Colocation Cooling Technologies: A Buyer's Guide
Cooling is the physical constraint that determines whether your racks can actually draw the power you are paying for. A facility may quote 20 kW/rack, but if the hall only has air cooling designed for 8 kW/rack, you will not energize that load. This guide compares the cooling technologies you will encounter in colocation facilities and explains how to verify installed capacity for your density profile.
Why cooling determines usable power
Power and cooling are a matched pair. The kW you commit to must be removable as heat from the hall. If the cooling system cannot extract the heat your servers generate, the facility will either cap your power draw or risk thermal shutdown. Always qualify cooling capacity at the hall level, not the building level—a facility may have high-density cooling in one hall and standard air cooling in another.
Cooling technology comparison
Most enterprise colo deployments (5–15 kW/rack) run in air-cooled halls with hot/cold aisle containment. Above 15 kW/rack, you need rear-door heat exchangers or in-row cooling. Above 40 kW/rack, liquid cooling to the chip or immersion is typically required.
| Technology | Typical density support | When to require it |
|---|---|---|
| Raised-floor air cooling | 5–10 kW/rack | Standard enterprise, low-density cabinets |
| Hot/cold aisle containment | 8–15 kW/rack | Moderate density, most modern halls |
| Rear-door heat exchangers | 15–35 kW/rack | High-density racks without liquid to the chip |
| In-row cooling (CRAC/CRAH adjacent) | 15–40 kW/rack | High-density zones within a mixed hall |
| Direct-to-chip liquid cooling | 40–100+ kW/rack | HPC, AI/ML training, GPU clusters |
| Immersion cooling | 50–200+ kW/rack | Extreme density, specialized workloads |
How to verify installed cooling capacity
Do not accept "we can support high density" without seeing the installed equipment. Future cooling promises do not energize your racks on your go-live date.
Use a site visit checklist to confirm the cooling in the specific hall you were quoted.
- Ask for the hall-level cooling design document, not the building marketing sheet
- Confirm the installed cooling capacity (kW) matches or exceeds your committed IT kW in that specific hall
- Ask whether cooling is N, N+1, or 2N redundant for your hall
- Verify whether high-density cooling is installed today or planned for a future phase
- Request the temperature and humidity setpoints the hall is designed to maintain
- Ask about chiller plant capacity and whether it is shared across halls or dedicated
Matching cooling to your workload
Standard enterprise workloads (web servers, databases, VMs) typically run at 3–8 kW/rack and are well-served by air-cooled containment halls. AI/ML and HPC workloads with GPU racks can push 30–80 kW/rack and require liquid or rear-door cooling from day one.
If you are planning a migration from standard to high-density workloads over your contract term, negotiate expansion rights into a hall that already has high-density cooling installed—not one where it is planned.
For GPU workloads, see cooling for AI training vs inference.
Cooling cost implications
Higher-density halls often command a $/kW premium because the facility has invested more in cooling infrastructure. But on a $/usable-kW basis, high-density cooling can be more efficient—you get more compute per square foot and per cabinet. Normalize TCO on $/usable kW/month, not $/cabinet/month, to compare fairly across density tiers.
Cooling choices also shape the answers to sustainable colocation questions.
Related Guides
Ready to compare providers?
Get matched quotes from multiple facilities or talk through your requirements with an expert.