India's AI data centre boom confronts water and power constraints

India is rapidly expanding AI‑focused data centres, with hyperscalers and local firms planning GPU‑dense facilities across Mumbai, Hyderabad, Chennai and Bengaluru. New builds are being designed around liquid cooling, as modern accelerators such as NVIDIA’s Blackwell GPUs exceed 1,000 W per chip and can demand up to 246 kW per rack.

The surge in capacity—already over 1.5 GW and projected to reach 6.5–17 GW by 2030—poses significant resource challenges. A 100‑MW data centre can consume about 2 million litres of water per day, and the sector’s total water use could approach 37.5 billion litres annually. Existing data centres used roughly 150 billion litres in 2024‑25, with forecasts that consumption will more than double by 2030. Many of the hub cities are classified as water‑stressed, and water availability is now a central factor in planning decisions.

Major tech companies have pledged tens of billions of dollars to build facilities in India, and the Adani Group alone pledged $100 billion, bringing the first wave of AI infrastructure investment to about $167.5 billion. Closed‑loop direct‑to‑chip liquid cooling is being promoted as a way to reduce water loss, which can be as high as 85 % with traditional evaporative systems.