A Crash Course in CRAC Cooling for Data Centers

CRAC means Computer Room Air Conditioner. It is precision cooling equipment designed to remove the continuous heat produced by servers while maintaining suitable airflow, temperature, and—where required—humidity in an IT environment.

A conventional CRAC is usually a compressorized, direct-expansion (DX) system. A CRAH, by contrast, generally uses a chilled-water coil supplied by a remote chiller plant. The names are useful, but vendors do not always use them consistently, so the important question is where the refrigeration cycle is located and what medium cools the room air.

As an Amazon Associate I earn from qualifying purchases.

What problem does a CRAC solve?

Nearly all electrical power consumed by servers, storage systems, network equipment, and power electronics eventually becomes heat. A CRAC removes that heat continuously and delivers the airflow needed to keep server inlets within their allowable environmental range.

That makes a CRAC more than an oversized office air conditioner. A complete precision-cooling design must address:

• Sensible heat from IT equipment.
• Air volume and static pressure at server inlets.
• Hot-air recirculation and cold-air bypass.
• Humidity where the equipment and operating envelope require it.
• Partial-load operation and low-load control.
• Redundancy during equipment or utility failures.
• Alarms, building-management-system (BMS) and data-center-infrastructure-management (DCIM) integration.
• Maintenance access and safe isolation.

The goal is not simply a comfortable room. It is stable, measurable thermal conditions at the equipment that matters.

ASHRAE’s data-center cooling guidance describes CRAC systems as compressorized systems and covers air-cooled DX, water-cooled DX, and related configurations.

How a DX CRAC works

A DX CRAC transfers heat using a refrigeration circuit. In plain language, the unit collects heat from indoor air and rejects it outside the IT space.

1. Return air enters the unit. Warm air from the data room, hot aisle, ceiling return, or return duct reaches the CRAC.
2. The indoor fan moves air across the evaporator coil. Refrigerant inside this cold coil absorbs heat from the air.
3. The compressor raises refrigerant pressure. This also raises its temperature so the captured heat can be rejected.
4. A condenser rejects heat. Depending on the design, heat goes directly to outdoor air or into a water or glycol loop.
5. An expansion device reduces pressure. The refrigerant becomes cold again before returning to the evaporator coil.
6. Conditioned air returns to the IT area. It may discharge beneath a raised floor, overhead, through ductwork, into a cold aisle, or close to a rack zone.

The cycle repeats as long as the room has a heat load and the system is operating.

Main heat-rejection arrangements

• Air-cooled DX: An outdoor condenser rejects heat directly to ambient air.
• Water-cooled DX: A water loop carries condenser heat away.
• Glycol-cooled systems: A water-and-glycol loop transfers heat to remote heat-rejection equipment and provides freeze protection where needed.
• Pumped-refrigerant designs: A refrigerant-pumping arrangement can reduce compressor operation under suitable conditions.
• Chilled-water systems: A remote chiller produces chilled water for a CRAH or chilled-water precision-cooling unit.

Therefore, “CRAC” should not automatically be treated as a synonym for “air-cooled DX.” Verify the manufacturer’s actual architecture.

CRAC versus CRAH

Attribute	CRAC	CRAH
Full name	Computer Room Air Conditioner	Computer Room Air Handler
Primary cooling mechanism	Usually an integral compressor and refrigerant circuit, commonly DX	Generally a chilled-water coil supplied by a remote chiller plant
Heat rejection	Outdoor condenser, water loop, glycol loop, or related system	Chiller plant, cooling tower, dry cooler, or other central system
Typical fit	Smaller, distributed, or modular IT rooms	Larger facilities with centralized chilled-water infrastructure
Main advantage	Room-level modularity and less dependence on a central water plant	Centralized plant efficiency and scalability
Main drawback	More local refrigeration equipment and potentially more distributed service work	Dependence on chillers, pumps, valves, water treatment, and heat-rejection equipment

This is a practical distinction, not an absolute law. Some product names blur the categories. When reviewing a submittal, ask: Does this room unit contain the compressor and refrigeration circuit, or does it receive chilled water from a separate plant?

ASHRAE’s technical handbook and Vertiv’s terminology glossary provide useful background, while the actual equipment submittal controls the project decision.

CRAC versus comfort HVAC

A rooftop unit, split system, or standard building air handler may be adequate for a small, low-density network closet—but it is not automatically suitable for a data room.

Data centers commonly have:

• High, continuous sensible loads, including nights and weekends.
• A need for year-round cooling, even when a comfort system would normally be in heating mode.
• Strict airflow-distribution requirements rather than merely an acceptable room-average temperature.
• Greater needs for filtration, alarms, monitoring, service access, and redundancy.
• Equipment that must continue operating through planned maintenance and selected failures.

The label alone is not decisive. An engineer should confirm the heat load, sensible capacity, airflow, operating range, controls, runtime, electrical supply, and backup strategy before a comfort unit is used for IT cooling.

Downflow and upflow CRAC designs

Downflow

A downflow CRAC generally takes return air from above and discharges supply air downward. It is often installed over or beside a raised floor, with perforated tiles or grilles distributing air to cold aisles.

This arrangement can work well when the raised floor is deliberately designed as a supply-air plenum. It performs poorly when floor openings, cable cutouts, pressure, and aisle layout are uncontrolled.

Upflow

An upflow unit generally discharges supply air upward or overhead and draws return air from the room or a defined return path. It can serve facilities without a raised floor, but the hot-air return route must be designed carefully.

In an open room, an upflow unit can re-entrain its own cool discharge air. ASHRAE guidance identifies separated mechanical galleries, ducted rear returns, and containment as ways to prevent this short circuit.

Airflow management is as important as cooling capacity

A CRAC does not cool a rack directly. It creates the temperature, pressure, and airflow path that allows the rack to reject heat.

A properly sized unit can still produce hot spots if the air distribution is wrong. A sound layout usually includes:

• Cold aisles: Rack fronts face the conditioned-air supply.
• Hot aisles: Rack rears discharge heat into a controlled return path.
• Containment: Doors, roofs, curtains, or ducted returns limit mixing between hot and cold air.
• Blanking panels: Empty rack spaces are closed so hot air cannot circulate through unused openings.
• Sealed penetrations: Cable cutouts and unused floor openings are sealed to reduce bypass air.
• Correct floor-tile placement: Supply openings belong where servers need air—not automatically in every aisle.
• Rack-level sensors: Measurements should reflect server-inlet conditions, not only a wall thermostat.

Controls may use underfloor pressure, containment differential pressure, supply-air temperature, return-air temperature, or combinations of these. Variable-speed fans can reduce energy use when the required airflow is lower, but the control sequence must preserve adequate airflow to the most demanding racks.

Excessive raised-floor pressure can increase fan energy and force air through unwanted leaks. Too little pressure can starve distant racks. Cabling, missing panels, open doors, and later construction changes can all undermine the original design.

Sensible and latent cooling

Data-center loads are dominated by sensible heat: heat that raises air temperature. Comfort HVAC systems often handle a larger latent component because occupants and outdoor ventilation introduce moisture.

For that reason, compare more than total capacity. A unit can have sufficient total cooling but insufficient sensible capacity for the IT load. Review:

• Sensible and total cooling capacity.
• Supply airflow and available static pressure.
• Entering- and leaving-air conditions.
• Outdoor design conditions.
• Fan power.
• Part-load performance and turndown.

Humidity control should follow the applicable equipment and facility operating envelope. Unnecessary humidification, dehumidification, or reheat consumes energy. Excessive humidification can also point to air leakage, over-ventilation, or a supply-air temperature that is lower than necessary.

How to size a CRAC system

A preliminary heat-load estimate can begin with:

Cooling load ≈ IT load + UPS and electrical losses + lighting + people + building heat gains

Because the IT load is electrical, its consumed power is approximately equal to the heat released into the room. That is a useful first estimate, not a final design calculation.

The final design should account for:

• Measured or specified IT load.
• Planned growth and deployment timing.
• UPS, PDU, transformer, and lighting losses.
• Envelope heat gain, infiltration, and ventilation.
• Fan heat and humidity-control loads.
• Local outdoor design conditions.
• Economizer operation and its changeover points.
• Required redundancy and maintenance capacity.
• Minimum-load behavior, cycling, and control stability.

Three sizing questions

1. Capacity: How many kilowatts or tons of cooling are required at the design condition?
2. Airflow: How much air must be moved, at what static pressure, and to which rack zones?
3. Reliability: What remains available if a unit, condenser, pump, chiller, control panel, or power source fails?

Do not solve growth and redundancy by automatically adding a percentage. Growth is a forecast; redundancy is an availability strategy. They should be documented separately.

Redundancy and availability

• N: Exactly enough cooling capacity for the design load.
• N+1: One additional unit or capacity increment beyond the required load.
• 2N: Two independent full-capacity cooling systems.
• Distributed redundancy: Several smaller units arranged so that one failure does not remove an entire cooling zone.

Redundancy must extend beyond the indoor cabinets. Review electrical feeds, generators, UPS-backed controls, condensers, chillers, pumps, cooling towers or dry coolers, water treatment, refrigerant circuits, and control networks.

Cooling for a UPS room may also require the same level of redundancy as cooling for the data room. See ASHRAE’s facilities guidance for related reliability considerations.

Economization and free cooling

Economization reduces compressor operation when outdoor conditions or heat-rejection conditions are favorable. It is not literally free: fans, pumps, controls, filters, water treatment, and maintenance still consume resources.

• Air-side economization: Suitable outdoor air is introduced directly, subject to filtration, humidity, contamination, and code controls.
• Water-side economization: A cooling tower, dry cooler, or heat exchanger reduces or bypasses mechanical chiller operation.
• Refrigerant-side or DX economization: The refrigeration system uses favorable outdoor conditions or refrigerant pumping to reduce compressor work.
• Indirect economization: Heat transfers through an air-to-air or air-to-fluid heat exchanger without directly introducing outside air into the IT space.

Evaluate outdoor temperature and humidity, air quality, freeze protection, water availability, damper and filter maintenance, controls complexity, local codes, and failure behavior. ASHRAE’s AI data-center framework discusses containment, raised supply-air temperatures within allowable thermal envelopes, and air-, water-, and refrigerant-based economization.

Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

Efficiency: look beyond the cabinet

COP is cooling output divided by electrical input under specified conditions. EER is a related imperial-unit measure. SCOP focuses on sensible cooling and can be particularly relevant to computer-room air conditioners.

At facility level, PUE is the ratio of total facility energy to IT-equipment energy. WUE relates water use to IT energy where water-consuming heat rejection is used. PUE is defined in ASHRAE’s guidance.

A highly efficient CRAC can be undermined by excessive fan speed, poor containment, inefficient pumps, dirty coils, bad setpoints, or an inefficient chiller plant. Request performance at the expected load and outdoor conditions—not just a nameplate number.

For U.S. projects, the Department of Energy regulates certain computer-room air conditioners. DOE states that final-rule testing requirements for covered equipment became mandatory on April 5, 2024. Verify the current federal rule, product scope, test method, and certification status at the DOE computer-room-air-conditioner page and the AHRI regulatory resource.

Refrigerants: verify the actual model

CRAC refrigerants vary by model, region, manufacturing date, and regulatory requirements. A current Vertiv DS105 page identifies R454B for that product, but this does not establish the refrigerant used by every CRAC family or market.

Before replacing or specifying equipment, verify:

• Refrigerant type and global-warming-potential classification.
• Charge size and leak-detection requirements.
• Technician certification requirements.
• Recovery and end-of-life obligations.
• Compatibility with existing condensers, piping, controls, and components.
• Local availability of refrigerant and replacement parts.

Never assume that a new unit can connect to an existing refrigerant circuit without an engineering compatibility review.

Maintenance checklist

Routine checks

• Filter loading and differential pressure.
• Supply and return temperatures.
• Humidity readings and sensor plausibility.
• Fan speed, vibration, and abnormal noise.
• Compressor starts, operating status, and alarms.
• Condenser-coil cleanliness.
• Condensate drains and pumps.
• Refrigerant or water/glycol leaks.
• Valve and actuator operation.
• Raised-floor obstructions and containment integrity.
• Sensor calibration and BMS/DCIM alarm communication.

Periodic service

• Electrical connections and insulation.
• Fan motor and bearing condition.
• Compressor performance.
• Refrigerant charge and leak inspection.
• Coil cleaning.
• Humidifier canisters or cylinders.
• Water quality, treatment, pumps, seals, and strainers on water-side systems.
• Automatic-transfer and generator behavior.
• Documented failover testing.

A unit can appear healthy while the room remains at risk. Common examples include a standby unit that is not enabled, a failed sensor reporting a plausible value, a restricted filter, a dirty condenser, a stuck chilled-water valve, a nuisance alarm that disabled a unit, or a running unit that has lost much of its cooling capacity.

Use qualified data-center HVAC technicians and a documented maintenance and failover plan.

When CRAC is the wrong choice

Choose CRAC or DX when:

• The site lacks a central chilled-water plant.
• The room is relatively small, distributed, or modular.
• Independent room-level cooling is valuable.
• Refrigerant service capability is available.
• Water use should be minimized.
• The climate and condenser arrangement support the design operating range.

Choose CRAH and chilled water when:

• A robust central chilled-water plant already exists.
• Centralized efficiency and scalability matter.
• The project supports many cooling units.
• The owner can operate chillers, pumps, valves, heat rejection, and water treatment.

Choose in-row cooling when:

• Rack densities vary substantially by row.
• Cooling needs to be placed close to high-density racks.
• Room-wide air distribution is inefficient.
• Capacity must be added incrementally without cooling the entire room.

Choose liquid or hybrid cooling when:

• Rack density exceeds what practical room airflow can support.
• Direct-to-chip cooling is required for AI or high-performance computing.
• Rear-door heat exchangers can capture a substantial share of rack heat.
• The facility needs higher supply-air temperatures or lower fan energy.

ASHRAE’s current AI framework discusses direct-to-chip and rear-door heat-exchanger approaches for approximately 50–100+ kW racks, while emphasizing that the correct architecture depends on the workload and thermal design. Many facilities will use a hybrid: room cooling for ordinary racks and close-coupled or liquid cooling for dense zones.

Buying and specification checklist

Ask manufacturers, representatives, and contractors for:

• Sensible and total capacity tables at project conditions.
• Airflow and available-static-pressure data.
• Part-load performance, fan turndown, and control sequences.
• Ambient operating limits and economizer changeover conditions.
• Redundancy, failure-mode, and controls documentation.
• Refrigerant details, charge, and service requirements.
• Sound, electrical, heat-rejection, and physical-access data.
• Warranty, parts availability, and response-time commitments.
• AHRI and DOE certification information where applicable.
• BMS/DCIM integration protocols, alarms, and points lists.
• Commissioning procedures, sensor calibration, and failover tests.

For enterprise equipment, pricing is normally quote-based rather than a retail checkout transaction. Installed cost depends on capacity, DX or chilled-water architecture, condensers or chillers, pumps, heat rejection, containment, electrical work, piping, controls, commissioning, local labor, and service contracts. Do not mistake an equipment-only figure—or a model number containing a capacity—for an installed project price.

Common mistakes to avoid

• Assuming every CRAC uses chilled water: Conventional CRACs are commonly compressorized and may be DX, air-cooled, water-cooled, or glycol-cooled.
• Assuming a larger unit is safer: Oversizing can cause short cycling, poor humidity control, low-load instability, higher fan energy, and greater single-unit failure impact.
• Judging the room by its average temperature: Rack-inlet hot spots and recirculation can remain hidden.
• Treating a raised floor as a complete airflow solution: Unsealed openings, poor pressure control, and misplaced tiles create bypass air.
• Calling economization free: Fans, pumps, controls, filtration, treatment, and maintenance still consume resources.
• Using nominal tons as the only sizing metric: Sensible capacity, airflow, static pressure, part-load operation, and redundancy are equally important.
• Assuming CRAC is universally suitable: High-density and AI deployments may require in-row, rear-door, direct-to-chip, or hybrid cooling.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.