UN Report Warns of AI’s Rapidly Growing Water and Land Footprint

UNU-INWEH finds global data centres could consume electricity equal to nearly 1 million terawatt-hours by 2030, with water and land impacts often obscured by a narrow focus on carbon emissions

A new report from the United Nations University Institute for Water, Environment and Health (UNU-INWEH) warns that the rapid expansion of artificial intelligence infrastructure carries a substantial and often overlooked water and land footprint, one that risks being obscured when AI’s environmental cost is assessed through carbon emissions alone.

AI’s Expanding Environmental Footprint

The rapid global expansion of artificial intelligence over the past several years has been accompanied by a corresponding surge in the construction and operation of data centres, the physical infrastructure underpinning AI training and deployment. Much of the public and policy conversation around this expansion has focused on energy consumption and associated carbon emissions, given the substantial electricity demands of large-scale computing operations.

However, electricity and carbon emissions represent only part of the environmental picture. Data centres also require significant water resources, primarily for cooling systems that prevent computing hardware from overheating, as well as substantial land area for construction and supporting infrastructure. As AI adoption accelerates and data centre construction expands into new regions worldwide, questions about these additional environmental dimensions — water and land use specifically — have begun to draw closer scrutiny from researchers and policymakers.

It is this gap in understanding that the new UNU-INWEH report seeks to address, presenting a more comprehensive accounting of AI’s environmental footprint that extends beyond carbon emissions alone.

Projected Scale of Electricity Demand

According to the report, electricity consumption by global data centres could reach 945 terawatt-hours by 2030 — a figure that illustrates the sheer scale of energy infrastructure required to sustain continued AI expansion. This projected demand underscores why data centre electricity use has become a significant consideration in broader energy and climate planning discussions worldwide.

Translating Electricity Demand into Water and Land Impacts

The report’s central contribution lies in translating this electricity figure into concrete water and land footprint estimates, offering a clearer sense of the physical resources required to support projected AI growth.

According to the report’s findings, the water footprint associated with this level of data centre electricity consumption would be equivalent to the basic annual domestic water needs of 1.3 billion people in Sub-Saharan Africa — a comparison intended to illustrate the scale of water resources involved by placing them alongside a concrete measure of human water need.

The associated land footprint, meanwhile, is projected to exceed 14,500 square kilometers, an area described in the report as roughly twice the size of the Jakarta metropolitan area. This land footprint reflects not only the physical space occupied by data centre facilities themselves, but also the broader infrastructure required to support their operation.

The Problem with Carbon-Only Assessments

A key finding of the report is that assessing AI’s environmental cost through carbon emissions alone can obscure important trade-offs between different environmental impacts. The report illustrates this point through the example of switching electricity generation from coal to bioenergy: while such a switch does reduce carbon emissions, it can simultaneously increase water use by more than thirtyfold.

This finding highlights a broader methodological concern raised by the report — that policy and industry decisions optimized primarily around carbon reduction may inadvertently worsen other environmental pressures, particularly water stress, if those trade-offs are not explicitly considered and disclosed.

Real-World Examples of Water Strain

The report points to specific real-world cases illustrating how AI’s water demands can strain local communities already facing water stress. In Querétaro, Mexico, and Montevideo, Uruguay, expanding data centre operations have drawn on local water supplies during periods of drought, according to the report’s findings. These cases serve as concrete illustrations of the tension that can arise when water-intensive digital infrastructure is sited in regions where water resources are already under pressure from climatic or other factors.

Recommendations for Policy and Industry

Based on its findings, UNU-INWEH calls for joint disclosure of carbon, water, and land footprints associated with data centre operations, rather than continued reliance on carbon-focused reporting alone. The report argues that this more comprehensive form of disclosure would allow policymakers, communities, and industry stakeholders to make more fully informed decisions about AI infrastructure development.

The report further recommends that siting and resource decisions for new data centre projects explicitly account for local water and land stress conditions, rather than treating these factors as secondary to considerations such as electricity availability or connectivity infrastructure. This recommendation reflects a broader push within the report toward integrating environmental risk assessment more directly into the planning process for AI infrastructure, particularly in regions where water resources are already limited or contested.

Broader Implications

As AI adoption continues to accelerate globally, the report’s findings suggest that water and land considerations are likely to become an increasingly important part of the broader conversation about AI’s environmental sustainability — one that has, to date, been dominated largely by discussions of energy use and carbon emissions. For regions already grappling with water scarcity or drought conditions, the expansion of data centre infrastructure may present a growing point of tension between technological development and local resource availability, underscoring the report’s call for more integrated, transparent environmental assessment going forward.