A new report from the United Nations University Institute for Water, Environment and Health projects that by 2030, the world’s artificial intelligence infrastructure could use enough water to meet the basic needs of 1.3 billion people for a year. This projection describes a global infrastructure trend, not the measured consumption of a single company or country. As AI expands, demand for electricity and cooling is rising, intensifying debate over the environmental footprint of the technology behind everyday digital services.
AI services run on data centers where servers operate continuously, drawing power and generating concentrated heat. The UN University report projects that data centers supporting AI could use 945 terawatt-hours of electricity annually by 2030, up from 448 TWh in 2025. It estimates an associated water footprint of 9.3 trillion liters, roughly the basic annual domestic water needs of 1.3 billion people, the population reference used for sub-Saharan Africa. That estimate includes water used directly for cooling and indirectly in electricity generation. It does not measure water consumed solely inside data-center facilities.
Cooling Systems Lead Water Demand
Water-based and evaporative cooling remove heat from servers, while power generation can add to a data center’s water footprint upstream. The balance depends on local conditions, including climate and the electricity mix, as well as how facilities cool servers and how intensively they use them. Google has said water cooling can cut a data center’s energy use by about 10% compared with air cooling. The trade-off is clear: in water-stressed areas, lower energy demand may come with greater reliance on water for cooling.
A 2025 study associated with Vrije Universiteit Amsterdam estimated that global AI systems could withdraw between 312.5 billion and 764.6 billion liters of water during the year, a modeled range roughly comparable to annual global bottled-water consumption. The study distinguishes water used at data centers from water consumed to generate their electricity. Withdrawal and consumption are different measures, and neither estimate provides a universal water-per-prompt figure. Impacts vary with the model and hardware, as well as the workload and the cooling system or power supply.
Broader electricity forecasts also point to rapid growth, but cover a wider category than AI alone. The International Energy Agency is cited as projecting data-center electricity use of roughly 485 TWh in 2025 and about 950 TWh by 2030. Those totals cover data centers as a sector, not AI workloads alone, so they should not be read as a direct measure of AI consumption.
Spain’s Data Center Debate
Aragón has become a major data-center development zone despite being one of Spain’s drier regions and experiencing repeated drought conditions. Amazon Web Services’ existing data centers there reportedly used about 68 million liters of water in 2025, equivalent to approximately 0.03% of annual industrial water demand in the Ebro basin, reported at 208 cubic hectometers. Regional reporting said actual use was 48% higher than the amount originally communicated to officials. That figure concerns existing operations and should not be treated as a measure of the company’s planned expansion.
The debate in Aragón also concerns the cumulative demands of planned facilities on water supplies and the electricity grid. AWS representative Fernando Blázquez said it was “a company policy not to publish water usage per site.” A basin-level share puts an operator’s current use in regional context, but does not settle questions about future demand or the combined impact of multiple campuses. Environmental groups are calling for closer oversight of water and energy impacts.
Transparency and Public Oversight
Major technology companies often report corporate-wide water use or data-center efficiency figures without separating AI workloads from conventional cloud services. That makes it difficult to assess the resource demands of individual services or compare facilities. The UN University and Vrije Universiteit Amsterdam figures are system-wide projections and modeled estimates, not direct readings from every AI facility. Their significance lies in the scale of the infrastructure trend and the need to account for on-site cooling and electricity generation.
Artificial intelligence’s growth will depend on technological advances and on how societies plan for its resource demands. Decisions about cooling technology and reclaimed water, along with electricity supply and local water availability, will shape the environmental impact, particularly in regions facing drought or competing demands for water.
Hidden resource consumption in everyday technology is not a new concern. Household appliances, for example, can have overlooked features that affect efficiency, as highlighted in a recent analysis of washing machine design quirks and their effect on laundry results (see this report).
