As artificial intelligence (AI) data centers (DC) emerge as significant consumers of national-level electricity, a new crisis of water scarcity threatens global AI infrastructure. A warning about the environmental costs of AI extending beyond carbon to include water and land has been raised in an official United Nations (UN) report.
According to the June 27 report from the United Nations University Institute for Water, Environment and Health (UNU-INWEH), annual water consumption by AI data centers is projected to reach 9.3 trillion liters by 2030.
This amount is sufficient to meet the basic annual water needs of 1.3 billion people in sub-Saharan Africa. The physical footprint of these data centers is estimated to exceed 14,500 square kilometers, roughly double the area of the Jakarta metropolitan region, due to the substantial water used for evaporative cooling to maintain server temperatures.
Kaveh Madani, director of UNU-INWEH, stated, "The perception of AI as merely software is outdated. AI requires physical infrastructure, including data centers, power facilities, cooling systems, transmission networks, semiconductors, land, and water."
The report also highlights a paradox where the transition to renewable energy for carbon reduction could exacerbate water and land issues. Lead author Dr. Miriam Akhzel noted, "We were most surprised to find that the most environmentally friendly choices from a carbon perspective could lead to worse outcomes in terms of water or land."
Geographic concentration is a significant concern. As of 2025, only 32 out of 196 countries (16%) have specialized AI data centers, with 90% of their capacity concentrated in the United States and China. Both countries operate large inland data center clusters, leading to water resource pressures in specific regions. In Querétaro, Mexico, the construction of data centers amid prolonged drought has sparked controversy over local water depletion, while in Ireland, data center power consumption has surged to 21% of total metered electricity, prompting a halt on new approvals near Dublin until 2028.
This situation presents new opportunities for major Asian hubs such as South Korea, Singapore, and Taiwan. Industry analysts suggest that these regions, with their island and peninsular geography facilitating seawater cooling and concentrated submarine cable infrastructure, are emerging as structural alternatives to the U.S.-China-dominated AI infrastructure.
South Korea is promoting the development of an AI data center cluster centered around Yongin and Anseong, and it possesses several key submarine cable hubs in East Asia. Singapore has already established itself as an Asian data center hub, although it faces constraints related to power and land. Taiwan, as a critical hub for AI semiconductor production, is seeing an increasing demand for data centers linked to its supply chain.
However, the advantage of water resources is not absolute. South Korea also grapples with issues of power grid overload due to concentration in the metropolitan area and structural limitations in renewable energy ratios.
Meanwhile, power supply issues remain a barrier to the expansion of AI data centers. Last year, global data center power consumption reached 448 terawatt-hours (TWh), ranking it 11th in the world, surpassing Saudi Arabia and following France. This figure is expected to rise to 945 TWh by 2030, with AI workloads projected to account for 40% of total data center power consumption.
Kaveh Madani emphasized, "There is little time left to ensure that the foundation of the AI technological revolution develops within global limits."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.