ChatGPT’s water usage is 4x higher than previously estimated

The rapidly expanding use of artificial intelligence, particularly large language models like ChatGPT, is creating unprecedented demands on water resources as data centers struggle to cool their increasingly powerful systems.

Updated water consumption data: Recent research from the University of California, Riverside reveals that ChatGPT’s water consumption is four times higher than previously estimated, with 10-50 queries consuming approximately two liters of water.

• The original study, “Making AI Less Thirsty,” based its calculations on 2020 OpenAI figures but has been revised following new data from Microsoft
• Professor Shaolei Ren, the study’s author, indicates that energy consumption and associated water usage are significantly higher than initial estimates

Technical infrastructure challenges: Modern AI server racks generate unprecedented levels of heat, requiring extensive cooling systems that consume massive amounts of water.

• Nvidia’s latest AI servers generate 120kW of power in a single rack, equivalent to 120 traditional space heaters
• Data centers require drinking-quality water for cooling systems, as impurities can damage sensitive equipment
• The United States hosts over 5,000 data centers, compared to approximately 600 in the UK

Environmental impact and corporate response: Major tech companies are reporting significant increases in water consumption while pledging to address environmental concerns.

• Google, Microsoft, and Meta have reported 17-22.5% increases in water consumption
• Tech giants have committed to becoming “water positive” by 2030, promising to return more water than they consume
• Water replacement often occurs in different locations from where it’s extracted, potentially affecting water-stressed areas

Global implications: The rapid expansion of AI infrastructure is creating environmental tensions in various regions worldwide.

• Chile has become a flashpoint for environmental activism, with Google forced to redesign a $200 million data center to use air-based cooling
• Irish data centers now consume 21% of the country’s electricity, up from 5% in 2015
• Water UK projects that new data centers could require water equivalent to a city the size of Liverpool

Innovation and solutions: Companies are developing various approaches to reduce water consumption in data centers.

• Digital Realty sources 43% of its water from non-drinking sources, including rainwater harvesting
• Iceotope’s precision liquid cooling technology eliminates the need for water cooling
• Google DeepMind has developed AI systems that reduce cooling energy consumption by 40%

Looking ahead: While AI systems present significant environmental challenges, they may also offer solutions to improve resource efficiency, creating a complex balance between technological advancement and environmental stewardship.

• AI tools are being deployed to optimize data center water usage
• Environmental concerns are driving innovation in cooling technologies and infrastructure design
• The industry faces increasing pressure to balance rapid growth with sustainable practices