The use of Artificial Intelligence (AI) takes up much discussion in sectors that range from agriculture to health care. According to an Organisation for Economic Co-operation and Development (OECD) working paper, “Measuring the Environmental Impacts of Artificial Intelligence Compute and Applications”, the development of AI algorithms comes with certain environmental costs such as an increased carbon footprint which exacerbates climate change-related challenges. The OECD is an international body of 38-member countries, primarily democracies with market economies, working to promote economic growth, prosperity, and sustainable development through data, analysis, and policy standards. It serves as a forum for governments to compare experiences, find solutions to common challenges, and set high standards for economic policy, covering areas like finance, AI, health, and climate action.
The report says that the global ICT industry (including hardware such as televisions) is estimated to be responsible for 1.8%-2.8% of global greenhouse gas (GHG) emissions; other calculations have a figure as high as 2.1%-3.9%.
It is worth noting that data about the carbon footprint of AI models and their use are not always authentic. A Google report published in August 2025, claims that a single text AI prompt consumes electricity of only 0.24 watt-hours. What may seem to be low levels of electricity consumption has also drawn criticism for the report’s incomplete and misleading conclusions.
In September 2024, an issue note by the United Nations Environment Programme (UNEP), on the environmental impact of the full AI life cycle, said that house AI servers may utilise 4.2 billion cubic meters (bcm) to 6.6 bcm of water in 2027, leading to water scarcity. The note also refers to a study which indicates that training a single Large Language Model (LLM) can generate almost 3,00,000 kilograms of carbon emissions.
Similarly, a June 2019 study, “Energy and Policy Considerations for Deep Learning in NLP”, has assessed that the training of a single large AI model entails the emission of over 6,26,000 pounds of carbon dioxide. This is equivalent to emissions caused by five cars in their lifespan. A look at the energy consumption in the use of ChatGPT, a widely used AI virtual Assistant, helps understand how AI algorithms impact the environment and contribute towards climate change. According to a study, “Navigating New Horizons” (UNEP, July 15, 2024), any request made through ChatGPT leads to energy consumption that is 10 times more than what it would be through a Google search.
In 2021, UNESCO released its “Recommendation on the Ethics of Artificial Intelligence”, which emphasised recognising the “negative impacts of AI on societies, environment”. These non-binding recommendations were adopted by around 190 countries. The United States and the European Union (EU) appear to be the prominent jurisdictions that have proposed legislation dealing with the environmental impact of AI, as the Artificial Intelligence Environmental Impacts Act of 2024 and The European Union’s resolution on harmonized AI rules.
Since there are global conversations about the carbon cost of AI use and deployment, India also needs to recognise the environmental costs of developing AI models. Current discussions now on AI and climate change are on how AI can help protect the environment, but without going into the demerits of developing large AI algorithms.
The first step to address this challenge is to carry out an exercise to measure the environmental impacts of developing and deploying AI models. In India, an Environmental Impact Assessment (EIA) study is mandatory as in the EIA Notification, 2006. While an EIA is often conducted to evaluate projects concerning the environment such as river water projects, its scope can be extended to include assessing the impact of development of AI algorithms on the environment.
The government could also focus on the establishment of measuring standards in order to assess AI’s impact. It can be done by involving stakeholders such as tech companies that are developing large-scale AI algorithms, think tanks, and non-governmental organisations that are working toward reducing carbon footprints and mitigating other environment-related challenges. This will help in building consensus on terminology, standards, and consistent indicators and reporting requirements, ultimately leading to informed policy decisions.
Another essential exercise is data collection which can be done by deploying sustainability metrics such as Greenhouse Gas emissions (GHG), energy, water and natural resources consumption that are utilised by AI algorithms. Environmental costs beyond energy, such as the impacts on freshwater and land use, could also be evaluated.
The government can also explore the possibility of making the environmental impact of developing and deploying AI models as part of environmental, social and governance (ESG) disclosure standards by the Ministry of Corporate Affairs and Security and the Securities and Exchange Board of India. It can take inspiration from the EU, as its Corporate Sustainability Reporting Directive (CSRD) framework requires the disclosure of emissions data from data centres and high compute activities, which includes training of LLMs.
The focus should shift to including AI as a part of solutions toward global sustainability goals. There are several AI sustainable practices that can be adopted to mitigate the adverse impact of AI on the environment such as deploying pre-trained models, using renewable resources to power data centres, and reporting AI-specific estimates. Hopefully, growing use of renewable sources of energy in India will prove to be beneficial, if the adverse impact of AI on the environment is audited regularly and improvised suitably.
