India’s rapid AI adoption risks undermining climate and resource goals without stronger governance

India’s push to scale artificial intelligence across health care, agriculture and industry is reshaping productivity, but it also carries a growing and under‑recognised environmental bill that threatens climate and resource goals unless checked now. The lead analysis by ForumIAS warns that rapid AI adoption expands energy demand, water use, electronic waste and raw‑material extraction, while related studies and policy reviews make clear these impacts are large, poorly measured and likely to rise without targeted governance.

Industry uptake in India is already high, with ForumIAS noting that roughly 70% of firms are running AI projects and more than 90% plan to use their own data to train models, well above global averages. That scale matters because the computational intensity of modern AI is concentrated: training a single large language model can produce carbon emissions in the order of hundreds of tonnes, and inference workloads can be orders of magnitude more energy‑hungry than ordinary web searches. According to a UNECE analysis, training one large model can generate about 300,000 kg of CO2; other sector studies report single‑model training runs consuming well over 1,000 megawatt‑hours of electricity. Those magnitudes convert rapid digital adoption into a material source of greenhouse gas emissions if electricity is not clean.

The infrastructure that supports AI, data centres, specialised accelerators and networking, drives much of the footprint. Global data‑centre energy use was reported at 415 TWh in 2024 and, by some projections, could more than double by 2030 if demand trends continue. Data‑centre cooling and hardware manufacture also create heavy water and materials burdens; the GreenAI white paper and United Nations Environment Programme estimates place server water consumption in the billions of cubic metres within a few years, while the manufacture of computing equipment consumes large quantities of raw materials and relies on rare earths mined with environmental costs. The Quint and other Indian reporting note that a large share of data centres are sited in water‑stressed states, amplifying local water‑security risks.

Operational characteristics of AI compound the policy challenge. High‑compute training involves volatile, spiky power demands that can strain grid flexibility and push operators toward diesel backup or fossil‑rich baseloads, worsening local air pollution and emissions. Model obsolescence further reduces the climate efficiency of prior investment: frequent re‑training and successive, larger models make much prior energy use effectively redundant. At the same time, hardware lifecycles for AI accelerators and edge devices are shortening, accelerating e‑waste flows; the GreenAI paper and UNFCCC initiatives underline that only a minority of global e‑waste is recycled safely, exposing communities and ecosystems to hazardous substances.

Reliable measurement is a central gap. ForumIAS and independent reviews highlight that current disclosures and lifecycle assessments are patchy: many corporate claims understate aggregate impacts by focusing on per‑query electricity, ignoring embodied emissions from manufacture, mining, water for cooling, and end‑of‑life disposal. Estimates for the ICT sector’s share of global emissions vary widely, typically 1.8%–3.9%, reflecting inconsistent methodologies and missing data on indirect effects. The Global Environment Facility and other bodies have urged standardised lifecycle frameworks to make the footprint auditable and comparable.

Policy responses are nascent but evolving. UNESCO recognised environmental harms in its AI recommendations, and regulators in the EU and the United States have proposed measures to rein in high‑compute emissions. The UNEP has recommended five measures , standardised impact measurement, mandatory disclosure, algorithmic efficiency improvements, green data centres and integration of AI policy with environmental regulation , as pragmatic building blocks. International reports and the GEF suggest embedding lifecycle assessments in national digital strategies to align innovation with ecological resilience.

For India those recommendations point to several concrete priorities that would be relevant to industrial decarbonisation practitioners and energy planners. First, integrate AI into the Environmental Impact Assessment regime or an equivalent digital infrastructure assessment so high‑compute projects undergo formal environmental scrutiny. Second, develop national standards and metrics for AI lifecycle measurement, covering energy, water, materials and e‑waste, that feed into regulatory compliance and corporate ESG reporting. Third, require disclosure of direct and lifecycle impacts for large training and hyperscale operations to improve market transparency and enable procurement policies that favour lower‑impact providers. Fourth, accelerate deployment of renewables and grid flexibility measures around data‑centre clusters, and incentivise water‑efficient cooling and closed‑loop systems where data centres locate in water‑stressed regions. Finally, strengthen e‑waste collection and safe recycling capacity to avoid offloading downstream harms as hardware is retired more rapidly.

These steps do not preclude beneficial uses of AI for emissions management and industrial optimisation, but they reframe policy so benefits are not achieved by shifting burdens elsewhere. According to the ForumIAS analysis and corroborating sector studies, aligning AI policy with climate and resource governance is necessary to prevent digitalisation from becoming another vector of unsustainable growth. For professionals engaged in industrial decarbonisation, the implication is immediate: integrate AI infrastructure exposure into corporate decarbonisation roadmaps, demand credible lifecycle data from technology suppliers, and engage with regulators to establish the measurement and procurement standards that will steer the sector toward lower‑impact computing.