AI's Hidden Environmental Toll: Carbon, Water, and Land Use

AI's Physical Infrastructure Demands Real-World Resources

Artificial intelligence may seem like a purely digital technology, but a new report from the United Nations University Institute for Water, Environment and Health reveals it depends on extensive physical infrastructure with measurable environmental costs. The analysis examines data centers, advanced chips, cooling systems, electricity grids, water resources, land, and critical mineral supply chains that power AI systems at scale.

According to the report first published by UNU-INWEH, AI's environmental impact extends well beyond carbon emissions to include water withdrawals and land use—footprints that don't always move in the same direction. Low-carbon electricity sources, for instance, may still carry high water or land requirements.

Why it matters

As AI becomes embedded in economies, public services, and daily life, understanding its full environmental cost is essential for sustainable deployment. The report frames this as a governance and justice issue: AI's benefits often cross borders while environmental burdens—from data center siting to mineral extraction—concentrate in specific communities and regions.

Location and Energy Source Drive Environmental Impact

The report's central finding challenges assumptions about AI efficiency. Environmental costs depend not just on how much electricity AI systems consume, but critically on where that electricity is generated and which energy sources power it. Every kilowatt-hour used to train, deploy, and operate AI models carries distinct carbon, water, and land implications based on the regional energy mix.

This geographic dimension means identical AI workloads can have vastly different environmental footprints depending on data center location. The analysis shows that major infrastructure trends, including rapid data center growth, combine with everyday use patterns—model choice, output length, modality, and the expanding use of text, image, and video generation—to shape AI's total impact.

A Framework for Responsible AI Development

The UNU-INWEH report calls for a responsible AI ecosystem grounded in six principles: transparency, efficiency by design, equity and environmental justice, lifecycle responsibility, global cooperation, and sustainable use. By making AI's carbon, water, and land footprints visible and comparable, the framework provides a practical basis for integrating AI considerations into energy, climate, water, and land-use planning.

The report emphasizes that addressing AI's environmental footprint requires moving beyond technical solutions alone. Without governance mechanisms that account for where costs fall, innovation risks shifting environmental burdens onto vulnerable communities already facing resource constraints.

Measuring What Matters

By quantifying environmental impacts across multiple dimensions, the report enables more informed decision-making about AI deployment. Organizations can compare the full resource implications of different models, deployment locations, and use cases rather than focusing solely on energy consumption or carbon emissions.

The analysis was published by the United Nations University Institute for Water, Environment and Health in its report "Environmental Cost of Artificial Intelligence: Carbon, Water and Land Footprints," marking the institute's 30th anniversary.