Contents
Introduction
As India accelerates AI adoption under the IndiaAI Mission, OECD and UNEP estimates show AI-driven ICT emissions nearing 3% of global GHGs, making environmental sustainability integral to technological leadership.
Environmental Footprint of India’s Rapid AI Proliferation
- Energy-Intensive Compute Economy: AI systems, especially Large Language Models (LLMs), are compute-hungry infrastructures. Studies estimate that training a single advanced AI model can emit 300,000–600,000 kg of CO₂, comparable to the lifetime emissions of multiple automobiles. In India, where over 70% of electricity generation remains coal-based, this intensifies the carbon lock-in risk, potentially undermining India’s Net Zero 2070 pledge announced at COP26.
- Water Stress and Thermal Externalities: AI depends on data centres requiring continuous cooling. UNEP (2024) projects global AI servers could consume 4.2–6.6 bcm of water by 2027. In India, hyperscale data centres clustered in Chennai, Noida, and Hyderabad draw water from already stressed aquifers, aggravating hydro-social stress and raising equity concerns, as recognised by the National Green Tribunal in multiple infrastructure cases.
- E-Waste and Resource Extraction: Rapid obsolescence of GPUs, TPUs, and AI accelerators contributes to India’s 1.8 million tonnes of annual e-waste (Global E-waste Monitor). The extraction of rare earths and lithium for AI hardware creates embedded environmental costs, often externalised to mining regions, contradicting principles of intergenerational equity.
Need for a Standardized Environmental Impact Measurement Framework
- What Gets Measured Gets Managed: Currently, AI’s environmental costs remain opaque and underreported. India lacks a statutory mechanism to quantify AI-related energy, carbon, and water footprints, leading to policy blind spots.
- Expanding the EIA Paradigm: Just as the EIA Notification, 2006 governs physical infrastructure, its scope can be extended to high-compute digital infrastructure, including AI model training and deployment. Metrics such as Power Usage Effectiveness (PUE), Carbon Usage Effectiveness (CUE), and Water Usage Effectiveness (WUE) should become mandatory disclosures.
- Global Best Practices: The EU’s Corporate Sustainability Reporting Directive (CSRD) and the U.S. Artificial Intelligence Environmental Impacts Act, 2024 illustrate how disclosure-driven governance aligns innovation with sustainability. India can localise these frameworks through SEBI’s ESG norms and MCA reporting standards.
Adopting ‘Green AI’: From Compute Maximalism to Efficiency
- Green AI vs Red AI: Traditional Red AI prioritises marginal accuracy gains regardless of energy costs. In contrast, Green AI emphasises algorithmic efficiency, frugality, and lifecycle sustainability.
- Technological Pathways:
- Model Optimisation: Techniques such as pruning, quantisation, and knowledge distillation drastically reduce compute needs.
- Pre-trained and Shared Models: Avoiding redundant training lowers cumulative emissions.
- Renewable-Powered Data Centres: Mandating green PPAs aligns AI growth with Panchamrit commitments.
- Edge AI: Decentralised computation reduces data transfer energy and latency.
- AI for Sustainability: Paradoxically, AI itself can enable climate action—optimising smart grids, precision agriculture, and disaster prediction—provided its own footprint is governed.
Way Forward:
- Aligning AI Sovereignty with Planetary Boundaries
- Institutional and Policy Integration.
- Introduce AI-specific environmental audits under the Energy Conservation Act.
- Create Energy-Star–like eco-labels for AI models.
- Incentivise Frugal AI research through targeted grants and tax credits.
Foster multi-stakeholder standard-setting involving industry, academia, and civil society.
Conclusion
Echoing Justice B.N. Kirpal’s environmental jurisprudence and UNEP’s lifecycle approach, India must ensure AI progress respects planetary limits, proving technological sovereignty and ecological stewardship can coexist.
