The carbon and water footprints of data centers and what this could mean for artificial intelligence

Alex de Vries-Gao 9 January 2026

The release of OpenAI’s generative artificial intelligence (AI) ChatGPT at the end of 2022 triggered significant global growth in the demand for AI applications. Since then, the power demand of the hardware supporting these applications has been rising rapidly. The International Energy Agency (IEA) estimated that AI systems accounted for 15% of total data center electricity demand (excluding cryptocurrency mining) in 2024,¹ and other recent research has suggested that AI systems represented 20% of total data center power demand by the end of 2024.² Given the increasing manufacturing capacity of the AI system supply chain, the total power demand of these systems could reach 23 gigawatts by the end of 2025.² This would mean that the share of AI system power demand would be almost half of the total data center electricity demand (47.4 gigawatts on average in 2024¹), approaching as much as a country such as the United Kingdom requires (30.7 gigawatts on average in 2023²). This rising power demand could have significant environmental implications. For instance, the IEA estimated that in 2024, the electricity generation for global data centers could have produced approximately 182 million tons of CO₂ emissions,¹ and they were responsible for 560 billion L of water consumption in 2023.¹

The carbon and water footprints of data centers are typically established by first assessing their locations and energy use, which are subsequently linked to characteristics of the respective electrical grids.³ The IEA, however, did not specify which parts of the estimated data center emissions and water consumption could be attributed to AI systems. The main challenge in determining such data is that even if a total AI power demand could be estimated, granular information regarding where AI systems are being operated cannot be obtained, despite past research emphasizing the importance of transparency in data centers⁴ and in information and communication technology (ICT) in general.^5,6 This information is crucial to assess the carbon and water intensity of the electricity generation sources used to power this hardware. Tech companies such as Google, Microsoft, Meta, and Amazon have been identified as the largest buyers of AI systems,⁴ but their corporate sustainability reports do not specify the impacts associated with these systems. Without further input from data center operators, these facilities reveal nothing about the equipment they use.

Although there are ways to estimate the global power demand of AI systems, it remains challenging to quantify the associated carbon and water footprints. The lack of distinction between AI and non-AI workloads in the environmental reports of data center operators means it is only possible to assess the environmental impact of AI workloads by approximating them through data centers’ general performance metrics. The environmental disclosure of tech companies is, however, often insufficient to

assess even the total data center performance of these companies.

Read the original article here