Microsoft
How much water does Microsoft use for AI?
Microsoft's water consumption jumped 34% in 2022, the year GPT-4 trained on its Iowa supercomputer and ChatGPT launched: the clearest fleet-level signal of AI's water cost on record.
+34%
year-over-year water consumption growth, 2022
68.5M gal
used in 2024 by its West Des Moines, Iowa campuses, where GPT-4 trained (local reporting)
Microsoft doesn't sell a frontier model of its own so much as it sells the floor the frontier runs on: Azure is OpenAI's compute provider, and Copilot rides on OpenAI models. That makes Microsoft's environmental reports the closest thing to a fleet-level measurement of AI's water cost.
The headline figure: in 2022, the year GPT-4 trained and ChatGPT launched, Microsoft's global water consumption rose 34% to roughly 1.7 billion gallons. Researchers connected the dots to specific places; local reporting put the West Des Moines, Iowa campuses where GPT-4 trained at 68.5 million gallons in 2024, drawn from the same watershed that supplies the city.
Microsoft has since promised 'water positive by 2030' and publishes replenishment numbers alongside consumption. Replenishment is real but happens in different watersheds than the withdrawals. The almond grower's question ('whose water, where?') applies to data centers too.
For your own Copilot or ChatGPT share, the calculator's frontier class covers it; Microsoft's contribution is the data center, not a separate model.
Your number
What does your AI actually use?
Pick your model, set your usage, get the number, with sources.
Sources
Cited on this page, in order of appearance.
- 1Microsoft (2023). Microsoft Environmental Sustainability Report. Microsoft corporate responsibility: data-center water consumption +34% YoY in 2022. (accessed 2026-06) official disclosure
- 2Wisconsin Public Radio (citing West Des Moines Water Works) (2025). Microsoft built 5 data center campuses in this Iowa city. Here’s what Wisconsin can expect.. WPR, 2025: Microsoft’s five West Des Moines data center campuses used 68.5M gal of water in 2024, 2.62% of the city system’s total, per the water works’ general manager. (accessed 2026-09) news
- 3Li, P., Yang, J., Islam, M. A., & Ren, S. (2025). Making AI Less "Thirsty": Uncovering and Addressing the Secret Water Footprint of AI Models. Communications of the ACM, 2025 (UC Riverside; arXiv:2304.03271): GPT-3-era inference consumes a 500 mL bottle per ~10-50 medium-length responses (on-site cooling + electricity-generation water); GPT-3 training evaporated ~700,000 L on site (~5.4M L including off-site electricity water). (accessed 2026-09) peer-reviewed
- 4Shehabi, A., Smith, S. J., Hubbard, A., et al. (2024). 2024 United States Data Center Energy Usage Report. Lawrence Berkeley National Laboratory, LBNL-2001637: 176 TWh electricity and 17B gal direct water in 2023, with 2028 projections. (accessed 2026-06) government