AI’s growth is depleting water reserves

AI is often portrayed as the harbinger of a prosperous, more efficient future. But the machines driving this revolution depend on a resource far older — and far more contested — than data or electricity: water.

- FRIEDERIKE ROHDE PAZ PENA

The village of La Esperanza near Microsoft's data centre in Querétaro, Mexico, is seen on July 26. As tech companies build data centres worldwide to advance artificial intelligence, vulnerable communities have been hit by blackouts and water shortages. ©2025 THE NEW YORK TIMES

(©2025 THE NEW YORK TIMES)

As the Heinrich B6ll Foundation’s recent Water Atlas makes clear, Al's rapid growth is depleting local water reserves around the world, from drought-stricken Chile to South Africa. Its physical footprint reflects a new form of colonial extraction; instead of silver and soy, now it is the cooling water that keeps the digital economy running.

While the debate about Al's energy use focuses on the power needed to train and operate large language models, what is often overlooked is the vast amount of water required to cool data centres, not to mention the water used in energy production and hardware manufacturing.

ChatGPT is a prime example. Training GPT-3 required roughly 700,000 litres of water for cooling alone. A Greenpeace study estimates that data centres will consume 664 billion litres annually by 2030, compared to 239 billion litres in 2024.

Al's benefits are concentrated in the Global North, yet its environmental costs increasingly fall on the Global South. In 2023, mass protests erupted in Uruguay over a proposed Google data centre as the country suffered its worst drought in 70 years. With reservoirs running dry, authorities began pumping brackish water from the Rio de la Plata estuary into public systems, granting Google permits to draw from the remaining freshwater reserves even as working-class families boiled salty tap water to drink.