The AI water footprint: A hidden cost that's not so tiny
The world is abuzz with the latest technological marvels, and at the forefront of this revolution is Artificial Intelligence (AI). But amidst the excitement, a hidden cost has emerged, and it's not just about the energy consumption. It's the water footprint of AI, and it's a topic that demands our attention and understanding.
In my opinion, the fact that AI's water footprint is often overlooked is a significant issue. While the energy consumption of AI has been widely discussed, the water usage is a more complex and less visible problem. The water footprint of AI is not just about the direct use of water in data centers, but also the indirect water use associated with electricity generation and the cooling systems that keep servers running.
One thing that immediately stands out is the discrepancy in estimates of AI's water use. A 2024 Washington Post analysis estimated that writing a single 100-word email with ChatGPT consumes approximately the volume of a standard bottle of water. However, a Google-authored arXiv paper reported that the median Gemini text prompt used only 0.26 milliliters of water. What makes this particularly fascinating is the realization that these estimates are not just about the number of drops or bottles, but about the complex interplay between different systems, time periods, assumptions, and boundaries.
From my perspective, the location of AI infrastructure is a critical factor in understanding the water footprint. A 2026 Guardian analysis revealed that 517 of 809 planned U.S. data centers were in locations that had been in drought conditions during the previous year. This raises a deeper question: how does the location of AI infrastructure impact the local water systems and communities? It's not just about the water used in data centers, but also the water needed to build and power them, and the water footprint of the electricity supply.
What many people don't realize is that the water footprint of AI is not just a local issue, but a global one. A 2023 paper by Pengfei Li, Jianyi Yang, Mohammad A. Islam, and Shaolei Ren estimated that global AI demand could account for 4.2 to 6.6 billion cubic meters of water withdrawal in 2027. This is a model-based projection, not a measurement of all AI systems today, but it gives scale to a problem that is otherwise easy to hide inside the smooth surface of a chat window.
In my view, the real question is not about the recklessness of every use of AI, but about the industry's growth outpacing public understanding of its physical demands. AI feels weightless because its interface is text on glass, but the machinery behind it is not. It's chips, servers, cooling towers, substations, power plants, supply chains, and local water systems. A 100-word email is small, but the infrastructure built to answer millions of such requests is not.
What this really suggests is that we need to take a step back and think about the broader implications of AI's water footprint. It's not just about the water used in data centers, but also the water needed to build and power them, and the water footprint of the electricity supply. Until those answers are reported consistently, the public will keep seeing AI as a clean digital service while communities near the pipes, pumps, and cooling systems deal with the physical cost. Personally, I think that this is a critical issue that needs to be addressed, and I hope that this article will spark a conversation about the hidden cost of AI's water footprint.
