52 AI Water Usage Statistics for 2026

ByShane Tran Updated on

AI runs on water as much as it runs on electricity. Data centers use it for cooling, and the power plants feeding them evaporate even more.

Here are 52 AI water usage statistics for 2026.

Top AI Water Usage Statistics

  • GPT-3 consumes roughly a 500 ml bottle of water for every 10 to 50 medium-length responses (Li et al.).
  • Global AI’s total water withdrawal is projected to reach 4.2 to 6.6 billion cubic meters in 2027 (Li et al.).
  • U.S. data centers directly consumed about 66 billion liters of water for on-site cooling in 2023, plus nearly 800 billion liters indirectly at power plants (LBNL).
  • Google consumed about 10.9 billion gallons of water across its data centers and offices in 2025, up 34% in a single year (Google).
  • Microsoft consumed 8,170 megaliters of water in FY25 at a datacenter water usage effectiveness of 0.27 liters per kWh (Microsoft).
  • The IEA estimates global data-center water consumption could rise from around 560 billion liters a year to about 1,200 billion liters by 2030 (IEA).
  • In July 2022, the month before OpenAI says it finished training GPT-4, Microsoft’s West Des Moines, Iowa cluster drew about 11.5 million gallons of water, roughly 6% of the district’s total (AP).

How Much Water Does AI Use?

A widely cited University of California, Riverside study traced the water behind generative AI back to its cooling towers and power plants, and its estimates anchor much of the discussion below.

1. GPT-3 consumes roughly a 500 ml bottle of water for every 10 to 50 medium-length responses, depending on when and where it runs (Li et al.).

2. Global AI’s total water withdrawal is projected to reach 4.2 to 6.6 billion cubic meters in 2027, more than the yearly water withdrawal of 4 to 6 Denmarks (Li et al.).

Source: Li et al., “Making AI Less Thirsty” (arXiv 2304.03271)

Water Use Per AI Query

Per-query water use is small on its own but adds up fast, and it swings widely with a data center’s climate and power source.

3. On the U.S. average, a single medium GPT-3 request consumes about 16.9 ml of water total, combining 2.2 ml of on-site cooling and 14.7 ml at the power plant (Li et al.).

4. Per-request water use swings with location, from about 7.6 ml in Texas to 47.5 ml in Washington (Li et al.).

5. OpenAI’s Sam Altman says the average ChatGPT query uses about 0.000085 gallons of water, roughly one-fifteenth of a teaspoon (Altman).

6. Data centers can evaporate roughly 1 to 9 liters of water per kilowatt-hour of server energy on-site, depending on climate and location (Li et al.).

Water to Train an AI Model

Training a large model concentrates months of water use into one facility over a short window.

7. Training OpenAI’s GPT-3 in Microsoft’s U.S. data centers can consume a total of 5.4 million liters of water, counting both on-site cooling and off-site electricity (Li et al.).

8. Of that, about 700,000 liters of clean freshwater is evaporated directly on-site to train GPT-3 (Li et al.).

9. In July 2022, the month before OpenAI says it finished training GPT-4, Microsoft’s West Des Moines, Iowa cluster drew about 11.5 million gallons of water, roughly 6% of the district’s total (AP).

Source: Associated Press, using West Des Moines Water Works records

Data Center Water Use

AI runs inside data centers, so the industry’s water footprint is the wider container for every model and query. Most of that water is never used on-site at all.

10. U.S. data centers directly consumed about 66 billion liters of water for on-site cooling in 2023 (LBNL).

11. Counting the water used at power plants to make their electricity, U.S. data centers had an indirect water footprint of nearly 800 billion liters in 2023 (LBNL).

Source: LBNL, 2024 United States Data Center Energy Usage Report, p. 55–57

12. Hyperscale and colocation facilities accounted for 84% of U.S. data centers’ direct water use in 2023 (LBNL).

13. U.S. data centers directly consumed 21.2 billion liters of water in 2014, most of it in enterprise facilities (LBNL).

14. Every kilowatt-hour a U.S. data center used in 2023 carried a national-average 4.52 liters of indirect water consumption at the power plant (LBNL).

15. The industry-wide average on-site water usage effectiveness (WUE) has held just over 0.36 liters per kWh across U.S. data centers through 2023 (LBNL).

16. A peer-reviewed study put U.S. data centers’ total operational water footprint at about 513 billion liters in 2018 (Siddik et al.).

17. Roughly three-fourths of U.S. data centers’ operational water footprint comes from indirect dependencies rather than on-site cooling (Siddik et al.).

18. The data center industry draws water, directly or indirectly, from 90% of U.S. watersheds (Siddik et al.).

Google Water Usage

Google ties its rising water use directly to demand for data centers, stating that as AI expands, so does demand for facilities “that require increased energy for operations and water for cooling.”

19. Google withdrew 14,689 million gallons of water across its global data centers and offices in 2025, up from 11,011 million gallons in 2024 (Google).

20. Google consumed about 10.9 billion gallons (41 billion liters) of water across its data centers and offices in 2025 (Google).

21. Google’s water consumption rose 34% from 2024 to 2025, which the company ties to rising demand for digital services (Google).

22. Google’s data centers alone consumed 10,523 million gallons of water in 2025 (Google).

Source: Google 2026 Environmental Report, p. 96

23. Google’s total water consumption climbed from 4,562 million gallons in 2021 to 10,869 million gallons in 2025 (Google).

24. Google’s water-stewardship projects replenished about 7.7 billion gallons in 2025, roughly 78% of its freshwater consumption (Google).

25. In 2025, 72% of Google’s freshwater withdrawals came from sources at low risk of water depletion or scarcity (Google).

26. In The Dalles, Oregon, Google’s data centers consumed 355.1 million gallons of water in 2021, about 29% of the city’s total (The Oregonian via DCD).

Microsoft Water Usage

Microsoft has invested in redesigning cooling to cut water, even as its overall footprint keeps climbing.

27. Microsoft’s total water withdrawals reached 13,266 megaliters (about 3.5 billion gallons) in FY25, up from 11,590 ML in FY24 (Microsoft).

28. Microsoft consumed 8,170 megaliters (about 2.2 billion gallons) of water in FY25, up from 6,693 ML in FY24 (Microsoft).

29. Microsoft reported a global average datacenter water usage effectiveness of 0.27 liters per kWh in FY25, a 25% reduction since its 2022 baseline (Microsoft).

30. Microsoft has pledged to become water positive by 2030, replenishing more water than it withdraws (Microsoft).

31. In FY25 Microsoft replenished over 14.2 million cubic meters of water, for the first time surpassing its global withdrawals (Microsoft).

32. About 50% of Microsoft’s FY25 water withdrawals (6,659 ML) came from areas with water stress (Microsoft).

33. Microsoft says its new zero-water-cooling datacenter design saves over 125,000 cubic meters of water per year per facility (Microsoft).

Amazon and Meta Water Usage

Amazon and Meta both report water-use effectiveness and community replenishment, though each discloses a different mix of figures.

34. Amazon’s AWS reached 53% of its water-positive-by-2030 goal in 2024, up from 41% in 2023 (Amazon).

35. AWS reported a global datacenter water usage effectiveness of 0.15 liters per kWh in 2024, a 17% improvement over 2023 (Amazon).

36. AWS’s active replenishment projects returned 4.3 billion liters of water to communities as of 2024 (Amazon).

37. AWS cut its datacenter cooling water needs by 946 million liters in North America in 2024 (Amazon).

38. Meta’s data centers withdrew 4,145 megaliters of water (about 1.1 billion gallons) in 2024 (Meta).

39. Meta’s annual datacenter water usage effectiveness was 0.19 liters per kWh in 2024 (Meta).

40. Meta’s restoration projects returned more than 1.59 billion gallons of water in 2024 across watersheds where it runs data centers (Meta).

41. Meta expects its restoration portfolio to restore 2.9 to 3.4 billion gallons of water annually once fully implemented (Meta).

AI, Water Stress, and Local Impact

Aggregate numbers hide where the water actually comes from. A large share of it is drawn in regions that were already short.

42. One-fifth of U.S. data center servers’ direct water footprint comes from moderately-to-highly water-stressed watersheds (Siddik et al.).

43. Nearly half of U.S. data center servers are powered, fully or partly, by plants sitting in water-stressed regions (Siddik et al.).

44. In the Phoenix region, indirect water use tied to data centers is projected to rise 400%, from about 2.9 billion gallons to over 14.5 billion gallons a year (Ceres).

The Future: AI Water Demand Projections

Nearly every projection points the same way: sharply up, and concentrated in the second half of the decade.

45. Global AI demand is projected to evaporate, or consume, 0.38 to 0.60 billion cubic meters of water in 2027 (Li et al.).

46. On-site water consumption attributed to AI in the U.S. alone could exceed 0.2 billion cubic meters in 2028 (Li et al.).

47. Counting indirect use, AI’s total water consumption in the U.S. alone is projected to reach up to about 2 billion cubic meters in 2028 (Li et al.).

48. Total on-site water consumption by all U.S. data centers in 2028 could double or quadruple the 2023 level, reaching about 150 to 280 billion liters (Li et al.).

49. U.S. hyperscale data centers alone are projected to consume between 60 and 124 billion liters of water directly by 2028 (LBNL).

50. The IEA estimates global data-center water consumption at about 560 billion liters a year now, rising to around 1,200 billion liters by 2030 in its Base Case (IEA).

Source: IEA, Energy and AI (2025), p. 242

51. Water consumption from chip manufacturing for data centers is projected to grow more than 50% to around 70 billion liters by 2030 (IEA).

52. The IEA estimates a 100 MW hyperscale data center in the U.S. consumes around 2 million liters of water a day, about the same as 6,500 households (IEA).

Conclusion

That wraps up our list of AI water usage statistics. The trend across every company and projection is the same: water use is climbing, and the infrastructure plans already in motion will push it higher through the end of the decade. We update this page as new data is published.