Green AI in Industry Quasi-Experimental Evidence from the Water Treatment Sector

This paper studies the environmental effects of the large-scale deployment of AI-driven monitoring systems across French wastewater treatment plants operated by a global leader in water supply services. Exploiting quasi-experimental variation in both the timing of adoption and outages thanks to high-frequency data, we estimate the causal impact of AI on energy use and carbon emissions. We find that AI has allowed treated plants to reduce their electricity consumption and carbon emissions by 5.4% and 6% respectively, and electricity costs by 8.2%, resulting in negative abatement costs, and still improved water effluent quality. The additional electricity demand generated by AI servers represents less than 1% of these savings. Beyond average effects, AI-equipped plants exhibit greater resilience to high operational stress, including extreme meteorological events and chemical pollutant peaks. They also improve load management by reallocating consumption from peak toward off-peak hours. Finally, we assess the aggregate implications of our findings for global climate and welfare using the DICE model. We find large estimated global welfare gains associated with our estimate of AI-induced CO2 reductions.