The artificial intelligence industry has a resource problem, and the United Nations wants receipts.

A report published on June 3 by the United Nations University Institute for Water, Environment and Health (UNU-INWEH) lays out the staggering environmental toll of AI infrastructure. The headline number: by 2030, global data centers dedicated to AI could consume approximately 945 terawatt-hours of electricity annually. That’s nearly triple the combined total electricity usage of Pakistan, Bangladesh, and Nigeria. Put differently, the machines learning to write your emails could soon account for almost 3% of the world’s anticipated electricity consumption.

The numbers behind the ask

Electricity is just the opening act. The report projects that water consumption tied to AI could reach 9.3 trillion liters per year by 2030. To make that number feel real: that’s equivalent to the basic annual domestic water needs of about 1.3 billion people.

Then there’s the physical footprint. The land required to support AI infrastructure could exceed 14,500 square kilometers.

And the waste. AI infrastructure could produce up to 2.5 million tonnes of electronic waste annually by 2030.

The report does not single out specific AI companies. It does note that environmental impacts vary significantly depending on location and energy sources.

Six principles, one deadline

The UNU-INWEH report doesn’t just wave a red flag. It attempts to build a framework around what it calls a “responsible AI ecosystem,” organized around six guiding principles: transparency, efficiency by design, equity and environmental justice, lifecycle responsibility, global cooperation, and sustainable use.

Transparency sits at the top of that list for a reason. The core recommendation is straightforward: AI companies should fully disclose their environmental footprints. Not just carbon, but water, land use, and e-waste too. The report also urges governments to enforce standardized environmental reporting, rather than relying on voluntary corporate disclosures.

The lifecycle responsibility principle is particularly pointed. Previous research from the UN Environment Programme had already flagged that current frameworks fail to measure the complete lifecycle impacts of AI technologies. That means the environmental cost of manufacturing chips, building data centers, running inference workloads, and eventually disposing of hardware is not being captured in any consistent way.

Another key recommendation: integrating AI infrastructure into national energy, water, and land-use planning.