NextFin News - The environmental footprint of global data centers is expanding to a scale comparable to major sovereign nations, with electricity consumption, carbon emissions, and water usage projected to double by 2030. According to a landmark report released Wednesday by the United Nations University Institute for Water, Environment and Health (UNU-INWEH), global data centers consumed 448 trillion watt-hours of electricity in 2025, a figure that exceeds the power usage of all but ten countries worldwide.
The report, titled "Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints," marks the first comprehensive effort by a United Nations body to quantify the multi-dimensional ecological impact of the artificial intelligence boom. Kaveh Madani, a water scientist and director of UNU-INWEH, noted that the current trajectory would see data centers accounting for nearly 3% of projected global electricity use by 2030, reaching 935 trillion watt-hours. Madani, who was recently awarded the Stockholm Water Prize, has long advocated for a "lifecycle" approach to environmental resources, often challenging the tech industry's narrative of "clean" digital growth.
While the tech sector frequently highlights efficiency gains, the UNU-INWEH findings suggest that the sheer volume of AI operations is overwhelming technical improvements. Last year’s data center operations generated approximately 208 million tons of carbon dioxide—roughly equivalent to the annual emissions of Argentina. Furthermore, the energy production required to sustain these facilities consumed 1.2 trillion gallons of water. The report warns that by 2030, if data centers were a nation, they would rank as the sixth-largest power consumer on the planet.
The rapid ascent of generative AI is the primary driver of this surge. AI currently accounts for roughly 20% of data center energy demand, but that share is expected to double to 40% by the end of the decade. The intensity of these operations is stark: a single ChatGPT-style query is estimated to be 200 times more energy-intensive than a basic text classification task, such as an email spam filter. Training requirements are also escalating; while GPT-3 required 1.3 billion watt-hours to train, subsequent iterations have consumed between 50 and 70 billion watt-hours.
Miriam Aczel, a co-author of the report and environmental policy researcher at UNU-INWEH, emphasized that the environmental burden is not evenly distributed. While the benefits of AI flow globally, the physical costs—ranging from water withdrawals for cooling to land use for infrastructure and mineral extraction for chips—are often concentrated in specific local communities. Aczel’s research typically focuses on environmental justice, and she argues that the "virtual" nature of AI masks a material system with heavy physical consequences.
Industry representatives have responded with a focus on utility and ongoing innovation. Caleb Max, President of the National Artificial Intelligence Association, argued that the "energy return on investment" of AI is transformative, citing benefits in food production, poverty reduction, and safety. Similarly, Josh Levi, President of the Data Center Coalition, stated that the industry remains committed to transparency and responsible growth. These perspectives suggest a growing tension between the immediate economic and social utility of AI and the long-term ecological constraints identified by the UN.
The report also highlights a "transparency gap" that complicates independent analysis. Fengqi You, a professor of energy engineering at Cornell University who was not involved in the study, noted that many tech companies do not fully disclose the water and energy consumption of specific facilities. You observed that the UN's involvement adds significant institutional weight to the push for better data, though he cautioned that the public should remain "concerned, but not panicked."
One of the more granular findings suggests that user behavior could play a role in mitigation. The report found that reducing the length of AI queries by 30% could lower the associated energy consumption by 25%. Madani pointed out that even small adjustments, such as being more concise and omitting unnecessary politeness in prompts, could collectively save an amount of electricity equivalent to the annual usage of 700,000 people in Africa. This highlights the paradox of efficiency: as AI becomes more integrated into daily life, the total environmental cost continues to climb despite individual technological refinements.
Explore more exclusive insights at nextfin.ai.
