

So does AI's water use matter, and does this data matter? 

Yes. The data clarifies the dominant narratives surrounding AI's environmental impact. The first narrative is that we must do everything we can to oppose the construction of new data centers, and if we do not, the Earth will dry up, or clean water might disappear. The second narrative is that only isolated communities will suffer from further data center expansion when not done sustainably. 

Narrative one operates on a very high growth rate that is highly unlikely in the short term (e.g., next decade). For example, you would be much better off investing your time in AI safety (misalignment), not sustainability, given the risk timeline. Narrative one, however, is successful in that it presents another challenge to the ruling class. It serves as a cathartic response and rallying cry to ruling class hegemony. It exposes their obliviousness to suffering given data center expansion seems to be one of their top priorities. Individuals in affected communities must continue to fight against new developments to protect their homes, but it might be best for media and outsiders to cover other topics for a higher impact. 

Narrative two, on the contrary, is too short a timeline. If AI expansion continues at this current rate (which is still unlikely given the benchmarks), the entire globe will face catastrophe. This is certainly a possibility, and it seems the risks haven't been calculated. 

So we must find a middle path between these two narratives. This is the case even if AI water use remains low in comparison to other industries for the next decade or so. This is also the case considering AI could be used to improve the efficiency of almost every industry, including oil and gas, to which my app does not account for given the diffuseness of indirect effects. For these reasons, it matters.