AI’s environmental costs threaten water, land and climate

Data centres, the global infrastructure powering AI, could consume 945 terawatt-hours of electricity annually by 2030—nearly triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria, countries collectively home to more than 650 million people. However, this is just the tip of the iceberg. Every unit of electricity used carries a “water footprint” for cooling and energy production, and a “land footprint” associated with power generation and supply chains.

Rethinking sustainability

According to a new study from the UN University (UNU), AI-related water consumption could equal the basic annual domestic needs of 1.3 billion people by the end of the decade. At the same time, its land footprint may exceed 14,500 square kilometres—roughly twice the size of the Jakarta metropolitan area.

The report highlights a critical gap: greenhouse gas emissions linked to training models are often prioritised, overlooking other costs. Solutions seen as “green” in one sense may worsen pressures in others, particularly where resources are scarce. Switching to certain renewables may reduce carbon emissions, but can significantly increase water consumption and land use.

Daily use is the main culprit

Public debate has centred on the energy to train models, but the study finds that day-to-day usage accounts for 80 to 90 per cent of total energy demand. One widely used AI service is estimated to process around 2.5 billion prompts per day, consuming hundreds of gigawatt-hours of electricity each year. Generating a single AI image can require more than a thousand times the energy of simple text classification, while video generation demands even greater resources.

Efficiency improvements alone are unlikely to offset rising demand due to the “rebound effect,” in which lower costs and improved performance drive higher usage, ultimately increasing total resource consumption.

Local burdens, global benefits

The environmental impacts of AI infrastructure are not evenly distributed. Costs are often concentrated in specific regions. In some countries, data centres account for a significant share of national electricity consumption, while in others, facilities are drawing heavily on water supplies, sometimes amid drought conditions. Additionally, AI infrastructure is projected to generate up to 2.5 million tonnes of e-waste annually by 2030, a burden likely to fall on lower-income countries with limited capacity for safe disposal. The production of critical minerals also raises concerns about environmental degradation and social inequities in extraction regions.

A widening digital and environmental divide

Expansion is creating new disparities. Over 90 per cent of AI-specialised computing capacity is concentrated in two countries: the United States and China. More than 150 nations lack significant domestic AI infrastructure. This imbalance limits economic opportunities and raises questions of environmental justice, as some countries bear the environmental costs without sharing in the benefits of AI-driven growth.

Towards responsible AI

UNU researchers stress this is not an argument against AI, but a call for urgent action to ensure the technology develops within planetary limits. The study outlines a framework for a “responsible AI ecosystem,” including transparency, efficiency by design, equity, lifecycle responsibility, global cooperation, and sustainable use. Governments are urged to integrate AI infrastructure into resource planning, while companies should design systems to minimise consumption. Ultimately, the future of AI will depend on governance choices made today.