Artificial Intelligence (AI) is increasingly projected as a transformative force across sectors such as health care, agriculture, governance, and education. However, the environmental consequences of developing and deploying AI systems have remained largely outside mainstream public and policy debate. As AI adoption scales rapidly, this oversight risks undermining climate and sustainability goals, particularly for countries like India that are simultaneously pursuing digital growth and environmental commitments.
Why AI’s Environmental Footprint Is Being Overlooked
Discussions on AI policy have largely focused on ethics, data privacy, and economic productivity. Environmental costs are often treated as secondary or incidental. Yet, an OECD working paper titled “Measuring the Environmental Impacts of Artificial Intelligence Compute and Applications” highlights that AI development carries significant ecological externalities, especially through energy-intensive computation and data infrastructure. According to the OECD, the global information and communication technology (ICT) sector contributes between 1.8% and 2.8% of global greenhouse gas (GHG) emissions, with some estimates placing the figure as high as 3.9%. AI systems form an increasingly large share of this footprint.
Carbon Emissions and Energy Use in AI Systems
Quantifying AI’s carbon footprint remains challenging due to limited transparency and inconsistent reporting. A 2025 sustainability report by claimed that a single text-based AI prompt consumes only 0.24 watt-hours of electricity. However, this figure has been criticised for ignoring upstream energy use such as data centre cooling, model training, and repeated inference at scale. Independent studies paint a more concerning picture. Research cited by the indicates that training a single large language model (LLM) can generate close to 300,000 kilograms of carbon emissions. Another widely cited 2019 study estimated emissions of over 626,000 pounds of carbon dioxide from training one large deep-learning model — comparable to the lifetime emissions of five cars.
Water Stress and the Full AI Life Cycle
Environmental costs are not limited to energy and carbon. In a 2024 issue note, UNEP warned that AI servers and data centres could consume between 4.2 and 6.6 billion cubic metres of water annually by 2027, primarily for cooling purposes. This raises serious concerns for water-stressed regions. The AI life cycle — spanning mineral extraction for hardware, energy use in training and deployment, and electronic waste — amplifies pressure on freshwater, land, and natural resources, yet these impacts remain poorly regulated.
Everyday AI Use and Invisible Energy Costs
Even routine AI usage contributes to cumulative environmental harm. A 2024 UNEP study noted that a single query made through consumes roughly ten times more energy than a conventional Google search. While marginal at the individual level, such differences become significant when multiplied across billions of daily interactions.
Global Policy Responses: Emerging but Fragmented
International institutions have begun acknowledging these challenges. In 2021, adopted its “Recommendation on the Ethics of Artificial Intelligence”, explicitly recognising AI’s negative impacts on society and the environment. At the national and regional level, the United States and the have moved further. Proposed legislation such as the U.S. Artificial Intelligence Environmental Impacts Act (2024) and EU-wide harmonised AI rules seek to integrate environmental accountability into AI governance.
India’s Regulatory Blind Spot
In India, policy conversations on AI and climate change remain largely focused on how AI can aid environmental protection — from climate modelling to resource optimisation — while ignoring the environmental cost of building and running AI systems themselves. India already mandates Environmental Impact Assessments (EIAs) under the EIA Notification, 2006, for infrastructure and industrial projects. Extending the scope of EIAs to include high-compute AI development and data centres could be a logical next step, especially as India positions itself as a global AI hub.
Measuring Before Regulating
A foundational requirement for regulation is reliable measurement. India could initiate a national framework to quantify AI-related emissions, energy consumption, water use, and land impact. This would require collaboration among AI developers, research institutions, civil society organisations, and environmental experts to standardise terminology, metrics, and reporting formats. Without such baseline data, policy interventions risk being either symbolic or misdirected.
AI and ESG Disclosure Norms
Another policy lever lies in disclosure. The environmental impact of AI systems could be incorporated into Environmental, Social, and Governance (ESG) reporting frameworks overseen by the Ministry of Corporate Affairs and the . The EU’s Corporate Sustainability Reporting Directive (CSRD), which mandates disclosure of emissions from data centres and high-compute activities including LLM training, offers a relevant model for India.
Making AI Compatible with Sustainability Goals
Mitigating AI’s environmental impact does not require abandoning innovation. Practical measures already exist: greater use of pre-trained models, powering data centres with renewable energy, improving hardware efficiency, and mandatory reporting of AI-specific environmental metrics. The policy challenge lies in shifting the narrative — from viewing AI only as a tool for sustainability to also recognising it as an activity that must itself be made sustainable.
What to note for Prelims?
- Share of ICT sector in global GHG emissions
- Environmental impacts of AI training and deployment
- Role of UNEP, OECD, and UNESCO in AI governance
- Energy and water use of data centres
What to note for Mains?
- Trade-offs between digitalisation and environmental sustainability
- Need for regulatory frameworks on AI’s ecological impact in India
- Role of ESG norms in governing emerging technologies
- Challenges in measuring and regulating AI-related emissions
