UN Warns AI Data Centers Will Use as Much Water as 1.3 Billion People Need by 2030

The United Nations University has published a report that for the first time brings together artificial intelligence's carbon, water and land footprints in a single place. The conclusions are stark: by 2030, AI infrastructure will consume resources on a scale comparable to the needs of entire continents, with the costs falling disproportionately on the regions that benefit least from the technology's growth.

The report was produced by the United Nations University Institute for Water, Environment and Health (UNU-INWEH). The authors note that previous assessments of AI's environmental impact focused almost exclusively on greenhouse gas emissions, overlooking the water consumption and land use of data centers and the power infrastructure that supports them. According to the researchers, this narrow accounting systematically understates the technology's true cost.

Daily use, not training

A key finding of the report challenges the popular belief that training large language models is AI's biggest environmental burden. According to UNU-INWEH, everyday use of already-deployed systems, known as inference, accounts for 80 to 90 percent of AI's total energy demand. That means the environmental cost scales with the number of users and queries, not just with the size and frequency of training new models.

The report's authors give a concrete example: ChatGPT alone handles around 2.5 billion queries a day, translating into roughly 383 gigawatt-hours of electricity consumption per year. The water footprint tied to its operation matches the minimum annual household needs of about 500,000 people in Sub-Saharan Africa.

Who bears the costs

The report highlights a stark geographic imbalance. More than 90 percent of the specialized computing power used to train and run AI models is concentrated in the United States and China. Meanwhile, more than 150 countries lack computing infrastructure of that class, yet they often bear the costs tied to mining the raw materials used in hardware production and disposing of the e-waste generated along the supply chain.

The researchers also stress that low-emission electricity does not automatically mean low water or land use. A data center powered entirely by renewable energy can still consume enormous amounts of water to cool its servers, while solar and wind farms occupy vast stretches of land that traditional carbon footprint accounting fails to capture.

Guterres calls for transparency

UN Secretary-General Antonio Guterres used the report's release to announce a new initiative aimed at the largest AI companies. He proposed an AI Environmental Transparency Initiative calling on tech firms to measure and publicly disclose the full environmental impact of their systems, covering carbon, water and land footprints together rather than in isolation.

AI is also thirsty for land, water and energy - Antonio Guterres, UN Secretary-General

No more hidden costs. No more shifting the burden onto those least able to bear it. It's time to come clean - Antonio Guterres, UN Secretary-General

The initiative, announced during Climate Action Week in London, calls on companies to power every data center with renewable energy by 2030 and to adopt stricter environmental standards limiting emissions and resource use. For now it remains an appeal rather than a binding legal commitment, so its effectiveness depends on voluntary cooperation from companies that have so far rarely published complete data on their data centers' water consumption.

What it means for Poland

For Poland, the report has practical relevance amid ongoing efforts to build a European AI gigafactory and a growing number of investments in domestic data centers. The more computing capacity that lands on the Vistula, the more pressing questions become about the water used to cool server rooms and the strain on local power grids, especially in regions where water access was already limited.

The UNU-INWEH report does not address Poland's situation directly, but it fits into a broader wave of criticism over the pace of data center construction worldwide, visible for instance in New York authorities' decision to pause construction of large AI data centers for a year, or in reports of bacterial contamination in Cheyenne's water supply during the construction of a Meta data center. The report's authors argue that without transparent data, it is difficult to hold a rational public debate about the location and scale of future investments.