The study has been published in the scientific journal Patterns.
Since the launch of ChatGPT at the end of 2022, demand for AI applications has grown explosively. In 2024, an estimated 15–20% of global data centre electricity consumption was used for AI applications. Data scientist Alex de Vries-Gao set out to map how large the global carbon and water footprints of AI systems could be in 2025. In earlier research, De Vries-Gao estimated that AI systems could require up to around 23 gigawatts of power in 2025. According to international energy statistics he uses as a reference, this is of the same order of magnitude as the average electricity consumption of the United Kingdom.
In this new study, he calculates what that would mean for the global carbon and water footprints of AI. To do so, he combined sustainability reports from major tech companies, scientific data on carbon emissions and water use per kilowatt-hour of electricity, and compared these with the International Energy Agency (IEA) report. On the basis of this analysis, he calculated how much water would be needed to supply the electrical power that AI is expected to demand, how much associated CO₂ emissions this would produce, and how this compares to the emissions and water use of countries and cities.
Carbon footprint comparable to New York City
These calculations show that AI systems could be responsible for 32.6 to 79.7 million tonnes of CO₂ emissions per year. For comparison: in 2023, New York City emitted 52.2 million tonnes of CO₂. De Vries-Gao says: “AI could therefore end up with a climate footprint comparable to that of a major world city or other energy-intensive sectors.”
AI as a major water consumer
De Vries-Gao, who has been studying the climate impact of AI for some time, then examined the water footprint of this technology. Data centres use large amounts of water to cool AI systems and indirectly through the generation of the electricity they consume. The International Energy Agency (IEA) estimated that data centres used a total of 560 billion litres of water in 2023. Drawing on corporate sustainability reports from Apple, Google and Meta, De Vries-Gao shows in his study that this indirect water use is significantly underestimated and is probably a factor of three to four higher than the official estimate. De Vries-Gao: “Based on my calculations, AI systems alone use between 312.5 and 764.6 billion litres of water. That is of the same order of magnitude as all bottled water consumed worldwide in a single year.”
More transparency
Major tech companies currently do not publish AI-specific figures on the energy and water use of their systems. Nor do they generally disclose much, if anything, about the performance of their data centres. In particular, very little is known about the water required to generate the electricity used by those data centres. In an explanatory note to a recent report on the environmental impact of its Gemini model, Google even stated that it did not wish to report this indirect water use because it does not fully control water consumption at power plants. According to De Vries-Gao, this ignores the fact that such water use directly results from the company’s own electricity demand, just as companies are already required under the Greenhouse Gas Protocol to report their indirect CO₂ emissions from purchased electricity. “It is highly problematic that these companies currently share so little information about this,” he says.
De Vries-Gao therefore calls on governments and companies to introduce stricter reporting requirements: full transparency at data centre level on energy use, carbon emissions and water consumption, with a clear distinction between AI and other applications. De Vries-Gao: “If we want AI to genuinely contribute to a sustainable future, we first need a clear picture of its environmental cost to society. That starts with transparency.”