Water Use Isn’t a Data Center Problem, It’s an AI Problem

2026-05-11 · Source: The Information · Depth: Unknown, quick

Summary

The increasing water consumption by data centers, particularly those supporting AI, is emerging as a significant environmental concern. Google's data centers alone consumed 5.6 billion gallons of water in 2022, a 20% increase from the previous year, primarily due to the cooling demands of powerful AI chips. Microsoft's water usage also surged by 34% in 2022, reaching 1.7 billion gallons. This escalating demand is exacerbated by the fact that many data centers are located in water-stressed regions, such as Arizona and Nevada, where water resources are already scarce. The training of large language models (LLMs) like GPT-3 can consume hundreds of thousands of gallons of water, highlighting the substantial environmental footprint of AI development and deployment.

Key takeaway

For data center operators and AI developers, understanding and mitigating the environmental impact of water usage is critical. Your infrastructure decisions, especially regarding cooling systems and data center locations, directly influence regional water scarcity. Prioritize deploying AI workloads in facilities that utilize advanced, water-efficient cooling technologies or are situated in areas with abundant, sustainable water sources to reduce your ecological footprint.

Key insights

AI's rapid growth significantly escalates data center water consumption, particularly in water-stressed regions.

Principles

• AI training is water-intensive.
• Data centers often locate in arid regions.

In practice

• Evaluate data center water footprint.
• Prioritize water-efficient cooling solutions.

Related on AIssential

• When it comes to total water use, AI data centers are a drop in the bucket — AI data center water use is globally minor but locally significant, prompting corporate stewardship.
• Can you explain to me the hatred of AI and its water usage? — AI's water usage controversy centers on data center cooling demands in water-stressed regions, impacting local communities.
• The Hidden Water Cost of Every Prompt — Water consumption for data center cooling is emerging as a critical environmental limit for AI.
• How some data center operators are tackling their water use problems — Data center water use is a complex, localized issue with inherent trade-offs between water consumption and energy efficiency.
• Nvidia wants to cut data center water use, but that’s not the same as fixing AI’s water problem — Nvidia's new cooling system addresses only a fraction of AI data centers' total water footprint by ignoring external electricity generation.
• The Cloud Is Not Weightless — AI's substantial and often hidden water footprint is a structural feature of the AI economy, accelerating climate stress.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Information.