AI’s water use now exceeds global bottled water consumption: study
text_fields
Artificial intelligence systems may now be consuming more water each year than the world’s total bottled water use, while producing a carbon footprint comparable to that of New York City, according to a new peer-reviewed study published in 2025.
The research raises fresh concerns about the environmental costs of AI at a time when global demand for AI services is accelerating rapidly.
The study, titled “The carbon and water footprints of data centres and what this could mean for artificial intelligence,” was led by Dutch researcher Alex de Vries-Gao.
It focuses on the data centres that power large-scale AI models and applications. Because major technology companies do not clearly distinguish between AI and non-AI workloads in their environmental disclosures, the researchers relied on estimates derived from emissions data, water use benchmarks, and sustainability reports from companies including Google, Meta, and Amazon.
Based on multiple modelling approaches, the study estimates that AI systems could be responsible for between 32.6 and 79.7 million tonnes of carbon dioxide emissions in 2025.
This level of emissions is roughly equivalent to the annual carbon footprint of New York City, one of the highest-emitting urban centres in the world. The findings on water use are even more striking. AI-related data centre operations are projected to consume between 312.5 and 764.6 billion litres of water annually, a figure that exceeds global bottled water consumption each year.
The research challenges the common assumption that training large models is the main environmental burden. Instead, it identifies AI inference, the continuous computation required to answer user queries, generate images, and run virtual assistants, as the dominant driver of energy and water use. The study argues that efficiency gains in data centres are being outpaced by rising demand, turning AI into a growing environmental and water security concern that can no longer be overlooked.