Data centers turn nearly all of the electricity they use into heat. Every kilowatt delivered to IT equipment must be removed continuously, 24 hours a day, with no interruption. Cooling is therefore one of the largest energy users in a data center after the IT load itself, and one of the systems most critical to uptime.
Data center cooling design differs from comfort cooling in almost every respect: the loads are dense, sensible and constant, the tolerance for failure is very low, and the equipment being protected has its own published environmental limits. This overview walks through the main design decisions and points to more detailed guides on each.
How data center cooling differs from comfort cooling
| Factor | Comfort cooling | Data center cooling |
|---|---|---|
| Load density | Tens of watts per square metre | Often kilowatts per rack, and rising sharply with AI workloads |
| Load profile | Varies with weather and occupancy | Nearly constant, year round |
| Latent load | Significant from people and ventilation | Very small; almost entirely sensible |
| Operating hours | Occupied hours | 8,760 hours per year |
| Failure tolerance | Discomfort | Equipment shutdown or damage within minutes |
| Environmental targets | ASHRAE 55 comfort | ASHRAE TC 9.9 equipment guidelines |
This is why office-grade comfort equipment is rarely suitable for server rooms.
Start with the environmental envelope
ASHRAE Technical Committee 9.9 publishes thermal guidelines for data processing environments, which define recommended and allowable ranges of temperature and humidity at the server inlet. The widely used recommended range for most equipment is 18 to 27°C (64.4 to 80.6°F) inlet temperature. Designing to the upper end of the recommended range, rather than cooling the room to office temperatures, allows warmer supply air and chilled water, more economizer hours and lower energy use. See ASHRAE TC 9.9 thermal guidelines.
Define the load
Cooling design starts with the IT load and its expected growth:
- Total IT capacity in kilowatts or megawatts, and how it will be phased.
- Rack density: average and maximum kilowatts per rack. Traditional enterprise racks might be a few kilowatts; high-performance and AI racks can be many times higher.
- Cooling method required by the IT equipment: air, liquid, or a mix.
- Other loads: UPS losses, lighting, electrical room heat, people and envelope gains.
- Redundancy level required by the business.
Density drives almost everything else. Air cooling handles moderate densities well with good airflow management. Beyond that, liquid cooling becomes necessary.
Air cooling
Airflow management
The most important principle in air-cooled data centers is to keep cold supply air and hot exhaust air separate. Hot and cold aisles, blanking panels, sealed cable cutouts and full containment prevent recirculation and bypass, which in turn allow higher supply temperatures and lower fan energy. See hot aisle and cold aisle containment.
Cooling units
- CRAC units (computer room air conditioners) have their own compressors and reject heat through condensers or dry coolers.
- CRAH units (computer room air handlers) use chilled water from a central plant.
- In-row coolers sit between racks for close-coupled cooling.
- Rear-door heat exchangers remove heat at the back of each rack.
- Fan walls and air handlers supply large halls from outside the white space.
See CRAC vs CRAH units for a comparison.
Raised floor or slab
Underfloor supply through a raised floor was the traditional approach. Many new facilities use slab floors with overhead or flooded-room supply and containment, which avoids underfloor obstructions and supports heavier racks.
Liquid cooling
As rack densities climb, especially for AI and high-performance computing, air cannot carry the heat efficiently. Liquid cooling options include:
- Direct-to-chip cooling, with cold plates on processors fed by a coolant distribution unit.
- Immersion cooling, single-phase or two-phase, where servers are submerged in dielectric fluid.
- Rear-door heat exchangers, which are a hybrid between air and liquid.
Liquid cooling allows much warmer facility water, improving free cooling potential and heat reuse opportunities. Most facilities with liquid-cooled racks still need air cooling for the remaining heat from other components and equipment.
Heat rejection and central plants
Heat removed from the white space must be rejected outdoors. Options include:
- Air-cooled chillers, with or without integrated free cooling coils.
- Water-cooled chillers with cooling towers.
- Dry coolers or adiabatic coolers for direct free cooling.
- Pumped refrigerant economizers on DX systems.
- Direct or indirect air-side economizers.
Canada's cool climate offers excellent free cooling potential. Facilities in Calgary, Edmonton and much of Ontario and Quebec can operate without mechanical cooling for a large share of the year if supply temperatures are set appropriately. See free cooling and economizers for data centers in Canada.
Redundancy and resilience
Cooling redundancy is typically expressed as N (just enough capacity), N+1 (one spare unit), N+2, or 2N (fully duplicated systems). Redundancy applies not only to cooling units but to pumps, chillers, piping paths, controls and power supplies. The Uptime Institute tier classification and the TIA-942 standard describe facility-level resilience. Thermal ride-through, the time the room can tolerate a loss of cooling, matters as much as installed redundancy. See data center cooling redundancy.
Efficiency metrics
Power usage effectiveness (PUE), defined by The Green Grid and standardized in ISO/IEC 30134-2, is total facility energy divided by IT energy. A PUE of 1.0 would mean all energy goes to IT. Cooling efficiency is usually the largest lever on PUE. Water usage effectiveness (WUE) tracks water consumption, which matters for facilities using evaporative cooling.
Design and commissioning process
- Define IT load, density, growth and redundancy with the owner and IT team.
- Select environmental targets from ASHRAE TC 9.9.
- Choose air, liquid or hybrid cooling based on density.
- Select heat rejection and economizer strategy for the local climate.
- Model airflow with CFD for complex or high-density layouts.
- Design controls and monitoring, including integration with DCIM.
- Commission with load banks under realistic heat loads, including failure scenarios.
Key takeaways
- Data center cooling serves dense, constant, sensible loads with very low tolerance for failure.
- ASHRAE TC 9.9 guidelines set the environmental envelope; warmer targets save energy.
- Airflow separation and containment are the foundation of efficient air cooling.
- Liquid cooling becomes necessary as rack densities rise, and enables warmer water and heat reuse.
- Canada's climate offers strong free cooling potential; redundancy and commissioning protect uptime.
Talk to an engineer
Have a project or a problem this guide touches on? Tell us about the building and we’ll be in touch within one hour.