Artificial intelligence and high-performance computing are changing data center cooling requirements. New processors consume more power and concentrate more heat within each rack, making conventional room-level air cooling less effective on its own.

Meanwhile, data center operators must reduce electricity demand, manage water responsibly, and maintain the reliability expected from mission-critical infrastructure.

These pressures are driving several developments:

• Closed-loop, direct-to-chip liquid cooling
• Higher-temperature cooling loops
• Air-cooled dry heat rejection
• Hybrid air-and-liquid cooling
• More precise environmental controls
• Modular cooling plants

At Flex Air, we design and manufacture custom air-handling units and modular mechanical systems for complex commercial and industrial facilities. In this article, we explain how these technologies work, where they provide value, and what project teams should consider when developing a sustainable data center cooling strategy.

What Is Sustainable Data Center Cooling?

Sustainable data center cooling removes heat reliably while limiting energy use, water consumption, emissions, and material waste.

An effective design must balance several priorities:

• Equipment temperature requirements
• Rack power density
• Local temperature and humidity
• Water availability
• Electrical capacity
• Redundancy and uptime
• Maintenance access
• Future computing demand
• Capital and operating costs

These priorities do not always align. Dry cooling can reduce on-site water use but require more fan or compressor energy during hot weather. Evaporative cooling can lower electrical demand but consume more water.

The most sustainable option depends on the facility, workload, location, and operating conditions.

Why Data Center Cooling Requirements Are Changing

Traditional data centers often distribute conditioned air throughout the data hall. Cool air enters the front of each server, absorbs heat, and exits through the rear.

This approach remains practical for many applications. However, AI processors and high-performance computing equipment produce much higher heat loads within a smaller area. Increasing room-level airflow may become inefficient or physically impractical beyond certain rack densities.

As a result, the industry is shifting toward removing heat closer to its source. Direct-to-chip and immersion cooling systems use liquid to capture processor heat before it enters the room.

Air handling still plays an important role. Power supplies, storage devices, networking equipment, lighting, and electrical systems continue to release heat into the space. Data halls may also require filtration, humidity management, pressure control, and general air circulation.

The future of data center cooling is therefore likely to include both air and liquid systems.

Are Data Centers Replacing Water Cooling with Air Cooling?

Some facilities are adopting air-cooled dry systems to reduce water consumption. However, the data center industry is not abandoning liquid cooling.

The key distinction is between using liquid and consuming water.

A closed-loop system circulates the same coolant between the IT equipment and the heat-rejection equipment. The system can then reject heat through an air-cooled dry cooler rather than an evaporative cooling tower.

In a typical closed-loop, direct-to-chip system:

1. Cold plates collect heat from processors.
2. Coolant carries the heat away from the servers.
3. A coolant distribution unit transfers heat to the facility loop.
4. Pumps move the heated liquid to outdoor heat-rejection equipment.
5. Dry coolers release the heat into the air.
6. The cooled liquid returns through the closed circuit.

This arrangement uses liquid for efficient heat transfer without continuously consuming water through evaporation.

Recent Development: Zero-Water Evaporation Designs

Water availability has become an important consideration when selecting data center locations and cooling systems.

Microsoft has introduced a data center design for AI workloads that uses closed-loop, direct-to-chip cooling. After the initial fill, the system recirculates its water rather than losing it through evaporation. Microsoft describes the design as consuming zero water for cooling during operation.

“Zero water for cooling” does not mean that the facility contains no water. It means the cooling process does not require a continuous water supply to replace evaporative losses.

This distinction is important when comparing cooling technologies. A liquid-cooled system can have low water consumption if it operates within a properly sealed loop.

Recent Development: Higher-Temperature Liquid Cooling

Direct-to-chip cooling places cold plates on CPUs, GPUs, and other high-heat components. Liquid absorbs heat directly from these components and carries it into the facility cooling system.

Newer equipment can operate with warmer cooling water. NVIDIA’s Vera Rubin platform, announced in 2026, uses a 45°C, or 113°F, liquid supply temperature.

When the cooling loop operates at a higher temperature, the facility may be able to reject heat through outdoor air without using mechanical chillers during suitable weather.

Potential benefits include:

• More hours of chiller-free operation
• Lower compressor energy
• Reduced cooling-tower dependence
• More effective use of dry coolers
• Improved heat-recovery potential
• Support for higher rack densities

Local climate remains critical. A dry cooler can only reject heat when outdoor conditions provide an adequate temperature difference. Hotter climates may require larger heat exchangers, increased airflow, or supplemental cooling.

Hybrid Cooling Provides a Practical Transition

Many data centers contain a mixture of conventional servers and high-density AI equipment. Replacing the entire cooling system may be unnecessary or impractical.

A hybrid cooling strategy can use different technologies for different loads:

• Custom air handlers for conventional server areas
• Direct-to-chip cooling for AI racks
• Rear-door heat exchangers for selected equipment
• Dry coolers during favorable weather
• Evaporative or mechanical assistance during peak conditions

Hybrid designs can also support phased expansion. Owners can add liquid cooling where rack density requires it while retaining efficient air cooling elsewhere.

The design team must clearly define how much heat each system will remove. This prevents the air system from being oversized after the liquid loop begins handling most of the processor load.

How to Improve Air-Cooling Efficiency

Air cooling remains a reliable and effective option for many data centers. Several design practices can improve its performance.

Separate supply and exhaust air

Hot-aisle and cold-aisle containment prevents heated server exhaust from mixing with cool supply air. This reduces recirculation and creates more consistent server inlet conditions.

Unused rack openings and gaps around cables should also be sealed. Even small openings can allow hot air to bypass containment.

Control airflow according to demand

Variable-speed fans allow the air-handling system to respond to actual cooling requirements. Reducing fan speed during lower-load periods can provide meaningful energy savings.

Controls should rely on server inlet temperature and pressure measurements rather than a single room thermostat.

Reduce unnecessary pressure drop

Filters, coils, dampers, sound attenuators, and transitions create airflow resistance. Higher resistance requires more fan power.

Designers should review the complete air path and select components that meet filtration and thermal requirements without creating excessive pressure loss.

Use economizers when conditions permit

Air-side economizers use suitable outdoor air to reduce mechanical cooling. Their performance depends on local temperature, humidity, and air quality.

Facilities in cooler climates may achieve more economizer hours than those in hot or humid regions.

Design for maintainability

Filters, coils, fans, sensors, and control components require regular inspection and service. Providing adequate access helps operators maintain the efficiency and reliability intended by the original design.

When Does a Data Center Need a Custom Air-Handling Unit?

Standard air-handling units may work for straightforward applications with common airflow, footprint, and control requirements.

A custom air-handling unit becomes valuable when the project requires:

• High airflow or static pressure
• Redundant fans or components
• Precise temperature and humidity control
• Specialized filtration
• Integration with chilled-water systems
• Unusual dimensions
• Restricted installation access
• Indoor or outdoor construction
• Specific casing materials or finishes
• Factory-integrated controls and piping
• Retrofit installation within an existing facility

Using a standard unit in a complex application can lead to extensive field modifications, inefficient airflow, limited service access, or a larger equipment footprint than the site can accommodate.

Flex Air designs custom air-handling units around the actual operating and installation requirements. Our capabilities include high-CFM and high-static applications, indoor and outdoor units, integrated piping, energy recovery, control packages, existing AHU retrofits, factory testing, and knockdown installation support.

Environmental Monitoring and Automated Controls

Cooling systems operate more efficiently when they respond to real conditions rather than fixed assumptions.

A data center environmental monitoring system may track:

• Server inlet and exhaust temperatures
• Supply and return air temperatures
• Humidity
• Differential pressure
• Airflow
• Rack-level power demand
• Fan and pump performance
• Chilled-water temperatures
• Filter pressure drop
• Liquid and refrigerant leaks

Controls can use this information to adjust fans, pumps, valves, and dampers.

For example, the system can reduce airflow when the IT load falls, increase cooling near a developing hot spot, or activate an economizer when outdoor conditions become suitable.

Monitoring also supports preventive maintenance. A rising filter pressure drop, declining coil performance, or unexpected temperature change may identify a problem before it affects server operation.

Comparing Common Data Center Cooling Methods

Cooling method	Water implications	Primary advantage	Key consideration
Conventional air cooling	Generally low on-site use	Familiar and widely supported	May require substantial fan and compressor energy
Evaporative cooling	Consumes water through evaporation	Efficient heat rejection in suitable climates	Local water availability
Dry cooling	Little or no evaporative water use	Reduces onsite water demand	Performance decreases as outdoor temperature rises
Direct-to-chip cooling	Low consumption with a closed loop	Removes concentrated heat efficiently	Requires compatible IT and facility infrastructure
Rear-door heat exchangers	Depends on heat-rejection method	Helps retrofit higher-density racks	Adds piping and maintenance requirements
Immersion cooling	Low consumption with a closed loop	Supports very high equipment densities	Requires specialized hardware and maintenance
Hybrid cooling	Varies by operating mode	Balances performance, energy, and water	Requires coordinated controls

This table provides a general comparison. Actual performance depends on system configuration, climate, workload, and operating strategy.

Questions to Ask When Selecting a Cooling System

Before choosing a cooling technology, owners and engineers should consider the following questions.

What rack densities must the system support?

Current and expected rack power density will determine whether air cooling remains practical or liquid cooling is required.

What heat remains in the room?

Even when cold plates remove processor heat, the air system may still need to manage heat from memory, storage, power supplies, networking equipment, and electrical infrastructure.

How much water will the facility consume annually?

Calculate water consumption across normal and peak operating conditions. Consider whether the system will use potable, reclaimed, or recycled water.

How will the system perform during extreme weather?

Annual averages can hide peak conditions. The design must maintain cooling capacity during the hottest and most humid expected weather.

What happens when equipment fails?

Mission-critical systems require clear redundancy strategies for fans, pumps, controls, heat exchangers, and power supplies.

Can the system expand with the IT load?

The facility should accommodate future racks and cooling modules without major disruption to existing operations.

Can technicians maintain the equipment safely?

Equipment layout should provide access to filters, coils, fans, pumps, valves, sensors, and control panels.

Measuring Data Center Cooling Performance

Power Usage Effectiveness, or PUE, compares total facility energy with the energy consumed by IT equipment. It helps operators understand how much electricity supports cooling, power conversion, and other non-IT systems.

PUE does not measure water consumption or account for the source of the facility’s electricity.

A broader assessment may include:

• Water Usage Effectiveness: Water consumed relative to IT energy
• Carbon Usage Effectiveness: Operational carbon emissions relative to IT energy
• Cooling energy per unit of IT load
• Fan and pump energy
• Chiller operating hours
• Economizer operating hours
• Cooling-water consumption
• Recovered heat
• Renewable-energy use

Teams should monitor these metrics throughout operation, not only during initial commissioning.

How Modular Cooling Plants Support Sustainable Design

Modular mechanical construction moves assembly and testing away from the project site and into a controlled manufacturing environment.

A modular cooling plant may include:

• Chillers
• Pumps
• Heat exchangers
• Piping
• Valves
• Controls
• Electrical components
• Structural supports

This approach can provide several advantages.

Factory integration and testing

Teams can inspect piping, verify controls, test equipment, and address issues before shipment.

Less onsite coordination

Preassembled systems reduce onsite piping, wiring, welding, and trade congestion.

Phased capacity

Owners can add cooling modules as IT demand grows instead of installing all projected capacity at the beginning of the project.

Consistent execution

Controlled manufacturing supports repeatable fabrication and quality procedures across multi-phase or multi-site projects.

Faster deployment

Module fabrication can occur while foundations, utilities, and other site work continue.

Flex Air manufactures modular mechanical systems for heating, cooling, and process-water applications. Our capabilities include custom chilled-water plants, integrated piping, in-house design support, factory testing, and quality-control procedures.

Frequently Asked Questions

Is air cooling more sustainable than water cooling?

Not in every application. Air cooling can reduce water use but may consume more electricity in hot conditions. Evaporative cooling can use less energy but requires water. The best option depends on climate, water availability, grid emissions, and server density.

Does liquid cooling consume water continuously?

Not if it operates in a closed loop with dry heat rejection. The system can recirculate the same liquid after its initial fill.

Why do AI data centers use liquid cooling?

AI processors produce high, concentrated heat loads. Liquid cooling removes that heat closer to its source and can support higher rack densities than conventional room-level air cooling.

Do liquid-cooled data centers still require air handlers?

Many do. Air handlers may still manage heat from components that are not connected to the liquid loop. They can also provide circulation, filtration, humidity control, and pressure management.

What is a modular data center cooling plant?

It is an offsite-manufactured system that combines cooling equipment, piping, controls, electrical components, and structural support in a transportable module.

Building a More Efficient Cooling Strategy

Sustainable data center cooling is not a choice between air and liquid. The most effective design may use liquid to capture concentrated processor heat, air to manage remaining room loads, and dry or hybrid equipment to reject heat outdoors.

Success depends on matching the cooling system to the workload, climate, water supply, building layout, and long-term expansion plan.

At Flex Air, we work with engineers, contractors, and facility owners to develop custom air-handling units and modular mechanical systems around specific performance and installation requirements.

By addressing airflow, equipment selection, controls, maintainability, and modular integration early, project teams can build cooling infrastructure that is efficient, reliable, and ready for future computing demands.