Cooling India's Booming AI Data Centres

Google's planned 1-GW hyperscale data centre in Visakhapatnam will use air-cooling amid local water concerns.
TCS HyperVault plans a 1-GW AI data-centre campus in Hyderabad using Direct-to-Chip (DTC) liquid cooling.

Capacity growth: ~520 MW (2020) → 1.5–1.57 GW (2026).
Electricity demand: Ministry of Power projects AI/data centres could add 26.3 GW of demand by 2031–32.
1:1 power-to-heat conversion: Nearly all electricity consumed by IT equipment becomes heat; a 1-GW data centre generates ~1 GW thermal load.
High-density AI racks: Traditional cloud servers: 5–10 kW/rack; modern AI racks: 120–150 kW/rack.
Thermal Wall: Physical limit of air cooling (~40 kW/rack) beyond which liquid cooling is necessary.
Cooling energy penalty: Fans, chillers and pumps consume a significant share of data-centre electricity.
Geographical concentration: Over half of capacity in Mumbai/Navi Mumbai; other hubs: Chennai, Hyderabad, Bengaluru, Delhi-NCR, Jamnagar.
Noise: Air-cooling a 1-GW facility can generate up to 100 dB low-frequency noise travelling several kilometres.
Carbon lock-in risk: Insufficient renewable capacity could increase coal reliance, complicating Net Zero 2070.
Urban Heat Island (UHI) effect: Waste heat released into urban surroundings raises local temperatures and AC demand.

Air-Cooling: Fans, CRAC/CRAH units; impractical above ~40 kW/rack.
Free Cooling: Uses naturally cool ambient air to reduce mechanical refrigeration.
Direct-to-Chip (DTC) Liquid Cooling: Cold plates on CPUs/GPUs; ideal for high-density AI/HPC.
Rear-Door Heat Exchanger (RDHx): Liquid-cooled heat exchanger on rack's rear door captures hot exhaust air.
Immersion Cooling: Servers submerged in non-conductive dielectric fluid:
Single-phase: liquid circulated through heat exchanger without state change.
Two-phase: liquid boils to vapour, condenses back.
Evaporative Cooling: Water absorbs heat and evaporates; energy-efficient but water-intensive.
Dry Cooling: Air-cooled heat exchangers; no water needed, suited to water-scarce regions but needs larger infrastructure.
Alternatives: Geothermal heat rejection, natural water cooling (lakes), waste-heat recovery for district/industrial heating.

Balancing AI growth with water scarcity, energy efficiency and climate commitments.
Technology choice (air vs liquid vs evaporative vs dry) has direct implications for local resource conflicts, emissions and sustainability.

Prelims 2020: AI capabilities (Ans: 1, 3 and 4 only).
Mains 2015: Advantages and security implications of cloud hosting vs in-house hosting for government businesses.