What if I told you that the carbon footprint of artificial intelligence (AI) data centres could rival that of New York City or even a small European nation by 2025? A recent study reveals this startling fact, highlighting that these facilities may consume as much water as is used annually by the global bottled water industry. This research estimates that AI systems operating in data centres could emit between 32.6 million and 79.7 million tonnes of carbon dioxide by 2025.
To put that into perspective, the report cites New York City’s total emissions at approximately 52.2 million tonnes of CO2 for 2023, making the projected emissions from AI remarkably significant. At the lower end of the estimate, AI’s carbon output could be similar to Norway’s total emissions, which were around 31.5 million tonnes for the same year, according to data from the European Environment Agency.
Data centres are expansive facilities that house servers running various online services, including cloud computing, video streaming, and AI applications. These servers generate substantial heat, necessitating water-based cooling systems to maintain safe operational temperatures. As advancements in AI and related technologies continue to accelerate, the demand for energy-intensive data centres—coupled with their need for cooling water—has surged.
The study also sheds light on AI's enormous water consumption, estimating it could range from 312.5 billion to 764.6 billion litres in 2025, which mirrors the global annual bottled water consumption. This water consumption includes both direct usage for cooling purposes and indirect use associated with electricity generation. Notably, indirect water use can be up to four times greater than direct consumption, yet many tech companies fail to disclose this critical information.
When we look at Europe, we find that it houses about 15% of the world’s data centres, coming in second to the United States, which dominates with around 45%. One major advantage Europe has is its cleaner electricity generation. The carbon intensity of European power grids stands at roughly 174 grams of CO₂ per kilowatt-hour, significantly less than the global average of 445 gCO₂/kWh and much lower than the U.S. figure of 321 gCO₂/kWh. Therefore, data centres operating in Europe have a considerably lower carbon footprint per unit of electricity they consume.
However, the study highlights a troubling lack of transparency among major tech companies. An examination of environmental reports from nine leading firms revealed a consistent absence of AI-specific environmental metrics. Despite acknowledging that AI significantly drives their increased energy consumption, none of the companies provided detailed disclosures regarding AI’s specific environmental impact.
The research utilized a top-down approach by combining public sustainability reports from major tech firms—such as Amazon, Apple, Google, and Meta—with estimates of AI electricity demand and grid intensity factors to assess AI’s environmental footprint. Yet, the author cautioned that there is considerable uncertainty surrounding these figures, as companies seldom differentiate between AI-related and non-AI computing activities.
"Additional disclosures from data centre operators are urgently needed to refine these estimates and responsibly manage the escalating environmental effects of AI systems," stated de Vries-Gao, the study's author. Other companies involved in this analysis include Baidu, Microsoft, Oracle, Tesla, and Tencent.
While none of these companies report AI-specific metrics, some like Google, Meta, and Microsoft noted significant increases in electricity consumption during 2023 and 2024, attributing this surge to their AI operations. "As the environmental impact of data centres grows at a rapid pace, so does the urgency for transparency within the tech industry," he added.
The research advocates for new policies that require the disclosure of additional environmental metrics. These should include the exact locations where AI systems operate, the scale of operations at each facility, and water usage effectiveness (WUE) values for individual sites.
Euronews Next reached out to the tech companies for comments but did not receive any responses before the publication of this report.
