AI’s 2025 Energy Surge: Water and Power Use Soars

The environmental footprint of artificial intelligence is expanding at a startling pace, with new research indicating its energy and water consumption has reached staggering levels. A recent study suggests the global AI industry this year will generate carbon emissions comparable to a major metropolis and consume freshwater resources on a scale that rivals global bottled water use. This analysis, while likely conservative, underscores a critical and growing challenge as AI becomes embedded in daily life.

The research, led by Alex de Vries-Gao, a PhD candidate at the VU Amsterdam Institute for Environmental Studies, points to a significant lack of corporate transparency. Tech companies often publish broad sustainability figures but rarely disclose the specific resources consumed by their AI operations. This data gap makes it difficult to grasp the full environmental toll. De Vries-Gao’s workaround involved piecing together estimates from hardware production forecasts, analyst reports, and public corporate statements to build a clearer picture.

His calculations estimate global AI power demand could reach 23 gigawatts this year, already exceeding the electricity used for Bitcoin mining. From this energy use, he projects annual carbon dioxide emissions between 32.6 and 79.7 million tons. To put that in perspective, New York City emits approximately 50 million tons annually. The carbon pollution from AI operations is now on par with that of a major global city, highlighting its substantial climate impact.

Water usage presents another major concern. Data centers require vast amounts of water for cooling systems to prevent servers from overheating. Furthermore, the power plants that supply electricity to these facilities are also immense water consumers, using it for cooling and steam generation. De Vries-Gao’s study projects that AI could use between 312.5 and 764.6 billion liters of water globally this year. This estimate surpasses prior research and signals a faster-than-anticipated escalation in resource demand.

Shaolei Ren, an associate professor at the University of California, Riverside who authored an earlier study on the topic, called the new findings “timely.” He noted that de Vries-Gao’s analysis is still “really conservative” because it focuses solely on the operational phase of AI. It excludes the substantial environmental costs accumulated during manufacturing, supply chain logistics, and electronic waste disposal at a device’s end of life.

The surge in building new data centers to support generative AI is also driving plans for more power plants, which compounds both water use and greenhouse gas emissions, especially if those plants rely on fossil fuels. This trend is fueling local opposition across the United States, where community pushback against new data center projects is frequently rooted in anxieties over strained water supplies and power grids.

The wide range in the study’s estimates, from emissions to water consumption, directly results from insufficient corporate disclosure. Sustainability reports often omit crucial details like indirect water use linked to electricity demand or the specific portion of resources dedicated to AI workloads. Environmental impacts can vary dramatically based on a data center’s location and the carbon intensity of the local power grid, making transparency about operational sites vital for accurate assessment.

“We really need to have that transparency, so we can start having that discussion,” de Vries-Gao argues. The current trajectory forces a societal question about priorities and equity. As AI’s capabilities grow, so too does its resource appetite, prompting a necessary conversation about whether this is a sustainable and fair path forward without greater accountability from the industry driving this innovation.

(Source: The Verge)