The AI water issue is fake (andymasley.substack.com)

A detailed debunk argues that fears about AI data centers draining U.S. freshwater are overblown. In 2023 all U.S. data centers used roughly 200–250 million gallons/day (~0.2% of U.S. freshwater withdrawals), with only ~50 million gallons/day consumed onsite (~0.04%). AI workloads are estimated to be about 20% of that footprint, so AI’s current water use is roughly 10.6 million gallons/day — about 0.008% of U.S. freshwater. Even accounting for aggressive forecasts (data-center electricity tripling by 2030 and AI energy use rising tenfold), projections put AI water use at around 0.08% of current U.S. freshwater — still tiny compared with major water users like agriculture or golf courses. The wider point for the AI/ML community: context matters. Many high-profile stories present large raw gallons without comparing them to total consumption, other industries, or the split between onsite use and offsite electricity-generation water. Local complaints tied to specific data centers have mostly stemmed from construction impacts (e.g., sediment and groundwater disruption), not routine operations. Policy implications: prioritize careful siting and construction safeguards, focus environmental scrutiny more on electricity and emissions than on water in most U.S. contexts, and use proportional, data-driven metrics rather than alarmist headlines.