AI data centers by 2030 to rival 650M people's power use, 1.3B people's water needs

On 3 June 2026, the United Nations University Institute for Water, Environment and Health (UNU-INWEH) released a landmark report, Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints, revealing that the environmental toll of artificial intelligence is far broader and more severe than carbon-focused assessments have acknowledged. The investigation, led by Professor Kaveh Madani (Director of UNU-INWEH), projects that by 2030 global data centers powering AI will consume 945 terawatt-hours of electricity annually—nearly triple the combined electricity use of Pakistan, Bangladesh, and Nigeria, countries collectively home to more than 650 million people. That same electricity will carry a water footprint equal to the basic annual domestic water needs of all 1.3 billion people in Sub-Saharan Africa and a land footprint exceeding 14,500 square kilometers, roughly twice the Jakarta metropolitan area, home to more than 32 million people.

The report fundamentally challenges how the industry measures sustainability. Researchers note that carbon emissions have been the dominant metric, but every kilowatt-hour of electricity used carries three footprints: carbon (from power generation), water (from cooling and power plants), and land (from energy infrastructure and supply chains). These footprints do not correlate. Switching from coal to bioenergy, for example, can slash the carbon footprint by an average of 70 percent while increasing the water footprint more than thirty-fold and the land footprint a hundred-fold. As Dr. Miriam Aczel, lead author and UNU-INWEH researcher, explains: "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, often in places that didn't ask for it." The report concludes that evaluating AI sustainability through a single metric can hide trade-offs and shift environmental burdens onto regions already facing water or land stress.

A critical reframing in the report concerns where energy consumption actually occurs. Public discussion has focused heavily on training large models—GPT-3 required an estimated 1.3 gigawatt-hours of electricity, while GPT-4 consumed between 50 and 70 GWh. However, once deployed, inference—the continuous running of models to answer everyday user prompts—dominates energy use, accounting for 80 to 90 percent of total AI energy consumption. ChatGPT alone is estimated to process around 2.5 billion prompts per day, translating to roughly 383 GWh of electricity per year for a single product. Offsetting the associated carbon emissions would require 2.6 million tree seedlings grown for 10 years, enough to cover a land area the size of Manhattan. The water footprint is equivalent to the minimum annual domestic water needs of roughly 500,000 people in Sub-Saharan Africa, and the land footprint equals over 800 football fields. Energy demand varies by orders of magnitude depending on the task: a typical conversational chat query is around 200 times more energy-intensive than basic text classification, while generating a single AI image can require around 1,450 times the baseline, and a single short AI-generated video can consume as much electricity as 200,000 spam classifications.

The report documents stark geographic inequality in both AI benefits and environmental burdens. Only 32 countries in the world host AI-specialized data centers, and 90% of that capacity is concentrated in 2 countries (the United States and China), while more than 150 countries have little or no access to sovereign AI compute. Ireland provides a concrete warning: in 2023, data centers accounted for 21% of total metered electricity, exceeding all urban households combined; the national grid operator has paused new approvals around Dublin until 2028 due to infrastructure strain. In Querétaro, Mexico, expanding compute infrastructure is drawing on water supplies amid prolonged droughts. In Uruguay, plans for a water-intensive data center coincided with a 2023 drought that depleted Montevideo's freshwater reserves, making tap water unsafe to drink. Meanwhile, AI infrastructure could generate 2.5 million tonnes of electronic waste annually by 2030, much of it processed in low-income economies with limited environmental safeguards, while critical minerals are extracted in jurisdictions with weak environmental oversight. Communities hosting data centers often do not benefit from the AI services running there, creating what Dr. Mir Matin, Manager of UNU-INWEH's Geospatial, Climate and Infrastructure Analytics Programme and a report co-author, calls an asymmetry that repeats historical patterns: "some places carry the costs and other places capture the benefits."

Professor Madani, who was recently named the 2026 Stockholm Water Prize Laureate, emphasizes that the problem will worsen without intervention due to the rebound effect (the Jevons Paradox): as AI models become more efficient and cheaper, they are used more frequently. Without explicit limits on tokens, resolution, or default output length, efficiency gains at the per-query level are easily erased by volume growth. The report calls for a responsible AI ecosystem built on six principles—transparency, efficiency by design, equity and environmental justice, lifecycle responsibility, global cooperation, and sustainable use—with recommendations directed at governments (integrating AI infrastructure into energy planning, water governance, and land-use permitting, and requiring standardized environmental footprint reporting), industry and AI developers (treating model selection, default outputs, and routing decisions as footprint determinants), users and deploying organizations (adopting fit-for-purpose use), data center operators (treating siting and energy procurement as footprint decisions), investors (treating electricity, carbon, water, and land footprints as material risks), communities and civil society (being involved early in data center siting decisions), and international institutions (supporting harmonized measurement standards and building compute capacity in excluded regions). Professor Tshilidzi Marwala, Rector of the United Nations University, framed the issue as a 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. Whether it does so equitably is now a governance question, not a technical one."