Environmental Cost of Artificial Intelligence: Carbon, Water, and Land Footprints

This UNU-INWEH research report looks beyond carbon-only assessments to measure the environmental impact of AI's rapid growth across three dimensions: carbon, water and land. It shows that AI relies on substantial physical infrastructure (data centres, advanced chips, cooling systems and global supply chains) and should be understood as a material system with real, measurable environmental costs. **Key findings:** - **Location matters:** AI's footprint depends not only on how much electricity is used, but on where that electricity is generated. - **Trade-offs across dimensions:** Low-carbon electricity does not automatically reduce water or land use. - **Environmental justice:** Burdens are concentrated in specific communities, while the benefits of AI are distributed globally. - **Usage choices count:** Infrastructure trends and everyday decisions, such as model selection, output length and modality, shape AI's overall footprint.