As AI and high-performance computing (HPC) push rack densities higher, liquid cooling is becoming an increasingly important part of the data center thermal management conversation. But that does not mean air cooling is going away.
For many data centers, the path forward will involve both.
Facilities may need to support traditional air-cooled infrastructure alongside high-density AI systems. Even within servers that use direct-to-chip liquid cooling, not every heat-generating component is necessarily connected to the liquid loop. Memory, storage, power supplies, networking equipment and other components can continue rejecting heat into the surrounding air.
That makes hybrid cooling more than a transitional approach. It is a thermal strategy designed around a simple principle: use the right cooling method for each part of the heat load while making sure the entire system works together.
Why Hybrid Cooling Is Becoming the Practical Path Forward
Data centers rarely operate at a single, uniform rack density. A facility may have relatively low-density storage and networking equipment, traditional enterprise compute and increasingly dense AI or HPC clusters operating within the same environment.
Replacing an established air-cooling architecture simply because some racks require liquid cooling may be neither practical nor necessary. This is especially true in existing facilities where operators must introduce higher-density equipment without rebuilding the entire data hall.
There is also no universal rack-density threshold that determines when an environment must move to liquid cooling. Server architecture, component heat flux, rack configuration, airflow and facility conditions all influence what a given cooling approach can support. The more useful question is where heat is being generated and how effectively it can be removed.
The same challenge applies in colocation and HPC environments, where equipment, workloads and rack densities can vary considerably. Some systems may require direct liquid cooling, while others remain well within the capabilities of air.
A hybrid approach gives operators the flexibility to match cooling resources to actual thermal requirements. A hybrid strategy is more than operating air- and liquid-cooled equipment in the same facility. It requires intentionally determining which portions of the heat load each cooling method will address—and how those systems will work together. Instead of treating air and liquid as competing technologies, the objective becomes determining where each can perform most effectively.
Understand What Liquid Cooling Does—and What It Doesn't
Direct-to-chip cooling addresses one of the fundamental challenges created by today's high-density computing: removing concentrated heat directly from components such as CPUs and GPUs.
By circulating coolant through cold plates attached to those components, heat can be captured much closer to where it is generated rather than relying solely on server fans and room-level airflow to remove it.
But removing heat from the highest-powered components does not necessarily remove the entire server heat load.
Other components may still be air cooled, leaving residual heat that must be removed from the rack and surrounding environment. The percentage will vary by equipment, but the underlying design consideration remains the same: introducing direct liquid cooling does not automatically eliminate the air side of the thermal equation.
A hybrid cooling strategy therefore starts with understanding the complete thermal profile of the equipment, including how much heat is captured by liquid and how much remains in the air. Liquid cooling does not eliminate the need for airflow management. It changes the heat load that airflow management is responsible for.
Give Air Cooling a More Targeted Role
In a hybrid environment, air cooling does not disappear. Its role changes.
Traditional air cooling may continue to support lower-density equipment throughout the facility while also addressing residual airborne heat from partially liquid-cooled systems.
That makes good airflow management just as relevant in a hybrid environment. Bypass airflow through open rack-mount spaces and cable openings can waste available cooling capacity, while recirculation can bring heated exhaust air back toward equipment inlets. Cable congestion can further restrict airflow paths and create localized thermal issues.
Uncontrolled bypass airflow, recirculation and cable congestion can still affect equipment inlet temperatures and cooling performance.
Blanking panels, air dams, seals and other airflow management techniques can help direct conditioned air through the equipment where it can do useful work. Containment can further separate supply and exhaust air, but it cannot compensate for poor airflow conditions within the cabinet. Addressing bypass paths, recirculation and cable obstructions first helps operators get more from available air-cooling capacity before determining where additional cooling approaches are needed.
Bring Cooling Closer to the Heat Load
As rack densities increase, another consideration is where heat is captured.
Traditional room-based cooling requires heat to move from the IT equipment into the data hall before it can be removed. Bringing thermal management closer to the rack can intercept that heat earlier and reduce the burden placed on room-level cooling systems.
Rear-door heat exchangers are one example. Installed at the rear of the cabinet, they use a liquid-filled heat exchanger to capture heat from server exhaust air before it enters the room.
Rack-level thermal management can also complement direct-to-chip cooling as part of a hybrid architecture. Direct liquid cooling can address concentrated heat from processors, while technologies such as rear-door heat exchangers can capture residual airborne heat from components that are not connected to the liquid loop. Together, the two approaches address different portions of the same rack-level heat load.
For mixed-density environments, that combination can provide another degree of flexibility. Operators can apply different cooling approaches where they are needed rather than designing every rack around the facility's highest-density equipment.
Design the Cabinet as Part of the Thermal Architecture
As cooling moves closer to IT equipment, decisions that once could be considered separately increasingly converge at the cabinet.
The cabinet must support the equipment itself while accommodating airflow, power distribution, high-density cabling, environmental monitoring and, where applicable, connections to liquid-cooling infrastructure. Each can affect thermal performance.
Cable congestion, for example, can restrict server airflow. Poor airflow management can allow hot exhaust air to recirculate toward equipment inlets. Increasing power density can create additional heat while simultaneously increasing the amount of power and cabling infrastructure that must fit within the cabinet.
Visibility also becomes more important. Temperature and humidity monitoring at the cabinet level can help operators understand changing environmental conditions and identify developing thermal issues before they threaten equipment performance or availability.
For that reason, the cabinet should not be treated simply as the enclosure around a cooling strategy. It is part of the thermal architecture itself.
Planning the cabinet, power distribution, airflow management, cabling, monitoring and cooling infrastructure together can help avoid solving one high-density challenge only to create another.
Build for a Cooling Strategy That Will Continue to Change
Few data centers will transition from predominantly air-cooled infrastructure to liquid-cooled infrastructure all at once.
Instead, cooling requirements are likely to evolve alongside equipment refresh cycles, new AI deployments and increasing rack densities. Different portions of the facility may move at different speeds.
That also creates an opportunity to introduce new cooling approaches incrementally. Operators can pilot a hybrid configuration at the rack or row level, measure thermal performance under real workloads and use those results to inform broader deployment. What works can then become a repeatable configuration rather than requiring each higher-density deployment to be engineered from scratch.
That makes flexibility one of the most important characteristics of a hybrid cooling strategy.
Rather than designing around a single cooling technology, operators can consider how their physical infrastructure will accommodate changing combinations of air cooling, rack-level heat removal and direct liquid cooling over time.
The objective is not to predict exactly what every future rack will require. It is to create an infrastructure foundation that can adapt as those requirements change.
The Future of Cooling Is Integrated
Higher rack densities are changing the way data centers approach thermal management, but the answer is unlikely to be a wholesale replacement of air with liquid.
Hybrid cooling allows operators to address concentrated heat where liquid cooling makes sense while continuing to use air effectively for the rest of the thermal load. The strongest strategies consider how those approaches interact from the component and cabinet to the data hall.
Chatsworth Products (CPI) helps data center operators approach high-density cooling as part of the broader physical infrastructure strategy—bringing cabinet design, airflow management, power distribution, monitoring and thermal management together around the needs of the IT equipment.
For a deeper look at the cooling approaches available for increasingly dense AI environments, download the white paper, Solving the AI Cooling Challenge: Lessons from the Front Lines of Data Center Cooling.