Artificial intelligence (AI) technologies have a significant hidden water footprint, with a single ChatGPT query consuming about half a liter of water. Research indicates that by 2027, AI could demand between 4.2 and 6.6 billion cubic meters of water, primarily from rivers and aquifers. This increasing consumption necessitates sustainable practices. Innovative solutions, like repurposing water from old coal mines for cooling, are being explored, highlighting the need for transparency in water usage as AI infrastructure expands.
The Hidden Water Footprint of Artificial Intelligence
Recent discussions surrounding artificial intelligence (AI) have highlighted an intriguing statistic: a single query on ChatGPT is estimated to consume about half a liter of water. This figure emerged during an international summit in Paris where experts examined the environmental impact of AI technologies. While the focus often rests on the energy consumption and resource use, the aspect of water usage frequently remains in the shadows.
AI technologies rely heavily on water for multiple purposes, particularly in their operational processes. For instance, Google’s data centers reported a 14% increase in water use in 2023, totaling 24 million cubic meters—equivalent to 9,600 Olympic swimming pools. Similarly, Microsoft, which has a significant stake in OpenAI, experienced a 22% rise in water consumption, reaching 7.8 million cubic meters during the same period.
Understanding the Impact on Global Water Resources
Research conducted by scholars from the University of California, Riverside, and the University of Texas at Arlington aims to quantify the water footprint associated with AI. Their findings suggest that by 2027, the water demand from AI could soar between 4.2 and 6.6 billion cubic meters, which would be four to six times the annual water consumption of Denmark or half that of the United Kingdom.
The study emphasizes that AI’s water usage is predominantly blue water—sourced from rivers, lakes, or aquifers—unlike agriculture, which relies largely on green water absorbed by soil. With the increasing demand for AI, the need for sustainable water practices becomes critical. Researchers caution that the local context significantly influences water consumption, as the type of water used for cooling varies across regions. For example, 10 searches on ChatGPT may consume 500 ml of water in Washington, while the same amount would be sufficient for 31 searches in the Netherlands.
Experts urge a nuanced understanding of water resource management, highlighting that the impact of AI on water usage cannot be generalized. The local availability and type of water used for cooling data centers are crucial factors that determine overall consumption.
In cities like Marseille, innovative solutions are being explored to mitigate water use. Digital Realty, an American data center operator, utilizes water from an old coal mine for cooling, which conserves drinking water resources and reduces energy consumption related to cooling operations. This method has reportedly allowed them to minimize air conditioning reliance and cut energy use by 20%. However, some officials argue that this approach diverts a resource that could serve other important urban needs.
France is also promoting strategies to recycle hot water from data centers to support nearby industries and residential areas. This initiative aims to strike a balance between technological growth and the essential needs of citizens, as officials emphasize the need for environmental stewardship in the face of expanding AI infrastructure.
Finally, as the demand for data centers grows, there’s a collective call from experts for improved transparency regarding water and energy consumption. Recommendations have been made for companies to disclose their water usage and adhere to environmental standards to ensure that the growth of AI technologies does not compromise vital resources for the communities they serve.