Power Usage Effectiveness (PUE) is one of the most widely used metrics for evaluating data center energy efficiency. For facility owners and operators, reducing PUE means lowering the amount of energy consumed by supporting infrastructure relative to the energy used by IT equipment.
Cooling is an important part of this equation. Chillers, cooling towers, dry coolers, pumps, CRAC units, CRAH units, and cooling distribution systems all contribute to facility energy consumption. An inefficient cooling architecture can increase operating costs even when server performance and IT utilization remain unchanged.
However, reducing data center PUE is not simply a matter of installing more efficient air conditioners or replacing air cooling with liquid cooling. The result depends on the complete thermal management system, including airflow distribution, cooling equipment selection, operating temperatures, heat rejection, control strategies, and site conditions.
This guide explains how advanced cooling solutions can help reduce data center PUE, which technologies are worth evaluating, and what facility managers and engineers should consider before investing in a cooling upgrade.
1. Understand PUE and the Role of Cooling Infrastructure
Before selecting new cooling equipment, operators need to understand what PUE measures and how cooling affects the result.
What is Power Usage Effectiveness?
PUE is calculated by dividing total data center facility energy consumption by IT equipment energy consumption.
For example, if a data center consumes 1,500 MWh of total facility energy while its IT equipment consumes 1,000 MWh, its PUE is:
The difference between total facility energy and IT energy represents energy used by supporting infrastructure, including cooling, power distribution, lighting, and other facility systems.
A lower PUE generally indicates that a greater proportion of facility energy is being delivered to IT equipment rather than supporting infrastructure.
However, PUE has important limitations:
It measures overall facility energy efficiency, not cooling efficiency alone.
It does not directly measure server computing efficiency.
It can change with IT load, even when cooling equipment performance remains similar.
It does not directly account for water consumption or the environmental impact of electricity generation.
For this reason, PUE should be evaluated alongside cooling-specific measurements, such as chiller electricity consumption, fan power, pump energy, and cooling water usage.
How does cooling affect data center PUE?
Cooling systems consume energy through several processes:
Moving air through server racks and cooling equipment.
Compressing refrigerant in mechanical cooling systems.
Circulating chilled water or other cooling fluids.
Rejecting heat through condensers, dry coolers, or cooling towers.
Controlling temperature, humidity, and system operation.
Poor airflow, excessive cooling capacity, inefficient chillers, and unnecessary pump or fan operation can all increase facility energy consumption.
The objective is to remove the required amount of heat while minimizing the energy used by the complete cooling system.
2. Optimize Airflow Management to Reduce Unnecessary Cooling Demand
Airflow management is often one of the first areas to investigate when trying to reduce data center PUE.
A cooling system may have sufficient capacity but still consume excessive energy because cold air does not reach server intakes efficiently.
Implement hot aisle and cold aisle containment
In a conventional data center, servers draw in cold air from the front of the rack and discharge heated air from the rear.
A hot aisle/cold aisle layout separates these two airflow paths.
Without effective separation, hot exhaust air can mix with cold supply air. This creates several problems:
Higher server inlet temperatures.
Uneven temperature distribution.
Increased cooling demand.
Unnecessary fan operation.
Greater risk of localized hot spots.
Cold aisle containment encloses the cold aisle to keep supply air directed toward server intakes.
Hot aisle containment encloses the hot aisle to collect server exhaust and direct it toward the cooling system return.
The correct arrangement depends on the facility’s rack layout, ceiling configuration, raised-floor system, cooling equipment placement, and maintenance requirements.
Eliminate bypass airflow and hot air recirculation
Not all cold air delivered by a cooling system reaches server equipment.
Some may bypass racks through floor openings, cable penetrations, or gaps around cabinets. At the same time, hot exhaust air may recirculate toward server intakes.
To reduce these problems, operators should inspect:
Missing blanking panels in server racks.
Unsealed cable openings.
Gaps around rack bases.
Incorrectly positioned perforated floor tiles.
Obstructed airflow paths.
Containment doors and roof panels.
Poorly arranged network and power cables.
These corrections can improve the effectiveness of existing cooling equipment without requiring a complete system replacement.
Balance airflow instead of simply increasing fan speed
Increasing fan speed may temporarily reduce high inlet temperatures, but it can also increase electricity consumption.
The better approach is to identify where airflow is insufficient and correct the distribution problem.
Measure temperatures at multiple rack inlet locations, review cooling unit supply and return conditions, and check whether airflow reaches the intended cabinets.
The objective is to provide adequate cooling at each rack while avoiding unnecessary airflow and fan energy.
3. Upgrade Precision Cooling Systems and Optimize Their Operation
CRAC and CRAH systems remain important in many traditional and enterprise data centers. Improving their performance can help reduce the energy consumed by cooling infrastructure.
However, the best configuration depends on the existing cooling plant, IT load, room layout, and operating conditions.
Optimize CRAC and CRAH systems
CRAC units typically use direct-expansion refrigeration to cool room air. CRAH units generally use chilled-water coils supplied by a central cooling plant.

Both systems can support efficient data center cooling when properly selected and controlled.
Key optimization measures include:
Use variable-speed fans. Variable-speed fan operation can reduce fan energy during periods of lower cooling demand. Fan speeds should be adjusted according to actual rack inlet temperatures and airflow requirements.
Coordinate multiple cooling units. Cooling units serving the same room should operate under a coordinated control strategy. Poorly coordinated systems may compete with one another, create uneven airflow, or operate at unnecessary capacity.
Optimize supply air temperature. Cooling setpoints should be based on IT equipment requirements and the facility’s operating envelope. Excessively low supply temperatures can increase refrigeration demand without providing meaningful operational benefits.
Maintain heat transfer surfaces. Dirty filters, obstructed coils, and poorly maintained fans can reduce system performance and increase energy consumption.
Evaluate in-row cooling for concentrated rack loads
In-row cooling places cooling equipment close to server racks rather than relying entirely on room-level air distribution.

This arrangement can shorten airflow paths and improve temperature control in areas where heat loads are concentrated.
In-row cooling may be suitable when:
Rack layouts create localized cooling requirements.
Room-based cooling has difficulty distributing air evenly.
Existing facilities need additional cooling capacity without major room reconstruction.
High-density racks require cooling closer to the heat source.
However, equipment placement, airflow direction, fan energy, maintenance access, and integration with existing cooling systems must be evaluated.
Consider rear door heat exchangers
Rear door heat exchangers capture heat from server exhaust air as it leaves the rack.
Depending on the design, they can be passive or actively assisted. Many configurations use water flowing through a heat exchanger at the rear of the rack.
By removing heat near the rack, rear door heat exchangers can reduce the amount of heat released into the data center room.
They are particularly worth evaluating for concentrated rack heat loads where conventional room cooling is becoming difficult.
However, their actual energy benefit depends on water temperatures, flow requirements, pump energy, heat rejection infrastructure, and the remaining room cooling load.
4. Use Efficient Chillers, Dry Coolers, and Free Cooling Technologies
Cooling equipment does not operate independently. Chillers, pumps, heat exchangers, dry coolers, and cooling towers form part of an integrated heat removal system.
Optimizing this system can have a significant effect on facility energy consumption.
Improve data center chiller efficiency
A data center chiller removes heat from chilled water and transfers it to a separate heat rejection circuit.
Its energy consumption depends on operating conditions, including:
Cooling load.
Chilled-water supply temperature.
Condenser temperature.
Outdoor conditions.
Compressor operating range.
Part-load performance.
Water flow and heat exchanger condition.
Operators should assess chiller efficiency across the facility’s actual operating conditions rather than relying exclusively on nominal equipment ratings.
Where the system design permits, optimized chilled-water temperatures and coordinated chiller staging can reduce unnecessary compressor operation.
Evaluate dry coolers for heat rejection
A dry cooler uses ambient air to remove heat from a circulating fluid through a heat exchanger.

Depending on outdoor conditions and the required cooling temperatures, dry coolers can support:
Partial or full chiller bypass.
Reduced mechanical refrigeration demand.
Heat rejection for suitable liquid cooling circuits.
More flexible cooling plant configurations.
Their performance depends heavily on ambient air temperature, heat exchanger design, fluid temperature, airflow, and operating conditions.
Use water-side economization where conditions allow
Water-side economization uses outdoor heat rejection equipment and heat exchangers to provide cooling without relying entirely on mechanical refrigeration.
A typical arrangement may include:
Dry coolers or cooling towers.
Heat exchangers.
Pumps.
Chilled-water or facility-water circuits.
Mechanical chillers for conditions when economization is insufficient.
The system can reduce chiller operating hours when outdoor conditions are suitable.
However, economization does not eliminate all energy consumption. Fans, pumps, and controls continue to operate.
Cooling tower configurations also require consideration of water consumption, treatment, scaling, corrosion, and local environmental requirements.
Compare common heat rejection technologies
Technology | Main function | Potential PUE benefit | Main consideration |
|---|---|---|---|
Data center chiller | Produces chilled water | Efficient operation can reduce mechanical cooling energy | Part-load performance and operating temperatures |
Dry cooler | Rejects heat through an air-cooled heat exchanger | Can support chiller bypass under suitable conditions | Outdoor temperature and heat exchanger capacity |
Cooling tower | Uses evaporative heat rejection | Can support efficient cooling plant operation | Water consumption and treatment |
Heat exchanger | Transfers heat between fluid circuits | Enables separation and optimization of cooling loops | Temperature approach, flow, and pressure drop |
The correct solution depends on local climate, water availability, cooling load, required temperatures, and the facility’s reliability strategy.
5. Adopt Liquid Cooling for High-Density AI and GPU Data Centers
AI training, high-performance computing, and GPU-intensive applications can create concentrated heat loads that are difficult to manage with conventional room-based air cooling alone.
Liquid cooling provides another way to remove heat by transferring it closer to the components that generate it.
It can support higher-density applications, but its effect on PUE depends on the complete cooling architecture.
Direct-to-chip liquid cooling
Direct-to-chip cooling uses cold plates mounted on selected high-heat components, such as CPUs and GPUs.
Coolant circulates through the cold plates, absorbs heat, and transfers it to a liquid cooling loop.
This approach reduces the amount of heat that must be carried away by server airflow.
Potential advantages include:
More direct heat capture from high-power components.
Reduced thermal burden on room-based air cooling.
Support for high-density server configurations.
Opportunities to use warmer coolant temperatures where the system design and IT equipment permit.
However, not every component is necessarily liquid-cooled. Memory, storage, power supplies, and other components may still release heat into the surrounding air.
The facility may therefore continue to require CRAH or CRAC systems for residual heat removal and room environmental control.
Use Coolant Distribution Units to manage liquid cooling loops
A Coolant Distribution Unit (CDU) manages coolant circulation and heat transfer in many liquid cooling systems.
Depending on its configuration, a CDU may provide:
Heat exchange between facility and technology cooling loops.
Coolant circulation and flow control.
Pressure and temperature monitoring.
Filtration and fluid management.
Alarm and leak detection functions.
Integration with facility monitoring systems.
The correct CDU selection depends on required flow, pressure, coolant type, heat load, temperature conditions, redundancy, and server compatibility.
Liquid cooling also introduces operational considerations such as connector compatibility, fluid chemistry, contamination, corrosion, leak detection, and pump reliability.
Consider immersion cooling for specialized applications
Immersion cooling places compatible IT hardware in a dielectric fluid.
Single-phase systems keep the fluid in liquid form, while two-phase systems use a phase change as the fluid absorbs heat.
Immersion cooling may be appropriate for specialized high-density computing environments, but it requires compatible hardware, fluid management, suitable service procedures, and a properly engineered heat rejection system.
Does liquid cooling automatically reduce PUE?
No. Liquid cooling can reduce the thermal burden on air-cooling systems, but it also introduces pumping, fluid circulation, heat exchange, and heat rejection requirements.
The overall PUE benefit depends on:
IT rack heat load.
Coolant supply and return temperatures.
CDU and pump efficiency.
Facility cooling water conditions.
Heat rejection equipment.
Residual air-cooling demand.
System control strategy.
A hybrid architecture may provide a practical balance: direct-to-chip cooling for high-heat components, combined with precision air cooling for residual heat and other facility spaces.
6. Improve Cooling System Controls and Real-Time Monitoring
Even well-designed cooling equipment can consume excessive energy when control strategies are poorly configured.
A facility needs accurate operating data to determine whether cooling equipment is delivering the required performance.
Monitor the right operating parameters
Useful monitoring points include:
Monitoring parameter | Why it matters |
|---|---|
Rack inlet temperature | Identifies potential IT overheating |
Rack outlet temperature | Helps evaluate heat removal |
Cooling unit supply and return air | Shows cooling performance and air distribution |
Chilled-water supply and return temperature | Helps evaluate cooling plant operation |
Water flow rate | Confirms whether the required cooling fluid is circulating |
Pump and fan power | Tracks energy consumption |
CDU pressure and temperature | Helps identify liquid loop abnormalities |
Cooling equipment alarms | Supports early fault detection |
Monitoring should be designed around the facility’s actual equipment and control architecture.
Integrate cooling equipment with BMS or DCIM
A Building Management System (BMS) or Data Center Infrastructure Management (DCIM) platform can provide centralized visibility into cooling equipment operation.
Useful capabilities may include:
Cooling unit status and alarms.
Temperature and humidity trends.
Fan and pump operating conditions.
Chiller staging and control.
Coolant flow and pressure monitoring.
Energy consumption tracking.
Equipment fault notifications.
With reliable monitoring, operators can identify unnecessary cooling operation, uneven temperature distribution, and abnormal equipment behavior before these issues become major operational problems.
Avoid overcooling
Overcooling occurs when cooling systems operate at lower temperatures or higher output than required.
It can result from conservative setpoints, poor sensor placement, outdated control sequences, or a lack of coordination between cooling units.
To address it:
Review the environmental requirements of installed IT equipment.
Compare rack inlet temperatures across the room.
Check sensor accuracy and placement.
Review cooling unit operating schedules.
Adjust control settings within approved equipment limits.
Monitor conditions after every adjustment.
Changes should be introduced carefully, with sufficient monitoring and appropriate operational safeguards.
7. Build a PUE Optimization Strategy Around Total Cost of Ownership
Advanced cooling equipment should be evaluated based on its contribution to the complete facility, not just its individual performance specifications.
A cooling system that has a higher purchase price may be justified if it reduces long-term energy costs and fits the facility’s operating conditions. Conversely, installing sophisticated equipment in a facility that does not need it may increase complexity without delivering a meaningful benefit.
Compare cooling strategies for different data center environments
Data center environment | Cooling strategy to evaluate | Key decision factors |
|---|---|---|
Traditional enterprise data center | Optimized CRAC/CRAH and airflow containment | Existing equipment, airflow, and room layout |
Medium-density server environment | CRAH, in-row cooling, or rear door heat exchangers | Rack heat distribution and cooling capacity |
High-density GPU data center | Direct-to-chip liquid cooling with residual air cooling | Server compatibility, CDU, and heat rejection |
Specialized high-performance computing | Immersion cooling or direct-to-chip systems | Hardware compatibility and maintenance |
New data center development | Integrated cooling plant and economization strategy | Climate, water, energy, and expansion plans |
What should buyers check before selecting cooling equipment?
Before purchasing or specifying equipment, engineering and procurement teams should request and evaluate:
Cooling performance
Total and sensible cooling capacity.
Rated and actual operating conditions.
Entering and leaving air or coolant temperatures.
Performance at expected partial-load conditions.
Energy consumption
Fan and pump power.
Compressor power where applicable.
Control and standby energy consumption.
Expected operating efficiency across the required load range.
System integration
Electrical requirements.
Water or refrigerant connections.
Flow and pressure requirements.
Control interfaces and monitoring protocols.
Compatibility with existing equipment.
Reliability and maintenance
Redundancy configuration.
Service access.
Spare parts availability.
Maintenance requirements.
Alarm and fault handling.
Warranty and technical documentation.
Long-term operating costs
Equipment purchase and installation.
Energy and water consumption.
Maintenance and replacement costs.
Infrastructure modifications.
Expected operating life and future expansion.
A proper assessment should compare the expected total cost of ownership under realistic operating conditions.
Frequently Asked Questions
1. What is a good PUE for a data center?
There is no single PUE value that is appropriate for every facility. PUE depends on climate, IT utilization, cooling architecture, power infrastructure, and measurement boundaries. It is more useful to compare a facility against its own operating history and relevant peer facilities with similar conditions.
2. Which cooling system is best for reducing data center PUE?
There is no universally best cooling system. Optimized CRAC/CRAH systems, in-row cooling, rear door heat exchangers, direct-to-chip liquid cooling, and immersion cooling can all be suitable in different applications. The correct choice depends on rack heat load, site conditions, IT hardware, and facility infrastructure.
3. Can cold aisle containment reduce PUE?
Cold aisle containment can improve airflow separation and reduce unnecessary mixing between hot exhaust air and cold supply air. This may allow cooling equipment to operate more efficiently, provided airflow distribution and control settings are properly optimized.
4. Does direct-to-chip liquid cooling improve PUE?
It can help reduce cooling energy in suitable high-density applications, particularly when it allows the facility to reduce mechanical refrigeration or optimize heat rejection. However, pump energy, CDU performance, residual air cooling, and facility water temperatures must also be considered.
5. How do dry coolers help reduce data center energy consumption?
Dry coolers reject heat using ambient air and a heat exchanger. When outdoor conditions and system temperatures permit, they can support partial or full chiller bypass, reducing mechanical refrigeration demand. Their performance depends on climate, heat load, and system design.
6. What information is needed to plan a PUE reduction project?
Start with total facility energy consumption, IT energy consumption, cooling equipment energy use, rack-level heat loads, inlet temperatures, chilled-water conditions, and site climate. These measurements help identify whether the main opportunity is airflow optimization, equipment upgrades, liquid cooling, or heat rejection improvements.
Conclusion: Optimize the Complete Cooling System, Not Just Individual Equipment
Reducing data center PUE requires a coordinated approach to thermal management.
Airflow containment, precision cooling optimization, efficient chillers, dry coolers, free cooling, and liquid cooling can all contribute to lower facility energy consumption when applied under suitable conditions.
For traditional data centers, improving airflow distribution and optimizing existing CRAC or CRAH equipment may be the most practical starting point. For high-density AI infrastructure, direct-to-chip cooling, CDUs, and hybrid cooling architectures may provide better control of concentrated heat loads.
The key is to select technologies based on measured thermal demand, operating conditions, reliability requirements, and long-term costs.
Planning a data center cooling upgrade or a new facility? Contact our team to discuss your project requirements and request a suitable configuration of precision air cooling, liquid cooling, and heat rejection equipment.