By Naveed Qazi
As one of the most influential technologies of our era, artificial intelligence is reshaping the sectors of public administration, financial system, scientific research, and everyday life. Behind its presence lies a deeply physical infrastructure, made up of data centres, semiconductor factories, fibre networks, and global mineral supply chains. The technology relies on an infrastructure dependent on electricity, water, land, and materials at a scale which is not widely apparent in public debate. However, what is conclusive now through international evidence is that AI is not environmentally neutral, even if its global footprint remains partially measurable.
The major starting point for the analysis on this issue would be the measured data on global digital ecosystems. The International Energy Agency reports that global energy consumption has been rising steadily, around 1 to 1.5 percent, and that electricity demand for data centres could double within the next few years, largely driven by high-performance artificial intelligence applications. Electricity use is just one part of the whole picture. There are also chances that AI could induce high water stress in certain regions, as cooling high-density systems require large volumes of water. Research from University of California, Riverside, particularly the work of Shaolei Ren and colleagues, postulate that artificial intelligence contributes to both direct and indirect water use.
Thirsty Algorithms: The Rising Demand for Global Energy and Water
Ren’s team found that standard conversational chatbots can consume roughly 500 millilitres of water for every 10 to 50 prompts. Training a single large model like GPT-3 can directly evaporate 700,000 litres of freshwater. According to United Nations University, Institute of Water, Environment and Health, the water footprint associated with training GPT-4 was about 600 million litres, enough to meet the minimum annual domestic water needs of 81,000 people in Sub-Saharan Africa, or to fill 237 Olympic-sized pools. These facts were also covered by journalist Olivia Tempest in Smart Water Magazine.
The material footprint of AI goes into hardware production and electronic use. Modern AI systems depend on a high degree of advanced chips, servers, and networking equipment that require energy-intensive manufacturing processes, and rare minerals such as lithium, cobalt, and rare earth minerals. The supply chains governing the hardware systems create environmental pressures during extraction, particularly in regions with weakened environmental regulation. When the hardware becomes non-operational, they are replaced at relatively fast cycles, adding to global electronic waste.
From Rare Minerals to E-Waste: The Material Burden of AI Hardware
The evolution of AI could contribute to increasing demand for high-performance hardware systems. The landmark study done in Nature Communications / Earth and Environment revealed the material burden of AI, showing that training a single large language model (LLM) requires between 1760 and 8,800 graphics processing units. The findings highlight that incremental model performance gains come at disproportionately high material costs, underscoring the need to incorporate material resource considerations into discussions of artificial intelligence scalability and sustainability. Similarly, IEEE Spectrum reflected on a scenario where LLMs will generate 1.2 million tonnes of e-waste between 2023 and 2030.
According to a United Nations University report, the benefits and burdens of the massive global expansion of AI are highly unequal. Several site-level cases show how globally distributed AI services create intense local pressures. In Ireland, data centres accounted for 21% of total metered electricity in 2023, exceeding all urban households. Similarly, in Queretaro, Mexico, expanding compute infrastructure is drawing on water supplies amid prolonged droughts. In Uruguay, plans for a water-intensive data centre coincided with a 2023 drought that depleted Montevideo’s freshwater reserves, making tap water unsafe to drink.
A Digital Divide: Environmental Justice and Global Disparity in AI Access
The report also concludes two important revelations: Firstly, ‘low-carbon’ is not automatically ‘low-water’ or ‘low-land’, and warns that evaluating AI sustainability through a single metric can hide trade-offs and shift environmental burdens onto regions already facing water or land stress. Secondly, only 32 countries in the world host AI-specialised data centres, and 90% of that capacity is concentrated in just two countries: the United States and China. Meanwhile, more than 150 countries currently have little or no access to sovereign AI compute. This imbalance not only limits economic opportunities but shows dominance of stronger, wealthier nations over weaker nations, raising questions of environmental justice. It also reflects a tussle between the Global North and Global South.
Hence, the analysis frames that the problem with the global footprint of AI is not just as an economic divide but also as an environmental justice issue where excluded countries bear critical minerals extraction and e-waste burdens while the strategic benefits flow elsewhere. That is why academicians such as Professor Tshilidzi Marwala want global systems of AI governance to be built sustainably and fairly. For a start, a practical solution would be the creation of new institutions, clearer rules, and sustainable and equal standards. One step would be mandatory disclosure of water and energy use associated with AI, as much of this information is currently confidential.
The Road to Accountability: Global Governance and the Future of Sustainable AI
There is also a need for lifecycle-based accounting for AI hardware, tracking environmental impact from mineral extraction through manufacturing, operations, and disposal. These evaluations would bring AI governance closer to the standards already used in heavy industry and energy sectors. Currently, global powers like the United States and China are tackling AI governance through a fragmented mix of targeted regulations, but more clarity is likely needed. At the same time, two laws in the European Union are already in deployment or near deployment. The EU AI Act (Regulation 2024/1689) mandates that providers report known or estimated model energy consumption. The second initiative, the EU AI Act Targeted Consultation on Energy Metrics, will build a standardised framework for measuring AI energy efficiency across the training and operational stages.
UNESCO’s recommendation on the ethics of artificial intelligence also explicitly establishes environmental and ecosystem flourishing as one of its central pillars. This framework addresses the complete lifecycle of AI technologies, positioning ecological sustainability as a non-negotiable component of responsible tech development. Hence, international governance efforts are pivotal to help build out measurement, standards, and reporting tools in an equitable way so that energy policy and investment strategies for managing hyperscale expansion and sustainability goals are addressed.
Naveed Qazi is a Contributing Editor at Science Matters and a published author. Hailing from Srinagar, Kashmir, with a postgraduate degree in International Business from the University of Hertfordshire, United Kingdom, he writes on science, technology, business, global affairs, and has contributed to all leading English-language newspapers in Kashmir. His professional experience spans hospital marketing, banking, construction operations and IT content development across India and the United Arab Emirates.