AI Could Make the Planet Thirstier

What price does the planet pay for every word generated by artificial intelligence? This question arises when examining the resources required to sustain these advanced models. Behind every interaction with a chatbot—a software program designed to simulate human conversation—lies a complex technological infrastructure whose operation consumes significant resources and leaves a substantial environmental footprint.

For example, generating a 100-word text with ChatGPT consumes, on average, approximately half a liter of water. While this amount may seem insignificant on the scale of a single query, its impact becomes enormous when viewed globally. If only 10% of the active population in the United States used this service weekly, annual water consumption would exceed 435 million liters, according to analyses by The Washington Post and researchers at the University of California, Riverside.

To operate efficiently, data centers require constant cooling. By 2030, water consumption associated with artificial intelligence could be equivalent to the annual needs of 1.3 billion people in Sub-Saharan Africa—nearly three times the combined yearly consumption of Pakistan, Bangladesh, and Nigeria, countries with a total population of approximately 650 million people.

Adding to these concerns are carbon dioxide (CO₂) emissions, which could reach 400 million metric tons. Likewise, the physical footprint of AI infrastructure is expected to occupy roughly 14,500 square kilometers—an area nearly ten times the size of Mexico City.

Sources: National Geographic, Ekopak Water, El País, The Washington Post.