Can Air-Cooling handle the heat from a 1 GW Data Centre?

Why in News?

Google has indicated that its planned gigawatt-scale data centre in Visakhapatnam will use air-cooling, following concerns about its water requirements. Meanwhile, TCS’s HyperVault explicitly envisages direct-to-chip liquid cooling for high-density AI workloads. The debate highlights the energy-water trade-off underlying India’s expanding digital infrastructure. 

Why are Data Centres Important?

• Data centres are specialised facilities that store, process, and manage digital information, forming the backbone of cloud computing, artificial intelligence (AI), fintech, e-commerce, and digital governance. 
• Data centres house servers, storage and networking equipment required to process and store digital information. Their expansion is being driven by:
  • Artificial intelligence and cloud computing
  • Digital India, UPI, e-commerce, 5G and Internet of Things
  • RBI’s payment-data localisation requirements
  • Cybersecurity and strategic data sovereignty
  • Demand for low-latency digital services.

They generate investment, support digital businesses and provide the infrastructure necessary for India’s emerging AI economy. India’s installed data-centre capacity increased from around 375-520 MW in 2020 to nearly 1.5 GW in 2025. It is estimated to reach 4.5-6.5 GW by 2030. Committed investments during 2019-25 were estimated at approximately $95 billion.

Why do data centres generate so much heat?

Data centres contain processors organised into servers, racks and clusters. Their billions of transistors switch electrical signals to process information. Electrical resistance, leakage currents, and repeated charging and discharging dissipate energy as heat.

Almost all electricity consumed by computing equipment ultimately becomes heat. 

• 1 GW = 1000 MW: If the IT equipment actually draws 1 GW, approximately 1 GW of heat must continuously be removed.
• Power versus energy: A constant 1-GW load operating throughout a year consumes 8.76 TWh of electricity.
• Capacity versus consumption: Announced capacity may represent the eventual campus build-out; actual demand depends on utilisation and construction phases.
• IT load versus total facility load: If 1 GW refers only to computing equipment, cooling and other supporting systems add to total electricity demand.

Heat must first travel from the chips into a cooling medium, and eventually into the external environment.

How do the major cooling technologies work?

• Air-cooling: Fans move air across components and carry heat away. Computer-room air conditioners use refrigerants, while air-handling systems may use chilled water. Separating hot and cold aisles prevents air from mixing and improves efficiency.
• Direct-to-chip liquid cooling: Coolant circulates through channels in metal cold plates attached to processors. It absorbs heat and carries it to a heat exchanger. Other components may still use air-cooling.
• Immersion cooling: Electronics are submerged in a non-conductive liquid. In single-phase systems, the liquid remains liquid; in two-phase systems, it boils and subsequently condenses. The latter uses the latent heat of vaporisation to transport heat effectively.
• Evaporative cooling: Water evaporates to remove heat, often through cooling towers. This can reduce electricity requirements but consumes water and becomes less effective in humid conditions.
• Dry cooling: Heat exchangers reject heat to outdoor air without deliberately evaporating water. Hot weather can require larger equipment or additional mechanical cooling.

Water-based liquids carry much more heat per unit volume than air, making them suitable for concentrated heat loads. However, liquid systems require additional plumbing, monitoring and maintenance. 

Why does conventional air-cooling struggle with AI workloads?

Air-cooling generally has lower initial costs, established maintenance practices and a large pool of trained technicians. It remains useful for lower-density racks, particularly where outdoor conditions permit free cooling.

However, high-density AI computing creates several constraints:

• Higher auxiliary electricity demand: Moving sufficient air through densely packed equipment requires powerful fans and cooling systems.
• Performance losses: Inadequate cooling can cause processors to reduce operating speed (thermal throttling) or shut down.
• Space requirements: Spreading equipment across more racks reduces heat density but increases building space and supporting infrastructure.
• Noise: Extensive fan and compressor operation can disturb nearby communities, making acoustic design and monitoring necessary.

Thus, air-cooling’s lower initial cost must be assessed against its lifetime electricity, space and performance costs.

NVIDIA’s GB200 NVL72 rack contains 72 Blackwell GPUs and 36 Grace CPUs. Its approximately 120-kW rack demand is addressed through a liquid-cooled design.

Why does location matter?

Cooling performance depends strongly on ambient conditions.

• Cool climates offer greater opportunities to use outside air and reduce mechanical refrigeration.
• Hot climates make heat rejection harder and can increase electricity demand precisely when the wider grid faces peak cooling demand.
• Humid climates limit evaporative cooling because moist air has less capacity to absorb additional water vapour.

For coastal Visakhapatnam, this implies that cooling choices should be evaluated against local temperature and humidity throughout the year, including extreme conditions.

Wider implications for India: 

• Digital competitiveness: Reliable computing infrastructure supports AI, cloud services, research and digital public services.
• Energy security: Large, continuous loads require generation capacity, transmission upgrades and reliable supply. The IEA’s 2025 base-case assessment projected global data-centre electricity consumption at approximately 945 TWh by 2030. 
• Water security: Cooling demand can compete with domestic and agricultural needs, especially during droughts.
• Indirect environmental costs: Low on-site water consumption does not eliminate water use or emissions associated with electricity generation.
• Environmental justice: Local communities may bear noise, water and infrastructure burdens while benefits accrue more widely.
• Climate resilience: Heatwaves, floods and coastal hazards must inform site selection and backup arrangements.

Way Forward:

• Adopt workload-specific cooling: Combine air-cooling for suitable equipment with liquid cooling for dense AI clusters.
• Disclose measurable performance: Report Power Usage Effectiveness and Water Usage Effectiveness, alongside absolute consumption. Lower ratios alone can conceal rising total resource use. 
• Assess cumulative local impacts: Evaluate water availability, electricity demand and noise across all proposed facilities in a region.
• Prioritise water-sensitive designs: Consider closed loops, dry cooling and treated wastewater where technically suitable.
• Scrutinise replenishment claims: Water restoration should address the affected watershed and relevant season; replenishment elsewhere does not necessarily resolve local scarcity.
• Integrate clean electricity and efficiency: Combine renewable supply, storage, efficient hardware and improved server utilisation.
• Explore heat reuse: Nearby industries or other users may utilise waste heat, although low temperatures and transport costs can limit feasibility.

India’s objective should be reliable computing with the lowest practical combined burden on energy, water and communities. For dense AI facilities, hybrid cooling is often a stronger engineering proposition than reliance on conventional air-cooling alone. 

UPSC Mains Practice Question:

Q. “Expansion of AI infrastructure presents an energy-water-environment challenge.” Discuss with reference to cooling technologies used in hyperscale data centres. 

Q. Examine the locational advantages and constraints for developing a world-class data centre industry in India. How can renewable energy integration address the sustainability and energy security challenges faced by this sector?

PDF