A new United Nations University report shows that data centers and AI collectively form an environmental footprint on par with nation-scale energy and water use, with a substantial growth trajectory through 2030.

Training and deployment of AI accounted for about a fifth of data-center load last year and could rise to nearly half by 2040, potentially driving AI electricity use to around 374 terawatt-hours—over nine times Nigeria’s current annual consumption.

Beyond electricity, AI’s footprint extends to increased chip production, e-waste, and water use, with large-model training consuming significant freshwater and inference consuming substantial water per request; demand also ties to minerals like cobalt, lithium, and rare earth elements.

Strategic governance measures offer the biggest potential to cut impact, such as choosing smaller models when feasible, aligning models to specific tasks, avoiding duplicate computation, favoring text over image/video, and encouraging concise outputs to save energy.

Individuals can shrink personal footprints by using conventional tools when possible and opting for simple behavioral changes, like questioning whether an AI-assisted search is truly needed.

Authors warn that cutting carbon emissions alone doesn’t automatically reduce water or land impacts, calling for proactive monitoring and mitigation before problems escalate.

Recommendations urge integrating AI infrastructure decisions with energy, water, and land-use planning; designing models with environmental impacts in mind; preferring lighter models; prioritizing greener energy procurement; and treating footprints as material investment risks with community involvement.

They also urge making footprints visible, regulators normalizing disclosures, and guiding users toward less energy-intensive tools.

Concrete policy steps include involving communities in data-center permitting, prioritizing energy- and resource-efficient models, avoiding high-default resource use like KI-enabled search, and mandating industry-wide environmental impact reporting.

Case studies show Brazil’s hydro grid has low carbon but high water and land footprints, the UK grid has notable land impact, and the Jevons paradox warns efficiency gains may raise total footprint without caps.

Experts warn that rapid expansion could collide with local resource pressures, underscoring the need for proactive, sustainable planning before infrastructure is locked in.

Even with efficiency gains, rebound effects may boost overall energy use, and company transparency about energy consumption remains limited.