MUMBAI: AI may be getting smarter, but the planet could still be picking up a bigger bill. A new report from the United Nations University Institute for Water, Environment and Health has warned that the artificial intelligence boom could dramatically increase global demand for electricity, water and land over the next five years, even as AI systems become more efficient.
At the heart of the warning is a centuries-old economic idea with a surprisingly modern twist. Researchers pointed to the “Jevons paradox”, a theory developed by economist William Stanley Jevons in the 19th century, which suggests that greater efficiency often leads to higher overall consumption rather than lower usage. In simple terms: when something becomes cheaper and more effective, people tend to use more of it.
The report argues that AI could follow the same path. As models become faster, cheaper and more powerful, adoption is expected to spread across industries, potentially driving resource consumption to unprecedented levels.
According to the study, AI-related activity could account for nearly 3 per cent of global electricity consumption by 2030. Annual electricity demand from data centres is projected to more than double, rising from 448 terawatt-hours last year to 945 terawatt-hours by the end of the decade. AI alone is expected to account for around 40 per cent of that demand, up from roughly one-fifth last year.
To put the scale into perspective, data centres consumed more electricity in 2025 than Saudi Arabia, the world’s 11th-largest electricity consumer. By 2030, their energy appetite could approach Japan’s current annual electricity demand.
The pressure extends far beyond power grids. Researchers estimate water consumption by data centres could surge from 4.5 trillion litres to 9.3 trillion litres by 2030 as cooling requirements intensify. Carbon emissions linked to data centre operations are projected to climb from 189 million tonnes to 399 million tonnes over the same period.
Land use is also expected to expand rapidly. The report forecasts that the footprint of data centres worldwide will more than double, growing from approximately 6,900 square kilometres today to more than 14,500 square kilometres by the end of the decade.
Kaveh Madani, Director of the United Nations University Institute for Water, Environment and Health and lead author of the report, said discussions around AI often overlook its physical reality.
“AI is not just software,” Madani noted, pointing to the vast network of data centres, power generation facilities, cooling systems, transmission infrastructure, semiconductor manufacturing, mineral extraction and land resources required to support the technology.
The report also highlighted growing concerns around global inequality. While only 32 countries currently host AI-focused cloud infrastructure, around 90 per cent of that capacity is concentrated in the United States and China. Meanwhile, countries supplying critical minerals and managing electronic waste often bear a disproportionate share of the environmental costs.
To address the challenge, researchers called for greater transparency from technology companies around energy use, water consumption and emissions. The report also urged governments to factor AI-driven demand into energy and climate planning and adopt a full lifecycle approach covering everything from mineral extraction to recycling and disposal.
The message is a timely reminder that efficiency alone may not be enough. As AI continues its rapid march into everyday life, the real challenge may not be building smarter machines, but ensuring their growing appetite does not outpace the planet’s resources.
