The UN University Institute for Water, Environment and Health (UNU-INWEH) has released a new global report that, for the first time, comprehensively assesses the environmental footprint of artificial intelligence infrastructure. Researchers examined not only carbon emissions but also water consumption and land resource utilization. Their findings indicate that current methods for evaluating the “eco-friendliness” of AI significantly underestimate its actual environmental burden.
In the opinion of the authors, the issue lies in the fact that, today, the sustainability of technologies is almost invariably measured by their carbon footprint. However, the data centers that power AI have a far broader impact, ranging from immense water consumption to the appropriation of large land areas for infrastructure and energy generation.
According to UN projections, by 2030, data center electricity consumption globally could reach 945 TWh – nearly triple the annual energy consumption of countries like Pakistan, Bangladesh, and Nigeria combined. For comparison, as early as 2025, data centers consumed approximately 448 TWh, surpassing the consumption levels of entire nations, including Saudi Arabia.
But the enormous electricity consumption is only part of the problem. The volumes of water required for cooling servers and generating electricity could, by 2030, equal the annual water consumption of 1.3 billion people in sub-Saharan African countries.
The third aspect is land. By 2030, over 14,250 km² of land could be needed for AI infrastructure, including the data centers themselves and their supply chains – this is roughly half the area of Belgium.
The report also dedicates specific attention to the so-called “green energy paradox.” Scientists note that efforts to reduce carbon footprints sometimes lead to an increase in other environmental problems. For instance, the switch from coal to biofuels might cut CO₂ emissions by about 70%, but it drastically increases water consumption by over 30 times and expands land use manifold.
As highlighted by the lead author of the study, Miriyam Achel, this situation is dangerous because it creates an illusion of environmental progress.
The report also details the already observable consequences of the rapid growth of AI infrastructure. For example, in Ireland, data centers consume over a fifth of the country’s total electricity, prompting authorities to restrict the issuance of new construction permits. In other regions, including Mexico and Uruguay, plans to expand data centers during drought periods have sparked public protests due to the risk of water scarcity for residents.
Concurrently, digital inequality is intensifying. According to UNU-INWEH, by 2025, specialized AI data centers will be concentrated in just 32 countries worldwide, with the vast majority of capacity located in the US and China. Developing nations are largely absent from this infrastructure but are often the ones bearing the environmental costs, including the rise in electronic waste, which could reach millions of tons annually by 2030.
The report’s authors issue a warning: the world has a limited window of opportunity to integrate AI development within genuine environmental frameworks. To achieve this, they believe it is insufficient to consider only the carbon footprint; when constructing new data centers, water and land consumption, as well as the impact on local ecosystems, must be mandatorily assessed.
About the Author
