Everybody calls these CRACs. Half the time they’re not. And if you’re planning an AI retrofit, getting that distinction wrong can send you down the wrong mechanical path.
CRAH vs CRAC is one of the most common mix-ups we run into, and it’s an understandable one.
Both units sit in the data center doing the same basic job, both look similar from the front, and both show up in the same conversations about room-level cooling. But they’re mechanically different machines, and which one you have (or which one you’re specifying for a new build) changes how your facility scales, how it fails, and what it costs to run.
If you’ve inherited a legacy room and aren’t sure which you’re standing in front of, or you’re planning a retrofit and need to know what you’re actually replacing, this distinction matters before you spend a single dollar.
Don’t Just Look at the Box in the White Space. Follow the Heat.
The box in the room tells you less than you’d think. The faster way to tell CRAC and CRAH apart — and to understand where liquid cooling fits in the same conversation — is to trace where the heat actually goes once it leaves the server:
CRAC: Server → air → evaporator/refrigerant → condenser/heat rejection → atmosphere
CRAH: Server → air → chilled-water coil → chilled-water loop → chiller/heat rejection → atmosphere
Liquid cooling: GPU/CPU → coolant → CDU → facility water → heat rejection
Once you follow that path, the CRAC-vs-CRAH question stops being about which box is sitting in the room and becomes about which infrastructure is actually doing the refrigeration work — and that’s the distinction that matters for scaling, failure planning, and cost.
What Is a CRAC?
A CRAC—computer room air conditioner—uses a DX (direct expansion) refrigeration cycle to cool the air. Its cooling circuit includes a compressor, refrigerant, evaporator, and heat-rejection system, which may be integral to the unit or connected remotely. A CRAC doesn’t depend on a central chilled-water plant to do its job; it makes its own cooling, the same basic principle as a standalone air conditioner, just built and sized for a data center’s continuous, high-sensible-heat load rather than a comfort-cooling application.
That self-sufficiency is the CRAC’s defining trait. It’s a good fit for smaller rooms, standalone facilities, or sites where running chilled-water piping isn’t practical or cost-justified. The tradeoff is efficiency: a compressor-based refrigeration cycle is generally less efficient at scale than a well-designed chilled-water system, and CRACs typically need a condenser (air-cooled or water-cooled) as a heat-rejection point outside the room.
What Is a CRAH?
A CRAH—computer room air handler—doesn’t make its own cooling. It’s a coil-and-fan unit: a CRAH relies on chilled water supplied by a central plant, district/campus loop, or other chilled-water source. The CRAH pulls warm room air across that chilled-water coil and pushes conditioned air back into the space. All the actual refrigeration work happens elsewhere, at the central plant, which typically serves multiple CRAH units across the facility.
Because a CRAH is just moving air across a coil rather than running its own refrigeration cycle, it’s mechanically simpler and, at data center scale, generally more efficient—the central plant can be optimized once and shared across the whole room instead of duplicating a refrigeration system in every unit. The tradeoff is dependency: a CRAH is only as good as the chilled-water plant behind it, and that plant is a shared point of both efficiency gain and risk.
The Core Mechanical Difference
| Factor | CRAC | CRAH |
| Cooling source | Self-contained (own compressor/refrigerant loop) | Chilled water from a central plant |
| Infrastructure dependency | Standalone — needs a condenser/heat-rejection point | Depends entirely on the central chilled-water plant |
| Typical efficiency at scale | Often attractive at smaller scale; system efficiency depends on DX design, ambient conditions and economization | Central chilled-water plants can deliver strong efficiency at scale, particularly when optimized for part load, water temperature and economization |
| Best fit | Smaller rooms, standalone sites, no existing chilled-water plant | Larger facilities already built around a central plant |
| Failure mode | CRAC architectures tend to localize refrigeration failures to individual units | CRAH architectures move more of the cooling dependency upstream to shared plant infrastructure. Reliability depends on how redundancy is designed at both the room and plant levels |
| Retrofit complexity | Straightforward — self-contained, drop-in replacement | Requires tie-in to existing chilled-water piping |
Both are room-level cooling—this comparison isn’t CRAH/CRAC vs. rack-level architecture like RDHx or CDUs. It’s a question of how the room gets its cooling, not where the cooling is delivered.
Why This Distinction Matters More as Density Rises
At moderate, uniform density, the CRAC-vs-CRAH choice is mostly an infrastructure and efficiency question—which one fits the facility you already have, or the one you’re building. Neither one, on its own, changes the ceiling you’ll eventually hit as individual racks climb into high-density territory (30, 50, 80+ kW). That ceiling is a room-level cooling limitation, not a CRAC-specific or CRAH-specific one.
Where the distinction does matter at high density is efficiency and total cost. A facility running CRAH units off a well-optimized central plant has more headroom to push PUE down as load grows, because the refrigeration side scales once, centrally, instead of unit-by-unit. A CRAC-based room hits its efficiency ceiling sooner, simply because every additional unit is running its own compressor.
That’s part of why so many legacy CRAC-based rooms get planned for a phased shift toward chilled-water CRAH architecture—or toward rack-level cooling like RDHx—as density requirements increase. (We cover that room-to-rack transition in more depth in RDHx vs. CRAH: Which Wins for High-Density Cooling?) The CRAC isn’t wrong for what it was built for; it just wasn’t built for the load profile a lot of facilities are now carrying.
It’s also worth being precise about where the confusion usually starts. “CRAC” got used as a catch-all term for years, back when most rooms genuinely were self-contained units—before central chilled-water plants were common at every scale. A lot of facilities teams still say “CRAC” out of habit even when the units on the floor are actually CRAHs tied to a central plant. That’s harmless in casual conversation, but it matters the moment you’re specifying a retrofit, ordering replacement parts, or trying to explain to a new hire why the unit in front of them doesn’t have a compressor.
Retrofit Considerations
If you’re evaluating a room today, a few practical questions decide the direction:
- Is there already a chilled-water plant on site, with spare capacity? If yes, CRAH is usually the more efficient path forward.
- Is this a standalone room or small facility with no existing central plant? CRAC’s self-sufficiency avoids the capital cost of building out chilled-water infrastructure for a room that doesn’t need it.
- What’s the realistic density trajectory for this room over the next 3–5 years? If it’s headed toward high-density racks, the room-level choice (CRAC or CRAH) becomes secondary to whether you’ll also need rack-level augmentation like RDHx.
- What does your maintenance team already know how to service? CRAC units may be easier to service as standalone assets because the refrigeration equipment is localized, while CRAH service often requires coordination with the central chilled-water system.
None of these have a universally right answer—they depend on what’s already built, what the budget supports, and where the facility’s density is actually headed.