UN Data Quantifies AI's Water and Land Footprint: Enough Water for 1.3 Billion People by 2030

Artificial intelligence is frequently discussed in terms of its symbolic outputs—algorithms, neural networks, and weightless cloud computing. Yet every generated image, synthesized report, and conversational prompt is produced by a vast, physical metabolism that converts material life-support systems into digital services. As the technology scales, the infrastructure required to sustain it is becoming comparable to the resource demands of entire nations.[3]

A comprehensive evidence pack published by the United Nations University Institute for Water, Environment and Health (UNU-INWEH) provides the most detailed quantification to date of this physical footprint. The central finding of the investigation is that evaluating AI's environmental cost solely through the lens of carbon emissions systematically mismeasures its true impact. Electricity used by AI systems also carries a massive water footprint and a sprawling land footprint, and these metrics do not always move in the same direction.[1][2]

The baseline data for electricity consumption illustrates the scale of the expansion. In 2025, global data centers consumed an estimated 448 terawatt-hours (TWh) of electricity, which would rank the sector as the world's 11th largest electricity consumer if it were a sovereign nation. By 2030, driven largely by the proliferation of AI workloads, that figure is projected to reach 945 TWh—nearly triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria, which collectively house over 650 million people.[1][2]

However, the most acute localized pressure comes from water consumption. The UNU-INWEH report projects that the water footprint associated with AI data centers will reach 9.3 trillion liters annually by the end of the decade. To contextualize this volume, 9.3 trillion liters is equivalent to the basic annual domestic water requirements of all 1.3 billion people living in Sub-Saharan Africa.[1][2]

This water demand stems from two primary vectors. First, power generation itself is highly water-intensive, requiring massive volumes for cooling thermal power plants and operating hydroelectric dams. Second, the data centers that house AI servers rely heavily on evaporative cooling systems to prevent high-performance processors from overheating. In warmer climates, a single hyperscale facility can consume millions of liters of fresh water daily, placing direct strain on municipal aquifers.[1][3]

The physical expansion of this infrastructure also requires vast tracts of land. The report estimates that by 2030, the land footprint necessary to support AI's energy infrastructure, supply chains, and data centers will exceed 14,500 square kilometers. This area is roughly twice the size of the Jakarta metropolitan area, a megacity home to more than 32 million residents.[1][2]

Furthermore, the rapid obsolescence of specialized AI hardware is generating a secondary crisis in electronic waste. The UN researchers project that AI infrastructure could produce up to 2.5 million metric tonnes of e-waste annually by 2030. Because only a fraction of global e-waste is formally recycled, much of this toxic material—containing heavy metals like lead and mercury—is expected to accumulate in low-income economies that lack the infrastructure to process it safely.[1][2]

A critical distinction in the data is the difference between model training and daily inference. Public debate has historically fixated on the immense energy required to train large language models. However, the evidence shows that inference—the process of generating responses to user prompts—accounts for 80% to 90% of total AI energy use once a model is deployed.[1]

The energy intensity of inference varies by orders of magnitude depending on the modality of the output. According to the UNU-INWEH data, a typical conversational chat query requires approximately 200 times the energy of a basic text classification task. Generating a single high-resolution AI image requires roughly 1,450 times that baseline, and rendering a short AI video can consume as much power as 200,000 simple text classifications.[1][2]

The evidence also highlights a complex trade-off problem in energy procurement. Transitioning data centers to low-carbon electricity sources does not automatically yield low-water or low-land operations. For example, the report notes that switching a facility's power supply from coal to bioenergy can reduce its carbon footprint by an average of 72%. However, this transition simultaneously increases the associated water footprint more than 30-fold and the land footprint 100-fold.[1][3]

The AI industry frequently points to rapid advancements in hardware efficiency and the deployment of liquid-immersion cooling as solutions to these resource constraints. Yet the UN researchers warn that efficiency gains alone are unlikely to reverse the trend due to the rebound effect, or Jevons paradox. As AI models become cheaper and more efficient to run, developers integrate them into more applications, driving a net increase in total usage that outpaces the per-query efficiency savings.[1][3]

A significant limitation in the current evidence base is the lack of localized transparency. While aggregate global projections are robust, the specific water usage effectiveness of individual, proprietary data centers remains largely opaque. Tech companies often report global averages, masking the severe strain placed on specific watersheds in arid regions where cooling demands are highest.[1][3]

Ultimately, the environmental footprint of artificial intelligence presents a profound challenge of equitable distribution. The economic and technological benefits of AI are highly concentrated, with the United States and China hosting 90% of specialized AI data center capacity. Conversely, the environmental burdens—ranging from water withdrawals and land occupation to mineral extraction and e-waste—are frequently outsourced to communities that receive a fraction of the technology's dividends.[1][2]

The UNU-INWEH report concludes that governing AI responsibly requires making its hidden metabolism visible. By integrating AI infrastructure into broader energy, water, and land-use planning, policymakers can ensure that the backbone of the digital economy develops within planetary limits, rather than shifting its costs onto vulnerable ecosystems.[1][3]