EN CZ
Bitter

Making AI Less Thirsty: The Hidden Water Footprint of ChatGPT and Large Language Models

Li et al. (2023) produced the first systematic analysis of AI's water footprint, an environmental cost that had been almost entirely overlooked:

  • GPT-3 training consumed approximately 700,000 liters of fresh water for cooling data centers
  • A typical ChatGPT conversation (20-50 questions) uses roughly 500 ml of water — about a standard water bottle
  • Microsoft's water consumption increased by 34% in 2022, largely attributed to AI workloads
  • Google's water consumption rose by 20% in the same period

Why Water Matters

Data centers require enormous amounts of water for cooling. This is particularly concerning because many data centers are located in water-stressed regions. As AI usage scales globally, the water footprint could exacerbate existing water scarcity crises.

Source

Li, P. et al. (2023). Making AI Less "Thirsty". arXiv:2304.03271.

Connected Research

You may also like