Key Points
- Training GPT-3 in Microsoft’s US data centers likely evaporated 700,000 liters of fresh water, equivalent to the amount needed to produce 370 BMW cars.
- A typical conversation with ChatGPT (10-50 queries) consumes roughly a 500ml bottle of water, a hidden cost that scales with millions of daily users.
- Data centers often rely on evaporative cooling, which loses water to the atmosphere; this water cannot be recycled back into local watersheds.
- AI’s water footprint is geographically concentrated, exacerbating droughts in places like Iowa, where on-site water consumption quadrupled in a year to cool servers.
- Unlike electricity, water is hyper-local and non-fungible—a data center can’t simply buy “water credits” from a wetter state to offset its impact in an arid zone.
Why It Matters
The water crisis reveals that AI’s sustainability challenge extends far beyond carbon emissions, demanding urgent transparency and strategic siting before growth is paralyzed by parched communities pushing back.
Read More
Explore more coverage: At the Arctic Games, Canada and Greenland vs. Trump Feels Like Its Own Sport, Mob Burns Congo Ebola Center Amid Rare Strain Outbreak, Rare Rubens notebook sheet goes on display in artist’s home city of Antwerp, I tried Amazon’s Bee wearable and am both intrigued and slightly creeped out
