-
CENTRES
Progammes & Centres
Location
The US and India must work together to balance AI innovation with sustainability by focusing on energy efficiency, transparent emissions reporting, and renewable-powered AI infrastructure
Image Source: Getty
Artificial Intelligence (AI) has become deeply integrated into modern society, reshaping industries, daily life, and economic structures. At its core, AI refers to systems that mimic human cognition and decision-making. While early AI concepts date back to the mid-20th century, recent breakthroughs in machine learning, computing capabilities, and vast data generation have propelled its rapid evolution. Today, AI’s potential to impact human lives is immense: AI could deliver an additional US$13 trillion to global economic activity by 2030, which is about 16 percent higher cumulative Gross Domestic Product (GDP) compared to current levels, or approximately an additional 1.2 percent growth in GDP per year. The US is investing heavily in AI infrastructure, with a notable surge in federal spending on AI, shifting from experimental projects to large-scale investments. An example of this is the US$ 500 billion Stargate project, a collaboration between OpenAI, Oracle, and SoftBank, which exemplifies the US’s drive to be competitive and build AI infrastructure.
The US is investing heavily in AI infrastructure, with a notable surge in federal spending on AI, shifting from experimental projects to large-scale investments.
But the US is not alone. India, too, is bidding to become a leading AI nation. With the Indian government hosting the next iteration of the AI action summit and announcing a national AI computing mission to develop indigenous AI models, India is also positioning itself as the hub for AI innovation and development. As both nations scale AI for economic gains, there are certainly opportunities but also risks—one of which is environmental. Given the high energy demands of AI systems, addressing their environmental footprint remains critical.
The environmental impact of AI arises across several stages of its value chain, which includes energy-intensive hardware like Graphical Processing Units or Tensor Processing Units, cloud platforms enabling large-scale computation, and end-user interfaces which power services like chatbots and image recognition. In terms of hardware, data centres, which form the backbone of AI operations, could account for up to 21 percent of overall global energy demand by 2030 when the cost of delivering AI to customers is factored in. Unable to keep up with the power demand, countries like Ireland, the Netherlands and Singapore have put in place moratoriums for the construction of data centres while the US has extended timelines for the construction of new data centres.
The International Energy Agency (IEA) projects that energy demand from AI is only expected to grow in the near future—electricity demand from data centres will increase from 460 TWh in 2022 to 800 TWh by 2026.
Training Generative AI models is extremely energy-intensive. Training a model like OpenAI’s GPT-3 is estimated to use just under 1,300 megawatt hours (MWh) of electricity, roughly equivalent to the annual power consumption of 130 US homes. Training the more advanced GPT-4 is estimated to have used 50 times more electricity than GPT-3. Further, once deployed, AI models need to be always on, constantly consuming energy. A request to ChatGPT requires an average of 2.9 watt-hours of electricity, nearly 10 times as much energy as the average Google search. Even creating an image with generative AI uses the energy equivalent of fully charging a smartphone. The International Energy Agency (IEA) projects that energy demand from AI is only expected to grow in the near future—electricity demand from data centres will increase from 460 TWh in 2022 to 800 TWh by 2026. The demand for water used in cooling systems in data centres is a growing concern as well. As data centre equipment becomes more densely packed into smaller spaces, the need for advanced cooling technologies rises, often relying on water sources in areas that are already under significant strain. To mitigate these environmental risks, governments and the private sector must proactively work towards embedding sustainability into AI ecosystem design.
Making AI sustainable involves reducing the material impact and footprint, without compromising on innovation. To balance innovation and environmental responsibility, action is needed across the AI value chain. Several strategies can be followed:
Building data centres in regions with easy access to renewable energy sources can ease pressure on existing resources and contribute to a lower carbon footprint. According to TechUK, only 10 percent of applications are latency-sensitive and need to be located near the end user. For over 90 percent of the applications, including training LLMs and GenAI models, data centres can be strategically located near renewable energy resources. AI can also help predict and analyse customer usage patterns, enabling automatic scaling of server capacity as needed. This optimises data centre infrastructure, preventing unnecessary energy waste. By automating this process, organisations can save time and lower carbon emissions.
Enhancing hardware efficiency is also essential. Utilising energy-efficient components and performing regular maintenance can greatly reduce emissions. A key approach to improving AI hardware’s energy efficiency is designing it specifically for AI workloads. This involves utilising specialised hardware platforms that execute AI tasks more quickly and with lower power consumption than general-purpose processors. Examples include ASICs (application-specific integrated circuits)—custom-designed chips optimised for specific AI applications like image recognition or natural language processing and FPGAs (Field-Programmable Gate Array)—reprogrammable hardware platforms that adjust to specific AI functions. FPGAs are often more energy-efficient than general-purpose processors, as they dynamically allocate only the necessary resources for a given task.
Assessing and reporting the environmental impact of AI systems enables organisations to grasp their lifecycle emissions and mitigate negative externalities.
Developing efficient AI models is just as crucial. Compact, domain-specific models designed for specific applications can produce identical results while using less processing power, thereby easing the strain on infrastructure and resources. Methods that enhance model efficiency for deployment on devices or systems with resource constraints, such as quantisation, distillation, and client-side caching, improve AI performance while promoting sustainability.
Above all, transparency is key to advancing sustainability initiatives. Assessing and reporting the environmental impact of AI systems enables organisations to grasp their lifecycle emissions and mitigate negative externalities. Developing standardised frameworks for monitoring and comparing emissions across the industry will ensure consistency and accountability. For example, unveiled at this year's AI Action Summit in Paris, the AI Energy Score—developed by Salesforce, Hugging Face, and Carnegie Mellon University—seeks to improve transparency regarding the environmental impact of AI models. Much like how ENERGY STAR revolutionised energy efficiency standards for appliances and electronics, this initiative provides a reliable benchmark for AI model sustainability. However, greater awareness and acceptance are still needed to refine and standardise the parameters for measuring energy consumption globally.
Among global initiatives aimed at strengthening collaboration between industries and governments to reduce AI-related emissions, the EU AI Act stands out for its strong emphasis on sustainable AI. The Act not only addresses the sustainability of AI itself but also promotes the use of AI for broader environmental goals. Notably, Article 40 mandates the development of harmonised standards to enhance AI systems' resource efficiency, ensuring energy-conscious operation throughout their lifecycle. Additionally, the Act encourages collaboration among AI developers, environmental experts, and social scientists to mitigate AI’s environmental impact.
In contrast, the United States and India lack similarly binding commitments to sustainable AI governance. Both the US and India are focused on leveraging AI for economic progress, yet there remains ample scope for collaborative initiatives that, while less binding than the EU AI Act, could still drive meaningful progress toward sustainable AI.
A key starting point for U.S.-India cooperation could be enhancing transparency in AI emissions reporting. By developing a common framework for measuring and disclosing carbon emissions and energy consumption across the AI value chain, both countries can lay the groundwork for policies that encourage sustainability. Establishing standardised Key Performance Indicators (KPIs) to assess the environmental impact of AI model training and long-term operational emissions could help drive data-driven policymaking. Additionally, launching a green AI certification scheme with evidence-based sustainability targets or incorporating emissions disclosures into voluntary commitments would encourage companies and consumers to make more environmentally conscious decisions.
By developing a common framework for measuring and disclosing carbon emissions and energy consumption across the AI value chain, both countries can lay the groundwork for policies that encourage sustainability.
Beyond bilateral efforts, multilateral initiatives like the Coalition for Sustainable AI announced at the Paris AI Action Summit and led by France, the UN Environment Programme (UNEP), and the International Telecommunication Union (ITU), hold significant potential. This coalition aims to standardise AI’s environmental impact assessments, incentivise energy-efficient AI development, and align AI solutions with global sustainability goals. While India has expressed support for this initiative, there is immense potential for the US to benefit in playing an active part in the coalition, particularly because it is a collaborative approach between governments, academia, civil society, and the private sector to green AI and use AI for greening the economy.
The revitalisation of iCET (United States–India Initiative on Critical and Emerging Technology) as TRUST (the Technology and Resilient US-India Strategic Trade Initiative) presents an opportunity to broaden bilateral cooperation across the entire AI value chain, including sustainability. During Prime Minister Narendra Modi’s visit to the U.S., he and President Donald Trump pledged to collaborate with private industry on a “U.S.-India Roadmap on Accelerating AI Infrastructure” by the end of 2025. This presents an opening for both nations to build AI infrastructure powered by renewable energy, including potential collaboration on nuclear energy and Green AI research. The two governments could also collaborate on specific frameworks that incentivise private companies to adopt sustainable practices, such as tax breaks for renewable-powered data centres or subsidies for developing energy-efficient hardware, emphasising how sustainable AI practices can lead to cost savings, operational efficiencies, and competitive advantages for businesses.
For these efforts to succeed, political leaders must create incentives for cleaner AI solutions while accounting for the carbon costs of AI development. Governments must strike a balance—removing barriers to private sector innovation while simultaneously encouraging greater investment in clean energy solutions. By taking these steps, the US and India can ensure AI growth aligns with global sustainability objectives while maintaining economic and technological competitiveness.
Sustainability must be embedded into the foundation of the AI ecosystem to support its long-term growth and resilience. Collaboration among governments, businesses, and innovators is essential to advancing AI while minimising its environmental impact. By integrating environmentally friendly practices with technological progress, we can unlock AI’s full potential without endangering the planet. As India develops its AI governance framework, prioritising sustainable AI is not just a policy imperative but also a strategic advantage—offering industries cost efficiencies, improved operations, and a stronger position in the global AI landscape.
Urmi Tat, Manager, Public Policy and Government Affairs, Salesforce, India
The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.
PREV
.png)