Overview
- India’s data centres used about 150 billion litres of water in 2024–25 and CEEW projects need could rise to roughly 358 billion litres a year by 2030 as capacity expands.
- Companies and researchers warn that training large AI models is water‑intensive — UNU‑INWEH estimates training ChatGPT‑4 required about 592 million litres — while everyday inference traffic will add steady ongoing demand.
- Operators are beginning to publish water figures, with Amazon reporting more than 9 billion litres for its global data centres in 2025, even as disclosure methods vary and complicate comparisons.
- More than 65% of India’s capacity is concentrated in a few metros and capacity climbed from about 375 MW in 2020 to 1,500 MW in 2025, raising local basin stress and prompting planning reviews, permitting changes, and some moratoria.
- Experts and industry recommend switching to reclaimed or seawater, closed‑loop and immersion cooling, careful site selection, and standardized basin-level planning, but each option carries trade‑offs in cost, land use, energy or carbon impacts.