Key Points
- Training GPT-3 in Microsoft’s U.S. data centers consumed an estimated 700,000 liters of freshwater — roughly equivalent to what it takes to produce 370 BMWs.
- Water evaporates in cooling towers and is used to humidify server rooms, meaning the water isn’t returned to the local watershed and is “consumed” rather than simply cycled.
- Running inference at scale is the hidden drain: 10–50 ChatGPT responses require about 500 milliliters of water, making continuous user interactions a massive cumulative load.
- AI’s water use varies wildly by location and season; a model trained in a drought-stricken desert will stress local reserves far more than one in a temperate, water-rich region.
- Researchers call for mandatory transparency, including water-use audits and runtime “water intensity” labels, so developers can schedule training in cooler, water-abundant zones.
Why It Matters
As AI becomes embedded in every industry, its invisible thirst could trigger battles over water rights, forcing communities to choose between sustainable tech growth and basic human needs. Without immediate disclosure standards, the environmental cost of your next AI-assisted task remains dangerousl
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
