Making AI Less "Thirsty": Uncovering and Addressing the Secret Water Footprint of AI Models

The most-cited estimate of AI water use. Its abstract reports that training GPT-3 in US data centers directly evaporated about 700,000 litres of freshwater; the paper also estimates roughly 500 ml consumed per 10 to 50 medium-length responses. A preprint, so treat the numbers as estimates; we use it because public AI water data is scarce.

Read the preprint itself →

We link to the publisher's page. Some journals show the abstract free and charge for full text; the abstract usually covers the findings cited here.