How to Select the Right Cooling System for Your Data Center

Selecting the right cooling system for a data center is one of the most consequential decisions in a facility's lifecycle. Get it right and you have a resilient, efficient thermal management platform that scales with your load. Get it wrong and you're looking at hot spots, redundancy gaps, and expensive retrofits.

This guide walks through the four primary cooling technologies — CRAC units, CRAH units, air-cooled chillers, and coolant distribution units (CDUs) — and the selection criteria that matter most for each.

CRAC vs. CRAH: Understanding the Difference

Computer Room Air Conditioning (CRAC) units are self-contained systems with their own refrigeration circuit — compressor, condenser, and evaporator all in one unit. They're well-suited for smaller deployments, edge sites, and facilities where a centralized chilled water plant isn't available or practical.

Computer Room Air Handlers (CRAH) units, by contrast, use chilled water supplied from a central plant. They're more energy-efficient at scale and better suited for large data centers where a chiller plant is already in place. The tradeoff: you're dependent on the chilled water infrastructure, which adds complexity and a potential single point of failure if not properly designed for redundancy.

Selection criteria: Available infrastructure, facility size, redundancy requirements, and total cost of ownership over a 10–15 year horizon. For new builds above 500kW of IT load, CRAH with a chilled water plant typically wins on efficiency. Below that threshold, CRAC units often make more economic sense.

Air-Cooled Chillers: When and Why

Air-cooled chillers are the backbone of the chilled water plant in most mid-to-large data centers. They reject heat to the outside air via condenser coils and fans — no cooling tower required, which simplifies water treatment and reduces maintenance overhead.

Modern air-cooled chillers with variable-speed compressors and EC fan technology can achieve PLV (part-load value) efficiencies that rival water-cooled systems in many climates. For facilities in moderate climates, free-cooling economizer options can dramatically reduce compressor runtime and operating costs.

Sizing considerations: Chiller capacity must account for peak IT load plus overhead for cooling the mechanical and electrical infrastructure. N+1 redundancy is the minimum for Tier III and above. Don't size to current load — size to the facility's design capacity with a clear path to add capacity as load grows.

Coolant Distribution Units (CDUs): The Liquid Cooling Path

As rack densities climb above 20–30kW per rack — driven by GPU clusters, AI training workloads, and high-performance compute — air cooling alone becomes increasingly inadequate. CDUs enable direct liquid cooling (DLC) by circulating coolant directly to server-side heat exchangers, removing heat at the source before it enters the room air.

CDUs sit between the facility's chilled water loop and the server-side coolant loop, managing pressure, flow rate, and temperature to protect sensitive compute hardware. They're the critical interface between facility infrastructure and the IT equipment.

When to specify a CDU: Any deployment with sustained rack densities above 20kW, AI/ML training infrastructure, or GPU clusters. Also consider CDUs for retrofitting high-density racks into existing air-cooled facilities where adding more CRAC/CRAH capacity isn't feasible.

Precision Air Handlers: Localized Cooling for High-Density Rows

In-row and overhead precision air handlers deliver cooling directly to the heat source — the server rack — rather than conditioning the entire room. In hot-aisle/cold-aisle containment architectures, they're often more efficient than perimeter CRAC units because they minimize the distance conditioned air must travel.

Precision air handlers work well in mixed-density environments where some rows run at high density and others don't, allowing targeted cooling capacity where it's needed without over-cooling the entire floor.

The Selection Framework

Before specifying any cooling equipment, work through these questions:

• What is the current IT load, and what is the design capacity of the facility?
• What is the average and peak rack density? Are there high-density zones?
• Is chilled water infrastructure available, or will this be a self-contained system?
• What redundancy tier is required (N, N+1, 2N)?
• What are the climate conditions at the site, and is free-cooling economization viable?
• Is liquid cooling required now, or is it a future consideration?

The answers to these questions drive the equipment selection. There's rarely a single right answer — the right cooling system is the one that matches your load profile, redundancy requirements, and total cost of ownership over the facility's design life.