AI's water consumption in 2026 is estimated to range between 312.5 billion and 764.6 billion liters, with per-prompt estimates varying by over 10,000 times due to infrastructure and location differences.
TL;DR
- AI's water consumption is a growing concern, with estimates for 2026 ranging from 312.5 billion to 764.6 billion liters.
- Per-prompt water usage varies significantly based on infrastructure, location, and measurement boundaries.
- AI's environmental impact extends beyond water consumption, including electricity demand, carbon emissions, and hardware production.
What happened
A 2025 study published in Patterns estimated that operating AI systems could consume between 312.5 billion and 764.6 billion liters of water in 2025, excluding supply-chain and end-of-life impacts.
The United Nations University reported that the electricity-associated water footprint of all data centers was 4.5 trillion liters in 2025, with AI accounting for about one-fifth of data center electricity use.
Per-prompt water usage estimates vary widely, with Google's Gemini Apps text prompt consuming 0.26 mL per median prompt, while Mistral's Le Chat consumes 45 mL for a 400-token response.
Why it matters
AI's water consumption is a critical environmental concern, with significant variations based on infrastructure and location.
Understanding the water footprint of AI is essential for developers, startups, and investors to make informed decisions about sustainability and resource management.
The lack of standardized measurement methods makes it challenging to compare water consumption across different AI models and data centers.
Key facts
- AI systems consumed an estimated 312.5 billion to 764.6 billion liters of water in 2025, excluding supply-chain and end-of-life impacts (Patterns study).
- Global data centers carried a 4.5 trillion-liter water footprint in 2025, with AI accounting for about one-fifth of data center electricity use (United Nations University estimate).
- US data centers directly consumed 17.4 billion gallons of water in 2023, with indirect water consumption associated with electricity generation estimated at 211 billion gallons (Lawrence Berkeley National Laboratory and Federal estimate).
- Global AI demand may withdraw 4.2 billion to 6.6 billion cubic meters of water in 2027, with consumption estimated at 0.38 billion to 0.60 billion cubic meters (Academic forecast in Making AI Less Thirsty).
- Google's Gemini Apps text prompt consumed 0.26 mL of water per median prompt in May 2025 (Google analysis).
- Mistral's Le Chat consumed 45 mL of water for a 400-token response, excluding the user's terminal (Mistral lifecycle assessment).
- Training GPT-3 with 175 billion parameters consumed 5.4 million liters of water, including 700,000 liters of direct on-site consumption (Making AI Less Thirsty).
- Mistral Large 2 consumed 281,000 cubic meters of water as of January 2025 after 18 months of use (Mistral AI lifecycle assessment).
Context
AI's environmental impact is a growing concern, with water consumption being a significant factor alongside electricity demand, carbon emissions, and hardware production.
The lack of standardized measurement methods and varying boundaries make it challenging to compare water consumption across different AI models and data centers.
Understanding the water footprint of AI is essential for developers, startups, and investors to make informed decisions about sustainability and resource management.