Research by Alex de Vries-Gao, affiliated with Vrije Universiteit Amsterdam, shows that AI systems could be responsible for between 32.6 and 79.7 million tonnes of CO₂ emissions annually by 2025. This places their climate impact on par with that of a major global city such as New York. AI-related water consumption could also reach hundreds of billions of litres per year. The study was published in the scientific journal Patterns.

Since the rise of large-scale generative AI, such as ChatGPT in late 2022, demand for AI applications has grown rapidly. In 2024, an estimated 15 to 20 percent of global data centre electricity consumption was attributed to AI workloads. Based on earlier work, De Vries-Gao estimated that AI systems could require up to 23 gigawatts of electrical power by 2025, comparable to the average electricity consumption of the United Kingdom.

The new study examines what this level of energy demand means for AI’s global carbon and water footprint. De Vries-Gao combined sustainability reports from major technology companies with scientific data on emissions and water use per kilowatt-hour and compared these figures with data from the International Energy Agency. The analysis shows that indirect water use, particularly for electricity generation, is consistently underestimated.

According to the study, AI systems alone may consume between 312.5 and 764.6 billion litres of water annually, a volume comparable to global bottled water consumption in a single year. At the same time, major technology companies publish little to no AI-specific data on the energy and water use of their data centres.

De Vries-Gao therefore calls for stricter transparency requirements. Governments and companies should provide full disclosure of energy use, CO₂ emissions and water consumption per data centre, clearly distinguishing AI from other digital services. Without such transparency, the public environmental costs of AI remain largely invisible as the technology continues to scale.