Data Center Cooling: CRAC and CRAH Units, Chillers and PUE

Cooling is the largest non-IT energy use in most data centers, and it runs every hour of the year. According to Lawrence Berkeley National Laboratory, U.S. data centers consumed about 4.4% of total U.S. electricity in 2023 (176 TWh, up from 58 TWh in 2014) and may consume 6.7% to 12% by 2028. As load grows, the efficiency of the cooling plant matters more each year. This guide explains how data center cooling works, why it degrades, how to measure it, and where different fixes belong.

How data center cooling works

Most enterprise and colocation data centers built in the last few decades use one or a mix of these approaches:

• CRAC units (computer room air conditioners): packaged units on the white floor with their own compressors (direct expansion, or DX), rejecting heat to outdoor condensers, dry coolers or condenser water.
• CRAH units (computer room air handlers): fans and a chilled-water coil, with no compressor of their own. The compressors live in a central chiller plant.
• Central chiller plants: water-cooled or air-cooled chillers, pumps and cooling towers, often with water-side economizers for “free” cooling in cool weather. Our water-cooled chiller and air-cooled chiller guides cover these in depth.
• Air-side economizers, in-row coolers, rear-door heat exchangers and, increasingly, liquid cooling (direct-to-chip or immersion) for high-density AI and HPC racks.

The temperature targets

ASHRAE Technical Committee 9.9 publishes the Thermal Guidelines for Data Processing Environments. As reproduced in DOE's 2024 Best Practices Guide for Energy-Efficient Data Center Design, the 2021 recommended range for air-cooled IT equipment inlet air is 64.4°F to 80.6°F (18°C to 27°C), with wider allowable ranges by equipment class (for example, 59°F to 89.6°F for class A1). Many older rooms still run far colder than they need to.

PUE in one paragraph

Power usage effectiveness (PUE) is total annual facility energy divided by annual IT equipment energy. A PUE of 1.6 means that for every 1 kWh delivered to IT, another 0.6 kWh goes to cooling, power conversion, lighting and other overhead. The DOE guide describes an average data center as having a PUE of about 1.6, with some recent super-efficient facilities below 1.1. Uptime Institute's 2025 global survey found that average PUE levels have shown little change for the sixth consecutive year, with improvements constrained by legacy infrastructure and regional barriers to efficient cooling. PUE measures infrastructure overhead, not how efficiently the IT itself works, so it should be read alongside absolute kWh.

A worked example of what cooling overhead costs

Take an illustrative facility with a steady 500 kW IT load. Over a year (8,760 hours) the IT equipment uses about 4.38 million kWh. At a PUE of 1.6, the whole facility uses about 7.0 million kWh, so roughly 2.6 million kWh goes to overhead, most of it cooling. If airflow fixes, setpoint changes and restored heat-transfer performance brought PUE to 1.5, total use would drop to about 6.57 million kWh, a saving of roughly 440,000 kWh a year. At an assumed $0.10 per kWh, that is about $44,000 a year before demand charges. The numbers are hypothetical; plug in your own load, PUE and rate, and use our savings calculation guide for the full method.

How data center cooling efficiency degrades

Airflow problems (the most common)

• Hot exhaust air recirculating to server inlets because of missing blanking panels, open cable cutouts or no aisle containment.
• Cold air bypassing the IT equipment and returning straight to the cooling units.
• Supply air set far below what the IT needs, to cover hot spots caused by the two problems above.
• CRAC/CRAH units fighting each other, some humidifying while others dehumidify.

The DOE guide notes that higher return temperatures, made possible by hot aisle/cold aisle separation, extend economizer hours significantly and allow lower fan airflow.

Mechanical and heat-transfer problems

• DX CRAC units: dirty filters and coils, low or high refrigerant charge, worn compressors, dirty outdoor condensers, and oil film inside evaporator and condenser tubes. These units often run near-constant load for many years, so gradual heat-transfer loss accumulates.
• Chillers: condenser tube scale and fouling, air in the condenser of low-pressure machines, poor condenser water flow, and part-load operation of oversized plants. DOE FEMP's O&M guide notes that any buildup on chiller heat-transfer surfaces insulates the tubes and requires a larger temperature difference between water and refrigerant.
• Controls drift: chilled-water setpoints left low, condenser water reset disabled, economizer changeover points never tuned.

How to measure cooling performance

For background on the efficiency metrics, see kW per ton, EER and COP explained.

Maintenance and operating changes that matter, in order

1. Fix airflow: blanking panels, cable cutout seals, containment, and removing perforated tiles from hot aisles.
2. Raise supply temperature step by step within the ASHRAE recommended range while monitoring inlet temperatures.
3. Coordinate CRAC/CRAH units so they do not fight over humidity; use variable-speed fans where available.
4. Tune the chiller plant: chilled-water reset, condenser-water reset, economizer changeover, and staging chillers to operate near their best efficiency. FEMP's O&M guide notes that on a centrifugal chiller, raising chilled-water temperature 2°F to 3°F can improve system efficiency by as much as 3% to 5%.
5. Heat-transfer maintenance: clean condenser tubes and coils, maintain water treatment, purge air from low-pressure chillers, verify refrigerant charge on DX units.

Where an internal oil-film treatment fits, and where it does not

Data centers are a good test of honesty about where a product applies, because so much of the opportunity is in airflow and controls, and some cooling equipment has no oil in circulation at all.

Where it does not fit:

• Oil-free compressors. Magnetic-bearing centrifugal chillers and other oil-free designs have no circulating oil, so there is no oil film to address.
• CRAH units, pumps, towers, economizers and liquid-cooling loops. These have no refrigerant circuit of their own.
• Airflow and setpoint problems. No refrigerant-side treatment fixes recirculation or overcooling.

Where it may be worth evaluating:

• Fleets of older DX CRAC units with oil-lubricated compressors that run at steady load year-round.
• Oil-lubricated screw or centrifugal chillers whose approach temperatures have crept up even after tube cleaning and water treatment are in order.
• Facilities that want to extend the life of existing cooling equipment while a larger redesign, such as a move to liquid cooling, is planned.

CryogenX4 says its treatment is a one-time application installed while the system runs, with no downtime, and that most installs take one day without modifying the system. That point matters in a facility where cooling cannot be interrupted, but it does not replace your own change-control process. According to the company, the treatment lifts oil film from internal heat-exchange surfaces and improves lubricity in the compressor; the company reports savings of up to 30%, and results vary by equipment condition. The company reports treating equipment from 1-ton split units up to 1,600-ton water-cooled chillers.

Questions to ask

• Which of our cooling units have oil-lubricated compressors, and which are oil-free?
• What are our IT inlet temperatures today, and how far below the ASHRAE recommended upper limit are we running?
• What is our cooling plant kW per ton, measured, not nameplate?
• Have condenser approach temperatures risen over the last two years?
• For any treatment: how will results be isolated from IT load changes and weather? See baselines and weather normalization.

Next step

Inventory your cooling units by type and lubrication, and pull a year of PUE and cooling kW data. Fix airflow and setpoints first. Then, if you have oil-lubricated DX units or chillers that are drifting, read our data center industry guide and contact CryogenX4 about a measured pilot on redundant equipment.