UNU-INWEH: AI Infrastructure Footprints Are Water and Land—Not Only Carbon

Policy and industry debates about AI infrastructure still default to carbon and megawatts. UNU-INWEH’s report forces water and land into the same frame as emissions, with concrete 2030-scale numbers that communities, utilities, and hyperscalers can argue over.

UNU-INWEH released a research report quantifying the carbon, water, and land footprints of the electricity used to train and run AI. Beyond greenhouse gases, researchers estimate that AI-driven data-center power use implies water demand on the order of the basic needs of 1.3 billion people by 2030 and a land footprint large enough to stress already constrained regions—especially when “green” carbon choices shift burdens onto water or land.

What the report measures

The study, Environmental Cost of AI’s Energy Use: Carbon, Water and Land Footprints (Aczel, Chamanara, Matin, Farsi, Marwala, Madani; doi: 10.53328/INR26RMA002), examines footprints tied to AI electricity use across major data-center hubs rather than treating carbon as the sole metric.

Key projected findings cited in UNU materials and contemporaneous coverage (TIME, UN News):

• Water: AI-related water use by 2030 could equal the basic annual domestic needs of ~1.3 billion people (on the order of Sub-Saharan Africa’s population scale in UN framing), covering both cooling and water embedded in power generation.
• Land: Power-generation and supply-chain land footprint may exceed 14,500 km² (5,590 sq mi)—roughly twice the Jakarta metropolitan area.
• Energy/emissions context: Rapid growth in data-center electricity demand is linked to AI expansion, with multi-footprint trade-offs across the world’s largest hubs.

Lead author Miriam Aczel emphasized that choices that look greenest on carbon can worsen water or land outcomes—underscoring the need for multi-metric siting and power planning.

Limits

• 2030 figures are model projections under stated assumptions—not measured 2026 consumption.
• Footprints are tied to electricity for AI workloads across hubs; facility-level cooling vs generation water splits vary by region.
• “Sustainable AI” marketing claims should be checked against multi-metric methods, not carbon alone.

Sources

• UNU-INWEH report collection: “Environmental Cost of Artificial Intelligence: Carbon, Water, and Land Footprints” (unu.edu/inweh, published June 3, 2026). Primary report page and citation.
• UNU-INWEH news release: “Rising Emissions, Depleting Water and Vanishing Land” (June 3, 2026).
• TIME: “AI Could Use as Much Water as 1.3 Billion People by 2030” (June 3, 2026).
• UN News summary of the UNU study (June 4, 2026).