7 Views
Artificial intelligence is changing the way data centres are designed, powered and cooled. As organisations deploy large language models, generative AI, AI inference, machine learning and GPU-intensive applications, conventional air cooling is increasingly challenged by higher rack power densities and concentrated heat loads.
This is particularly important for Data Center India, where the market is expanding rapidly alongside cloud, AI and digital infrastructure demand. India’s data-centre capacity increased from around 375 MW in 2020 to approximately 1,500 MW in 2025, according to India’s Ministry of Electronics and Information Technology. The government also reports that advanced cooling technologies, including direct-to-chip liquid cooling and immersion cooling, are being adopted to reduce water usage.
At the same time, JLL’s 2026 India data-centre research reports that capacity stood at around 1.6 GW in mid-2026 and projects expansion to 6 GW by 2029, with AI workloads contributing to the acceleration in demand.
For AI-focused facilities, therefore, cooling is no longer simply a facility-management concern. It is becoming a core part of the compute architecture.
Traditional data centres primarily use air to remove heat from servers. Fans move warm air away from processors and other components, while cooling systems condition the air before it is circulated back into the data hall.
Liquid cooling takes a different approach. Instead of depending primarily on air to transport heat away from high-power components, it uses a liquid coolant with much greater heat-transfer capability.
One important architecture is direct-to-chip liquid cooling, where cold plates are positioned directly against heat-generating components such as GPUs and CPUs. The coolant absorbs heat and transfers it through a cooling loop and heat exchanger.
NVIDIA describes direct-to-chip cooling as a design in which a cold plate is placed directly on the GPU, while noting that other server components can still require air cooling in hybrid architectures.
This makes liquid cooling particularly relevant for modern AI servers.
AI workloads are fundamentally different from many conventional enterprise workloads.
Training and inference can involve large numbers of GPUs operating continuously. As GPU performance increases, the amount of electrical power consumed by individual accelerators and entire racks also increases. Most of that electrical energy ultimately becomes heat that must be removed.
NVIDIA notes that data-centre rack densities have risen dramatically, citing hyperscale facilities capable of supporting more than 135 kW per rack, compared with roughly 20 kW historically.
This creates several challenges:
Higher heat density
Greater cooling requirements
Increased facility power demand
Greater thermal-management complexity
Higher demands on rack and floor design
Potential limitations on conventional air-cooling systems
JLL similarly notes that increasing AI rack densities require changes to data-centre design, including higher power consumption, floor loading and heat-generation requirements.
The AI Rack Density Challenge
A conventional enterprise server rack may operate at relatively moderate power density. AI infrastructure can be substantially denser because multiple high-performance GPUs are installed in the same rack.
For example, NVIDIA’s current rack-scale AI platforms include fully liquid-cooled architectures such as the GB300 NVL72, which integrates 72 Blackwell Ultra GPUs and 36 Grace CPUs.
This illustrates why cooling has become a fundamental consideration when planning next-generation data centers in India.
Liquid has significantly different thermal characteristics from air. NVIDIA states that liquid is nearly 1,000 times denser than air, enabling it to transport heat more effectively.
In practical terms, direct-to-chip liquid cooling can:
Cold plates positioned close to the GPU or CPU provide a direct path for heat transfer.
This reduces dependence on room air as the primary mechanism for transporting heat away from the chip.
As more computing power is concentrated into each rack, the cooling architecture needs to scale alongside it.
Deloitte India identifies increasing rack density as a major driver for advanced cooling solutions such as liquid cooling and rear-door heat exchangers.
Air-cooled systems require fans and substantial airflow to move heat through the facility. Liquid cooling can transfer heat closer to its source, potentially reducing the cooling infrastructure required for the same computing load.
Thermal management directly affects sustained computing performance. Keeping GPUs within their required operating temperatures helps AI infrastructure maintain predictable performance under demanding workloads.
Liquid cooling can allow operators to rethink rack layouts, cooling distribution and facility infrastructure as they move toward AI-oriented data halls.
Major Types of Liquid Cooling for Data Centers
Liquid cooling is not one single technology. Data-centre operators can choose from several architectures depending on workload density, infrastructure and deployment requirements.
Direct-to-chip cooling uses cold plates attached directly to processors.
A typical system includes:
GPU/CPU → Cold Plate → Coolant Loop → CDU → Heat Exchanger → Facility Cooling System
A Coolant Distribution Unit (CDU) manages the transfer between the technology cooling loop and the facility cooling infrastructure.
This approach is particularly relevant for high-density GPU clusters.
Rear-door heat exchangers are installed at the back of server racks and capture heat from exhaust air before it enters the data hall.
They can be useful for increasing density in existing facilities without completely replacing the server cooling architecture.
However, they may not provide the same thermal capability as direct-to-chip systems for the highest-density AI racks.
With immersion cooling, servers or components are submerged in a dielectric fluid.
The fluid absorbs heat directly from the equipment, eliminating much of the conventional air-cooling path.
Immersion cooling can support very high-density environments, but it introduces different operational, maintenance and fluid-management requirements.
NVIDIA’s technical discussions identify direct-to-chip and immersion cooling as different approaches, with immersion systems involving specialised dielectric fluids and operational considerations.
Liquid cooling is particularly relevant to India’s data-centre expansion because the country is simultaneously dealing with rising compute demand, power requirements and sustainability considerations.
The Government of India reported in March 2026 that data-centre capacity had reached approximately 1,500 MW in 2025 and that electricity demand from data centres is estimated to reach 13.56 GW by 2031–32. The government also specifically identified direct-to-chip liquid cooling, adiabatic cooling and immersion cooling among technologies being adopted to minimise water usage.
JLL’s 2026 research identifies established markets including Mumbai, Chennai and Pune, while Hyderabad, Visakhapatnam and Delhi NCR are among locations seeing growing infrastructure activity.
For operators planning new AI facilities, cooling technology therefore needs to be considered alongside:
Power availability
Renewable-energy sourcing
Water availability
Climate conditions
Network connectivity
Land and floor loading
GPU density
Disaster recovery
Facility scalability
How Liquid Cooling Can Improve Data Center Efficiency
Cooling itself consumes energy.
NVIDIA has reported that cooling can account for up to 40% of a data centre’s electricity consumption, although the actual proportion varies significantly by facility design and operating conditions.
Liquid cooling can improve efficiency by moving heat away from processors more directly and reducing some of the energy required for high-volume air movement and mechanical cooling.
Earlier NVIDIA and Equinix testing reported that liquid-cooled data centres could use around 30% less energy for the same workloads in the tested configurations, with the cited example achieving a PUE of approximately 1.15 versus 1.6 for its air-cooled comparison. These figures are specific to those tested designs rather than a universal performance guarantee.
This distinction is important: liquid cooling does not automatically make every data centre more efficient. Overall efficiency depends on the complete facility architecture, including pumps, CDUs, heat exchangers, chillers, cooling towers, water systems and operating temperatures.
Liquid Cooling and Water Conservation
Water efficiency is becoming an increasingly important consideration for data-centre development.
India’s government has explicitly highlighted the relationship between cooling technology and water consumption and reported industry adoption of technologies designed to minimise water use.
Liquid cooling can support closed-loop designs in which coolant continuously circulates between the server-side cooling system and heat-rejection infrastructure.
This does not mean every liquid-cooled facility is automatically water-free. The ultimate water requirement depends on how heat is rejected from the facility.
Possible approaches include:
Closed-loop liquid cooling
Liquid-to-liquid heat exchange
Dry coolers
Adiabatic systems
Chilled-water systems
Hybrid cooling architectures
Immersion cooling
Therefore, when evaluating a data center in India, operators should examine total water consumption rather than looking only at the server-level cooling technology.
Liquid Cooling for NVIDIA GPU Infrastructure
The growth of GPU computing is one of the strongest reasons for the adoption of liquid cooling.
Modern NVIDIA AI platforms increasingly incorporate liquid-cooled configurations. For example, NVIDIA’s GB300 NVL72 is described as a fully liquid-cooled rack-scale platform containing 72 Blackwell Ultra GPUs and 36 Grace CPUs.
NVIDIA also lists liquid-cooled configurations of its RTX PRO Server platform, including an MGX 6U design with eight liquid-cooled RTX PRO 6000 Blackwell Server Edition GPUs.
This trend has implications for Indian AI infrastructure providers offering:
GPU cloud
GPU servers
AI infrastructure
AI inference
LLM hosting
Machine-learning training
High-performance computing
Generative AI platforms
As GPU deployments become denser, cooling must be engineered as part of the overall GPU infrastructure rather than treated as a separate facility layer.
Despite its benefits, liquid cooling introduces additional design and operational requirements.
Infrastructure Changes
Existing air-cooled data centres may require significant modifications before they can accommodate high-density liquid-cooled racks.
Higher Initial Complexity
Operators need additional equipment such as CDUs, piping, manifolds, pumps, heat exchangers and monitoring systems depending on the architecture.
Maintenance Requirements
Liquid systems require appropriate procedures for leak detection, coolant management, component replacement and maintenance.
Facility Design
New AI facilities need to consider liquid cooling during the design stage rather than trying to retrofit infrastructure after racks have already been installed.
Skilled Workforce
Technicians and facility teams need expertise covering both IT hardware and liquid-cooling infrastructure.
A 2026 Data Center Dynamics analysis focused specifically on India highlights the importance of selecting the appropriate architecture early in greenfield AI data-centre projects, particularly as rack densities rise.
How Indian Data Centers Can Prepare for High-Density AI
Organisations planning AI-ready infrastructure should consider the following approach:
Cooling infrastructure should not be designed only around today’s hardware. Future GPU power and rack density should be considered during facility planning.
Define expected kW-per-rack requirements before selecting the cooling architecture.
For high-density GPU deployments, direct-to-chip cooling should be evaluated as part of the initial infrastructure design.
Power delivery and thermal management are closely connected. Increasing compute density increases both electrical and cooling requirements.
Indian data-centre projects should evaluate local water conditions and select an appropriate heat-rejection strategy.
AI infrastructure is evolving quickly. Modular cooling and power infrastructure can make future expansion easier.
The Future of Liquid Cooling in Data Center India
India’s data-centre industry is moving toward increasingly AI-oriented infrastructure.
JLL’s latest India research projects a rise from around 1.6 GW of data-centre capacity in mid-2026 to 6 GW by 2029, while highlighting AI-driven demand and the need for advanced cooling technologies.
At the technology level, the transition is also continuing toward increasingly integrated liquid-cooled AI systems. NVIDIA’s current platforms already include rack-scale liquid-cooled architectures, while its 2026 materials describe next-generation systems moving toward even broader liquid cooling.
This suggests that the future Data Center India landscape will increasingly be shaped by the combination of:
AI + GPUs + High-Density Racks + Liquid Cooling + Renewable Power + Advanced Networking
Liquid cooling is therefore not simply a replacement for traditional air conditioning. It is becoming an important infrastructure layer for building data centres capable of supporting the next generation of AI workloads.
The growth of AI is forcing data-centre operators to rethink how computing infrastructure is designed, powered and cooled. Higher GPU performance and rack density create concentrated thermal loads that can be difficult to manage efficiently with traditional air cooling alone.
For Data Center India, this transition is especially significant. India’s data-centre capacity is expanding rapidly, while government and industry projections point toward continued growth in AI and high-performance computing. Advanced cooling technologies such as direct-to-chip liquid cooling, rear-door heat exchangers and immersion cooling can help operators address the thermal, power and sustainability challenges associated with high-density infrastructure.
Liquid cooling is a thermal-management technology that uses liquid coolant to remove heat directly from high-performance components such as GPUs and CPUs. It is increasingly used in AI data centers where rack densities are too high for traditional air cooling alone.
AI workloads rely heavily on GPUs that generate substantial amounts of heat. Liquid cooling can remove this heat more efficiently and support high-density GPU infrastructure while helping data centers manage power and cooling requirements.
The major approaches include direct-to-chip liquid cooling, rear-door heat exchangers and immersion cooling. Direct-to-chip cooling is particularly relevant for high-density AI and GPU deployments.
Liquid cooling can reduce the energy required for certain cooling operations, particularly by reducing dependence on high-volume airflow. However, actual energy savings depend on the complete data-center design, including pumps, heat exchangers, chillers and heat-rejection systems.
Yes. Liquid cooling is well suited to high-density GPU infrastructure because it can transfer heat directly from GPUs and CPUs. This makes it increasingly relevant for AI training, inference, HPC and other compute-intensive workloads.
Businesses should evaluate GPU rack density, power availability, cooling architecture, water usage, network connectivity, scalability, reliability and operational expertise before selecting a data-center provider.
