AI data centres could double power and water use by 2030

• AI could drive 40% of data centre power use by 2030.
• UN researchers said AI relies on data centres, grids, chips, land, and water.

Data centres are projected to use almost twice as much electricity and water by 2030, according to a report by the United Nations University Institute for Water, Environment and Health (UNU-INWEH). The report links the increase to the infrastructure required to run artificial intelligence systems.

The report said data centres used 448 terawatt-hours of electricity worldwide last year, exceeding Saudi Arabia’s total annual electricity consumption. AI accounted for about one-fifth of that total.

By 2030, annual electricity use by data centres is projected to reach 945 terawatt-hours. AI is expected to account for 40% of global data centre electricity use by then. The report said data centres are also expected to account for nearly 3% of projected global electricity use.

Water use is also expected to increase. Data centres consumed 4.5 trillion litres of water last year. The report said the amount could meet the needs of more than 600 million people in Sub-Saharan Africa.

That figure is projected to rise to 9.3 trillion litres by 2030. Water demand is linked to cooling systems and the electricity used to run data centres.

The report said large-scale water withdrawals can strain aquifers and river systems, particularly in arid or groundwater-depleted regions.

The report also estimated that data centres produced 189 million tonnes of carbon dioxide emissions last year. By 2030, emissions are expected to rise to 399 million tonnes.

AI infrastructure

Kaveh Madani, director of the institute and lead author of the report, said public debate often treats AI as software while overlooking the infrastructure behind it. “AI is also physical infrastructure,” he said, referring to data centres, electricity generation, cooling systems, transmission networks, chips, minerals, land, and water.

UNU-INWEH said the report examines water and land footprints alongside carbon emissions.

Miriam Aczel, a United Nations University environmental policy researcher and study co-author, said that about 90% of AI-related power use comes from operational requests rather than model training.

Those operational requests include prompts, searches, and media-generation tasks.

The report said a typical ChatGPT-style query is about 200 times more energy-intensive than basic text classification. It said image and video generation require more computing power, while larger models require more electricity to train.

The physical footprint of data centres is also expected to expand. The report said their land use could increase from 6,900 square kilometres in 2025 to more than 14,500 square kilometres by 2030.

The institute warned that AI deployment could place additional pressure on land, power, and water resources if governments do not account for environmental costs. It also said data centre expansion could contribute to rising electronic waste.

The report also examined life-cycle impacts from AI hardware and supporting infrastructure.

Policy and disclosure pressure

The European Union is separately preparing minimum energy-efficiency standards for data centres, along with a possible sustainability label covering water use and clean energy supply.

The report noted that AI can support efficiency improvements, including by helping optimise electricity grids and reduce waste. However, it said overall electricity and water demand is still expected to rise as additional data centre capacity is built.

Madani said efficiency gains do not always reduce total energy use if improved performance or lower costs lead to more frequent use. He said AI will not run out of water or electricity globally. However, poorly planned data centre growth could add pressure in specific locations already facing resource constraints.

The International Energy Agency has noted that data centres tend to concentrate in specific locations, creating grid integration challenges.

Investors have also pressed Amazon, Microsoftand Google for more information on water use and conservation at US data centres.

The researchers said limited company disclosure makes it difficult to assess where data centres are located, how large they are, and how much energy and water they consume. Fengqi You, a Cornell University energy engineering professor who was not part of the report, said: “We cannot manage what companies do not disclose.”

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology eventsclick here for more information.