The rapid rise of artificial intelligence (AI) is creating a hidden environmental burden that could rival the resource needs of entire nations, according to a new report by the United Nations University Institute for Water, Environment and Health (UNU-INWEH). The report,
'Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints', warns that by 2030 the world's data centres powering AI systems could consume 945TWh (terawatt hours) of electricity annually and require9.3trn (trillion) litres of water every year — enough to meet the minimum domestic water needs of all 1.3bn (billion) people living in Sub-Saharan Africa.
Researchers say the environmental consequences of AI are being significantly underestimated because most discussions focus solely on carbon emissions while overlooking the equally important impacts on water resources, land use and electronic waste.
"AI is not just software," the report states. "It is physical infrastructure built on data centres, advanced chips, cooling systems, electricity grids, water resources, land and critical mineral supply chains."
AI's Electricity Demand Approaching Country-scale Levels
According to the report, global data centres consumed an estimated 448TWh of electricity in 2025. If data centres were treated as a country, they would rank as the world's 11th-largest electricity consumer. By 2030, electricity demand is projected to more than double to 945TWh, equivalent to almost 3% of projected global electricity consumption.
The report notes that this projected consumption is nearly three times the combined annual electricity use of Pakistan, Bangladesh and Nigeria, countries with a combined population exceeding 650mn (million) people.
Researchers estimate that AI workloads accounted for about 20% of total data-centre electricity consumption in 2025 and could reach 40% by 2030. Under that scenario, AI alone would consume about 378 TWh of electricity annually — more than nine times Nigeria's electricity consumption.
Professor Kaveh Madani, director of UNU-INWEH and leader of the investigation team, said the study was not intended as an argument against AI but rather a call for responsible development.
"This report is not a case against artificial intelligence, a technological transformation that is improving the lives of billions of people around the world. It is a call for using it responsibly and addressing its unintended impacts proactively to make it sustainable and equitable," he said.
Water Footprint Equivalent to the Needs of 1.3bn People
Perhaps the most striking finding relates to water consumption.
The report estimates that the water footprint associated with projected 2030 data-centre electricity use will reach 9.3tn litres annually. Researchers calculated that this volume is equivalent to the minimum annual domestic water requirements of the entire population of Sub-Saharan Africa.
The water demand stems not only from cooling data-centre servers but also from water used throughout electricity generation systems.
The study argues that policymakers frequently overlook this dimension of AI infrastructure because environmental assessments typically focus on greenhouse gas emissions.
"Every kilowatt-hour of electricity used to train or run an AI system also carries a water footprint, from cooling and power generation, and a land footprint, from energy infrastructure and supply chains," the report notes.
Researchers warn that efforts to reduce carbon emissions alone may inadvertently worsen water stress. The report highlights that switching from coal to bioenergy can reduce carbon emissions by about 70% while increasing water consumption more than 30-fold and land use more than 100-fold.
Lead author of the report, Dr Miriam Aczel said the findings challenge conventional assumptions about green energy transitions.
"What surprised us most is how often the choices that look greenest from a carbon perspective end up worse for water or for land. If we keep judging AI sustainability by carbon alone, we might think that renewables make AI infrastructure clean, but that is solving one problem while creating other problems."
Land Footprint Larger Than Many Metropolitan Regions
The report also quantifies AI's land footprint, another environmental dimension rarely discussed in public debates.
Researchers estimate that the land required to support projected 2030 electricity consumption by data centres will exceed 14,500sqkm (square kilometres). According to the report, this is roughly twice the size of the Jakarta metropolitan area and nearly ten times larger than Mexico City.
The study argues that electricity generation, transmission infrastructure, renewable energy installations and associated supply chains all contribute to land-use pressures that are frequently omitted from sustainability discussions.
ChatGPT and Everyday AI Use Driving Most Energy Demand
The report challenges another widely held assumption — that training AI models is the primary environmental concern.
While training advanced systems remains extremely energy-intensive, researchers found that inference, the ongoing process of responding to user prompts after deployment, now accounts for between 80% and 90% of total AI energy consumption.
ChatGPT alone is estimated to process about 2.5bn prompts every day.
At a conservative estimate of 0.42 watt-hours per prompt, this translates into approximately 383Gwh (gigawatt hours) of electricity annually for a single AI product.
Researchers estimate that supporting ChatGPT's current usage requires enough water to meet the annual domestic needs of roughly 500,000 people in Sub-Saharan Africa.
AI-generated Videos Emerging as Major Environmental Concern
The study identifies AI-generated images and videos as a rapidly growing source of energy demand.
According to the report, a typical AI-generated image consumes around 2.9 watt-hours of electricity, making it about 1,450 times more energy-intensive than basic text classification tasks.
Video generation is even more resource-intensive.
Researchers estimate that a short AI-generated video can consume as much electricity as 200,000 spam-classification operations. A high-complexity AI video can require more than 415 watt-hours of energy per clip. (Page 9)
The report calculates that the electricity-associated water footprint of a single AI-generated image is about 29 millilitres, while a complex AI-generated video requires about 4.1 litres of water — nearly equivalent to two days of drinking-water needs for one person.
Local Communities Bearing Environmental Costs
The report argues that AI's benefits and environmental burdens are distributed unequally across the world.
Several case studies illustrate how local communities often absorb the resource costs of infrastructure serving users elsewhere.
In Ireland, data centres accounted for 21% of total metered electricity consumption in 2023, exceeding electricity use by all urban households. Concerns over grid capacity have led authorities to pause new approvals in the Dublin area until 2028.
In Mexico's Querétaro region, expanding data centre infrastructure is drawing on water supplies amid prolonged drought.
In Uruguay, plans for a major data centre coincided with a severe drought that depleted freshwater reserves around Montevideo.
Dr Mir Matin, co-author of the report, said the pattern reflects a growing environmental justice issue.
"If you map where data centres are getting built against where water stress is worst, you tend to see the same regions in some instances. And the communities living near these sites are not necessarily the ones using the AI being run there."
AI Infrastructure Concentrated in Two Countries
Researchers also warn that AI infrastructure is becoming increasingly concentrated.
Only 32 countries currently host AI-specialised data centres, while more than 150 countries have little or no sovereign AI computing capacity.
The report estimates that over 90% of AI-specialised cloud computing capacity is concentrated in just two countries — US and China.
Professor Tshilidzi Marwala, rector of the United Nations University and UN under-secretary-general, described the imbalance as both a technological and governance challenge.
"The concentrated development of AI infrastructure in the privileged areas of the world is creating a large digital divide that poses profound challenges in the equitable development of AI."
E-waste and Mineral Extraction Add to Environmental Burden
Beyond electricity and water consumption, the report warns that AI's physical infrastructure creates substantial environmental impacts across its entire lifecycle.
The production of advanced AI hardware relies on critical minerals such as lithium, cobalt, gallium and rare earth elements, often extracted in regions with weaker environmental protections.
By 2030, AI infrastructure could generate as much as 2.5mtpa (million tonnes per annum) of electronic waste, equivalent to discarding nearly 250 Eiffel Towers every year. (Pages 10 and 13)
Researchers argue that many of these environmental costs are borne by low-income countries that receive few of the economic benefits associated with AI development.
Call for Global Governance and Accountability
The report concludes that AI's environmental footprint should be viewed as a governance challenge rather than merely a technical issue.
UNU-INWEH recommends a six-principle framework for a 'responsible AI ecosystem' centred on transparency, efficiency by design, equity and environmental justice, lifecycle responsibility, global cooperation and sustainable use. The framework is presented in.
Among its recommendations, the report calls on governments to incorporate AI infrastructure into energy planning, water governance and land-use regulation. It also urges companies to disclose standardised information about carbon, water and land footprints associated with AI systems.
The central message, researchers say, is not to halt AI innovation but to ensure it develops within environmental limits.
As AI becomes embedded in everything from healthcare and finance to education and communication, the report warns that the world can no longer treat artificial intelligence as a purely digital technology.
Instead, policy-makers must recognise that every AI prompt, image and video is supported by a vast physical infrastructure consuming electricity, water, land and minerals on a scale increasingly comparable with that of nations.
The question, the report suggests, is no longer whether AI will transform society, but whether that transformation can occur without transferring environmental costs to communities that derive little benefit from the technology itself.
