New research has laid bare the colossal environmental footprint of the artificial intelligence boom, suggesting the technology's data centres could produce more carbon dioxide this year than entire small nations or the city of New York.
The Staggering Scale of AI's Carbon Footprint
In a study published on Monday in the peer-reviewed journal Patterns, data scientist Alex de Vries-Gao estimated the carbon emissions from electricity used to power AI systems globally. His calculations put the figure for 2025 at between 33 million and 80 million metric tons of CO2.
To put that into perspective, the higher estimate would surpass the total 2024 emissions of countries like Chile and Czechia (both 78m tons), Romania (71m tons), and even the entire metropolis of New York City, which emitted 48 million tons of greenhouse gases last year.
De Vries-Gao's analysis builds on his prior work, which estimated AI's global electricity demand could reach 23 gigawatts by the end of 2025, a figure approaching the total power consumption of a nation the size of the United Kingdom. He combined this with International Energy Agency data on data centre emissions to arrive at the alarming range.
A Thirst Problem: AI's Massive Water Consumption
Perhaps even more contentious is the study's assessment of AI's water use. De Vries-Gao calculated the sector's total water footprint for 2025 to be somewhere between 312 billion and 767 billion litres.
This staggering range, if accurate, means the water consumed to cool AI servers and generate their power could exceed the 46 billion litres of bottled water humanity drinks annually. The researcher cautioned, however, that estimating water use is "even more difficult to assess" than carbon emissions due to a severe lack of public data and corporate secrecy.
The issue has already sparked local protests. In Newton County, Georgia, the arrival of a Meta data centre has been linked to rising water prices, damaged wells, and a projected water deficit by 2030. Similarly, a report warned that planned data centres could increase annual water stress in Phoenix, Arizona, by 32%.
Corporate Secrecy and a Call for Transparency
A central theme of the report is the "significant uncertainty" surrounding these figures, directly attributed to the non-disclosure practices of major AI firms like Microsoft, Google, and OpenAI. This lack of transparency extends to water usage, where companies and local governments have sometimes fought to keep details secret.
"Without transparent data, the biggest opportunities for mitigating the climate impacts of data centres and AI cannot be easily identified," de Vries-Gao wrote. He emphasised that further disclosures from operators are essential to improve the accuracy of environmental assessments.
The study notes that environmental impact varies wildly between facilities, depending on location, cooling methods, and the local energy grid's carbon intensity. Some experts argue the public concern is overstated, pointing out that the estimated water footprint largely consists of "indirect" use by power plants, which often return water to the local system.
Nevertheless, the soaring demand for high-end computation has turned data centre resource consumption into a hot-button issue. Recently, 230 environmental groups urged the US Congress to impose an immediate national moratorium on new facilities, highlighting the growing political and ecological tension surrounding the infrastructure of the AI age.
