The chart this week comes from Andy Masley, a researcher who has spent the past year challenging claims about AI’s environmental footprint. His interactive visualisation compares US water consumption across industries using public supply (potable) water. US meat and poultry processing consumes 108 billion gallons per year. All US golf courses drink 48 billion. Semiconductor fabs use 9 billion. And all US AI workloads? And all US AI workloads projected to use 45 billion by 2030.
But experts caution against dismissing concerns outright. “In the near term, it’s not a nationwide crisis,” says Cornell professor Fengqi You. “But in locations that have existing water stress, building these AI data centres is gonna be a big problem.” Jonathan Koomey, a leading researcher on data centre efficiency, agrees: “Each situation has to be evaluated in the context of the specific design of the facility being proposed.”