AI Data Centers Will Consume Water for 1.3 Billion People by 2030

Artificial intelligence's environmental costs extend far beyond electricity and carbon emissions, according to new research from the United Nations University Institute for Water, Environment and Health. By 2030, AI infrastructure will consume enough water to meet the needs of 1.3 billion people, while requiring three times the electrical power used by Pakistan, Bangladesh and Nigeria combined—nations with a total population of 650 million.

The investigation, titled "Environmental Cost of AI's Energy Use: Carbon, Water, and Land Footprints," examines the physical infrastructure behind AI systems that most users perceive as purely digital. Already in 2025, global data centers consumed an estimated 488 terawatt-hours of electricity—enough to rank as the world's 11th largest electricity consumer if treated as a single nation, positioned between France and Saudi Arabia.

The hidden physics of cloud computing

"Most people understand AI as a digital technology, as a virtual thing, as something that is in the clouds," said Kaveh Madani, director of the UN institute and co-author of the report. "What we tried to do in this report was to remind people that there's some physics to all of this."

That physical reality includes a supply chain stretching from mineral extraction in developing nations through hardware manufacturing, data center construction, energy-intensive operations requiring substantial cooling water, and eventual electronic waste disposal.

Geographic inequality in costs and benefits

The report frames AI infrastructure as an environmental justice issue. Ninety percent of AI computing capacity is concentrated in the United States and China, while communities in other countries—and disadvantaged populations within those two nations—disproportionately bear environmental costs including mineral extraction impacts, electronic waste, and water shortages.

Lead author Dr. Miriam Aczel noted that renewable energy solutions don't necessarily reduce all environmental harms. "What surprised us most is how often the choices that look greenest from a carbon perspective end up worse for water or for land," she stated. Biofuels, for instance, carry water footprints 70 to 400 times larger than some fossil fuels, while hydropower dams create significant ecological impacts and reservoir evaporation.

Why it matters

As tech companies race to expand AI capabilities, the infrastructure buildout could undermine broader decarbonization goals if rising power demand prevents the retirement of fossil fuel plants. The concentration of benefits in wealthy nations while environmental costs fall on vulnerable communities mirrors historical patterns of resource extraction—what Madani termed "a new form of imperialism" where critical minerals fuel technologies that local populations cannot access.

Calls for transparency and responsibility

The researchers emphasize that AI technology itself is neutral—its impacts depend on deployment choices. They call for greater transparency in environmental reporting, government regulation of data center placement, and user awareness of AI's resource intensity. "We now have an AI consumerism that we should be worried about," Madani said, suggesting users can reduce demand through more selective AI tool usage.

The findings were first reported by Democracy Now! in an interview with Madani, who previously served as deputy head of Iran's Department of Environment and received the Stockholm Water Prize in early 2026.