The United Nations University Institute for Water, Environment and Health (UNU-INWEH) released a landmark report on its 30th anniversary that fundamentally reframes how we need to think about artificial intelligence's environmental impact. Titled "Environmental Cost of Artificial Intelligence: Carbon, Water and Land Footprints," the study—authored by Aczel, Chamanara, Matin, Farsi, Marwala, and Madani (doi: 10.53328/INR26RMA002)—moves beyond the carbon-only lens that has dominated AI sustainability discussions. The central finding should alarm anyone paying attention to where this industry is heading: every kilowatt-hour used by AI carries carbon, water, AND land implications, and these footprints don't always move in the same direction.

Beyond Carbon Myopia

The report's most important contribution is exposing a dangerous oversimplification that has allowed AI companies to claim green credentials based on renewable energy alone. Low-carbon electricity is not automatically low-water or low-land. Hydroelectric power, for instance, generates minimal carbon but requires massive water storage and land acquisition. Nuclear energy produces negligible carbon emissions while consuming significant water for cooling. This interconnected reality means the industry can't simply buy renewable energy certificates and call it sustainable—AI's environmental cost depends on WHERE that electricity is generated and WHICH specific sources power it.

The Material Reality of AI

While Silicon Valley loves to position AI as pure software—a few electrons here, some algorithms there—the UNU-INWEH report strips away that mythology. AI is a material system with measurable environmental costs baked into its entire lifecycle: data centers requiring land, advanced chips demanding rare earth minerals extracted from specific regions, cooling systems drawing water from local aquifers, and electricity grids stressed by unprecedented demand spikes. The benefits of these systems flow across borders and sectors globally, but the environmental burdens—e-waste, water withdrawals, land use changes, mineral extraction impacts—concentrate in specific communities and regions often far removed from where the technology is consumed.

What Your Queries Actually Cost

The report details how everyday usage patterns shape AI's footprint in ways most users never consider. Model choice matters—a smaller, efficient model versus a large frontier system can have orders of magnitude difference in energy consumption. Output length directly correlates with resource use; longer generations require more computation. Modality shifts the calculus significantly—generating an image or video consumes far more resources than text, and real-time voice interactions add their own layer of demand. The growing trend toward AI-generated content across platforms multiplies these individual impacts into substantial aggregate environmental burdens that most organizations aren't even measuring.

Governance as a Justice Issue

Perhaps most critically, the report frames AI's environmental footprint not merely as a technical problem requiring better efficiency metrics, but as a governance and justice challenge demanding structural responses. The authors call for a responsible AI ecosystem grounded in six pillars: transparency about actual resource consumption, efficiency by design rather than afterthought optimization, equity and environmental justice protections for affected communities, lifecycle responsibility extending beyond deployment to manufacturing and disposal, global cooperation on standards, and sustainable use practices that acknowledge real-world constraints.

Key Takeaways

  • Low-carbon energy does NOT equal low-water or low-land impact; the three footprints must be evaluated together
  • AI infrastructure concentrates environmental burdens in specific geographic regions while benefits distribute globally
  • Individual usage choices—model selection, output length, modality—have measurable aggregate impacts at scale
  • The industry cannot solve this with renewable energy certificates alone; systemic governance reform is required

The Bottom Line

The UNU-INWEH report makes clear that AI's environmental cost isn't a PR problem for the tech industry to manage—it's a justice issue requiring regulatory teeth and transparent reporting requirements. Until governments mandate disclosure of water withdrawal data, land use impacts, and full lifecycle analyses alongside carbon metrics, the public remains in the dark while communities near data centers bear costs they'll never see compensated. The era of AI companies self-reporting their green credentials ends now—or it should.