In a groundbreaking report released by the United Nations University Institute for Water, Environment and Health (UNU-INWEH), the environmental toll of artificial intelligence (AI) data centers is revealed to be far more complex and far-reaching than previously understood. By 2030, these facilities are projected to consume an astounding 945 terawatt-hours (TWh) of electricity worldwide—almost triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria, countries with a combined population exceeding 650 million. This exponential energy demand signals a new era where AI infrastructure rivals national energy footprints, compelling urgent attention to its sustainability.
Critically, the UNU-INWEH report challenges the prevailing narrative that centers solely on carbon emissions as the main environmental cost of AI. While carbon footprint remains a major issue, the report elucidates that every kilowatt-hour used to power AI systems also imposes substantial demands on water and land resources. Water is extensively consumed in electricity generation and cooling processes, whereas vast tracts of land are necessary for both the physical infrastructure of data centers and the intensive supply chains that support them. Notably, these three environmental vectors—carbon, water, and land footprints—do not necessarily correlate. Transitioning to low-carbon energy sources like biomass may reduce emissions but drastically increase water and land consumption, revealing complex trade-offs previously unaccounted for.
The global scale of this multifaceted footprint is unprecedented. Data centers standalone would constitute the world’s 11th largest electricity consumer, trailing only behind economic powerhouses such as France and outranking Saudi Arabia. In 2025, they already consumed approximately 448 TWh, nearly half of their projected use five years later. This surge is driven largely by the phenomenon known as inference: the continuous deployment phase where AI models actively process billions of real-time user queries. Contrary to the popular spotlight on the energy-intensive training phase, inference accounts for an estimated 80 to 90 percent of total AI energy consumption. For example, ChatGPT alone processes roughly 2.5 billion prompts a day, amounting to about 383 gigawatt-hours (GWh) annually—comparable to the carbon offset that would require planting 2.6 million trees cultivated over a decade.
Energy utilization per inference query varies dramatically across AI tasks. Basic conversational exchanges consume about 200 times more energy than rudimentary text classification, but more resource-intensive outputs such as generated images can require 1,450 times that baseline energy. Even more alarming, the generation of short AI videos may consume as much electricity as 200,000 spam email classifications. Key decisions—often hidden from end-users—such as model selection, prompt length, and default output resolution critically dictate the environmental footprint of AI applications, pointing to an urgent need for transparency and configurable user controls.
Paradoxically, ongoing advancements in AI efficiency may exacerbate rather than mitigate environmental impacts due to the “rebound effect” or Jevons Paradox. As inference models become more efficient and cost-effective, user demand multiplies, offsetting per-query energy savings with scaling volume. Without firm policy limits on factors like token counts, query resolution, and output durations, gains in efficiency risk being entirely negated, precipitating unchecked growth in aggregate energy consumption.
The report exposes glaring disparities in the geographical distribution of AI infrastructure, with over 90 percent of AI-specialized data centers concentrated in just two countries, the United States and China. Meanwhile, more than 150 countries lack sovereign AI compute capacity altogether. This uneven global geography not only consolidates economic and security advantages but also concentrates environmental burdens related to critical mineral extraction and the growing tide of electronic waste. Extractive activities for minerals essential to AI hardware frequently occur in jurisdictions with weak environmental governance, deepening socio-environmental injustices.
On a more localized scale, the expansion of AI infrastructure is intensifying resource stresses in vulnerable regions. For instance, data centers in Ireland consumed 21 percent of the nation’s metered electricity in 2023—surpassing residential consumption and prompting authorities to freeze new permits until 2028. In drought-stricken Queretaro, Mexico, and Uruguay’s capital Montevideo, energy and water demand for AI operations are compounding water scarcity crises, with particularly profound implications for community health and wellbeing. Compounding this, e-waste generated by AI-related infrastructure could reach 2.5 million tonnes annually by 2030, an equivalent weight to nearly 250 Eiffel Towers discarded each year, burdening waste management systems and host communities further.
Responding to these multifaceted challenges, the UNU-INWEH report proposes a comprehensive six-principle roadmap for sustainable and equitable AI development. This framework emphasizes transparency, design-driven efficiency, equity and environmental justice, lifecycle responsibility, global cooperation, and sustainable usage. It assigns tangible responsibilities across the AI ecosystem—from governments integrating infrastructure into energy and water planning, to AI developers prioritizing footprint-conscious model selection, and from operators conducting cumulative environmental impact assessments to international entities supporting harmonized metrics and capacity building in underserved regions.
This detailed environmental analysis underpins the imperative that future governance must adopt multidimensional metrics encompassing carbon, water, and land footprints jointly. Solely focusing on carbon obscures trade-offs and does not fully capture AI’s planetary impact. Effective policy must engage not only the training phase but the far more expansive inference activities—addressing default settings on AI platforms and promoting usage patterns that align with environmental limits. Moreover, community involvement and environmental justice should be central to decisions on data center siting and operational oversight, ensuring that the benefits and burdens of AI technologies are shared fairly and sustainably.
Investors and financial institutions emerge as crucial agents of change, urged to incorporate multi-resource footprint risks into due diligence concerning AI infrastructure investments. Given their leverage, they represent some of the fastest means to realign market incentives toward sustainability. Meanwhile, international cooperation is vital to curb cross-border displacement of environmental costs and promote equitable AI development, particularly aiding countries currently excluded from sovereign compute capacity to build locally relevant systems.
In sum, the era of AI demands a paradigm shift: technological advancement must be coupled with conscious stewardship that respects planetary boundaries and social equity. The digital revolution offers immense promise, but its backbone—massive energy-hungry infrastructure—requires robust oversight and measured governance to avoid unintended environmental consequences. Careful measurement, transparency, and shared responsibility across global stakeholders offer a way forward where AI can continue to enhance prosperity without undermining the Earth’s finite resources.
Subject of Research: Environmental impacts of artificial intelligence data center electricity use, including carbon, water, and land footprints.
Article Title: Environmental Cost of AI’s Energy Use: Carbon, Water and Land Footprints.
News Publication Date: 3 June 2026.
Web References:
- United Nations University Institute for Water, Environment and Health (UNU-INWEH) report: https://unu.edu/inweh/collection/environmental-cost-of-AIs-Enrgy-Use-Carbon-water-and-land-footprints
- DOI: 10.53328/INR26RMA002
References:
Aczel, M., Chamanara, S., Matin, M., Farsi, A., Marwala, T., Madani, K. (2026). Environmental Cost of AI’s Energy Use: Carbon, Water and Land Footprints. United Nations University Institute for Water, Environment and Health (UNU-INWEH), Richmond Hill, Ontario, Canada.
Keywords: AI, artificial intelligence, data centers, energy consumption, carbon footprint, water footprint, land footprint, sustainability, environmental policy, rebound effect, digital divide, electronic waste, critical minerals, environmental justice
