Data Center: Cooling Technologies, Energy Use & Water Challenges

Subject: GS 03: Science and Technology

Context: Google’s proposed 1-GW data center in Visakhapatnam has reportedly opted for air-cooling amid concerns over the water requirements of large data centers.

• At the same time, Tata Consultancy Services’ proposed one-gigawatt HyperVault facility is being designed with direct-to-chip (D2C) liquid cooling for high-density artificial intelligence computing.

UPSC Online Preparation

About Data Center

• A data center is a specialised facility that houses servers, storage systems, networking equipment, and supporting infrastructure to store, process, and manage large volumes of digital data and computing workloads.
  • They are an important and integral component of the overall Digital India Infrastructure ecosystem.
• Data Center Architecture: A data center consists of several layers of computing infrastructure:
  • Transistors: Billions of transistors form a modern processor.
  • Processors: Processors perform computational operations and generate heat.
  • Servers: One or more processors are combined into a server.
  • Racks: Multiple servers are housed in a rack.
  • Clusters: Multiple racks are organised into clusters.
  • Data Center: Multiple clusters together form a large data center.

Why Does a Large Data Center Generate Heat?

• Role of Transistors: Modern processors contain billions of transistors that manipulate electrical currents to perform computational operations.
• Heat Generation: Electrical resistance and the charging/discharging of transistors convert part of the electrical energy into heat.
• Cumulative Heat: With millions of processors operating simultaneously, this heat accumulates on a very large scale.
• Thermal Wall: As computing power and heat density increase, conventional air cooling faces a physical limit in how much heat it can efficiently remove from a rack. This constraint is often referred to as the “thermal wall.”
• Heat Removal: A 1-gigawatt data center can, in principle, generate approximately 1 gigawatt of heat, making efficient heat removal essential.

How Does a Data Center Move Heat?

• Heat Generation: Heat originates at the transistor level during computational operations.
• Chip-Level Heat Transfer: Heat generated in the silicon die is first transferred to a heat sink, such as a cooling plate.
• Heat Removal: The heat is then transferred through the cooling system and ultimately released into the environment.
• Cooling Technology: The method used after the heat sink depends on the cooling technology adopted by the data center.

Proposed Air Cooling Technique by Google

• Working: Fans circulate cooled air through server racks, collect hot air, and cool it again using Computer Room Air Conditioning (CRAC) systems with refrigerants or Computer Room Air Handling (CRAH) systems with chilled water.
  • Configurations: Hot-aisle/cold-aisle containment separates hot and cold air; free cooling uses cooler outside air; while rear-door heat exchangers use chilled water to absorb heat from outgoing hot air.
• Advantages of Air Cooling:
  • Lower Upfront Cost: Air cooling generally has lower infrastructure costs; liquid cooling can carry a 7–10% premium due to specialised piping, leak-detection systems and coolant distribution units.
  • Mature Technology: It is a proven technology with a mature pool of technicians and established maintenance protocols.
  • Suitable for Low-Density Racks: Engineers often prefer air cooling for racks producing 20 kilowatts (kW) or less.
  • Climate Advantage: In cool or arid conditions, air-side economisers can reduce energy used for mechanical cooling by up to 70%.
• Limitations of Air Cooling
  • Thermal Wall: Air cooling can remove at best around 40 kW per rack, whereas modern Artificial Intelligence (AI) racks may generate 120–150 kW.
  • Higher Energy and Ownership Costs: Cooling high-density racks requires extremely high airflow, which increases power consumption and the total cost of ownership.
  • Limited Scalability: Air cooling alone is considered physically and economically impractical for a 1-gigawatt (GW) data center.
  • Acoustic Pollution: Large numbers of fans, chillers and air-handling units can generate significant noise, with some equipment producing up to 100 decibels (dB).

Click to Know UPSC OnlyIAS Coaching Centres

Other Cooling Technologies Used By Data Centers

• Direct Liquid Cooling: 
  • Working: A liquid coolant is brought directly to the heat-generating components, enabling more efficient heat transfer than air.
  • Cold-Plate Cooling: A metal plate with coolant channels is placed directly against the processor; the coolant absorbs heat and carries it to a heat exchanger.
  • Advantage: Liquids generally have a higher heat capacity than air, making liquid cooling suitable for high-performance computing.
  • Limitation: Components that do not require liquid cooling may still need conventional air cooling.
• Immersion Cooling: 
  • Working: Electronic components are immersed directly in a non-conductive liquid coolant.
  • Types: In single-phase immersion, the coolant remains liquid; in two-phase immersion, it boils after absorbing heat, and the vapor is subsequently condensed.
  • Advantage: Direct contact between the coolant and electronics enables highly efficient heat transfer and supports very high power densities.
  • Limitation: Requires specialised equipment and more complex coolant management.
• Evaporative Cooling: Heat is transferred to water, which is then evaporated in cooling towers or evaporative condensers; it is energy-efficient in low-humidity climates but water-intensive.
• Dry Cooling: Uses ambient air and finned heat exchangers to dissipate heat; it reduces water consumption but requires larger heat-exchange surfaces, particularly in warm climates.

Check Out UPSC CSE Books

Visit PW Store