Liquid cooling What is liquid cooling?
Liquid cooling removes heat from servers and data center equipment. Coolant is circulated near or through heat sources within IT equipment. Because liquids can carry heat much more effectively than air, liquid cooling can help data centers manage highpower processors, graphics processing units (GPUs), accelerators, and dense computing (e.g., HPC and AI) systems that often exceed the limits of conventional air cooling.
Time to read: 8 minutes | Published: October 20, 2026
Table of Contents
Key Takeaways
- Liquid cooling removes heat more effectively than air cooling and can support the high-power CPUs, GPUs, and accelerators used for AI and HPC.
- Direct liquid cooling, closed-loop cooling (liquid-assisted air cooling), rear door heat exchangers, adaptive rack cooling, and immersion cooling address different server, rack, and facility requirements.
- The right approach depends on workload demand, rack density, facility readiness, efficiency goals, serviceability, and total cost of ownership.
Why is liquid cooling important for modern data centers?
Liquid cooling is increasingly important because AI and compute-intensive workloads generate much more heat than ever before. Modern data centers must remove more concentrated heat for reliable operation, system availability, and sustained performance. AI, HPC, and other accelerated workloads place powerful demands on IT, creating cooling needs that conventional approaches may not handle.
Conventional air cooling remains appropriate for many data center environments with rack densities under 20kW. However, as chip power and rack density rise, fans and room-level HVAC must move greater volumes of cooled air than many data centers and IT systems can provide. Liquid cooling addresses this constraint by removing heat more efficiently and closer to high-power components.
By cooling capacity, liquid cooling can help data centers add computing capacity within available floor space and cooling power, maintain more consistent component temperatures, and prepare future processor generations. Depending on the system and facility design, it may also reduce the power envelopes required for IT, meaning more systems in the same rack space by as much as 25%
How does liquid cooling work?
Liquid cooling works by capturing heat from servers. The heat is transported from the heatgenerating components (e.g., CPU, GPU, and sometimes memory, storage, fabric, and power) with a liquid coolant. The coolant circulates through cold plates on the components and transfers it through a heat exchanger and ultimately to facility cooling equipment for dissipation outside of the data center.
Servers generate this heat as they convert electrical energy into computing work. The process uses heat-transfer interfaces, coolant loops, pumps, manifolds, heat exchangers, and facility cooling equipment.
Heat capture
Heat capture occurs when liquid absorbs heat close to the components that generate it. In direct liquid cooling, cold plates are attached to high-heat components such as CPUs and GPUs, and coolant flows through channels inside the plates without coming into direct contact with the electronics. Other types of liquid cooling can capture heat from server exhaust air or immerse compatible hardware in a dielectric fluid.
Heat transport
Heat is transported out of the server components, through hoses. The hoses connect into a manifold in the back of the rack to consolidate all the coolant lines into one flow. Once the heated coolant is consolidated in the manifold, it is piped to a coolant distribution unit. (CDU). The CDU has two primary purposes. The first is to remove heat from the incoming coolant. The heat is pumped out of the CDU into the facility water loop to move the heat outside of the data center. Secondly, the CDU must also maintain the right flow, pressure, and temperature back to the IT rack.
Heat dissipation
Heat dissipation occurs when a heat exchanger moves captured heat from the IT/technology loop to the facility water loop. The facility water loop then rejects that heat through equipment outside of the data center such as dry coolers, cooling towers, or chillers, depending on local design and climate. In suitable facilities, the heated liquid may also be reused to heat nearby residential spaces or industrial processes.
What are the different types of liquid cooling?
Liquid cooling includes several different approaches that capture heat at the component, server, rack, or room boundary level. Organizations may use one method or combine liquid and air cooling in a hybrid design.
Direct liquid cooling
Direct liquid cooling, sometimes called direct-to-chip cooling, sends coolant to cold plates mounted on high-heat components such as CPUs and GPUs. It is commonly used in AI and HPC systems because it removes a large share of heat from the source.
Some direct liquid-cooled systems still use fans for memory, storage, power supplies, networking, or other components, while fanless DLC (fully liquid-cooled architectures) can eliminate server fans. Heat is removed from the data center through facility water, reducing heat load on a data center when compared to air-cooling.
Heat flow: Component > manifold > CDU > facility water > Outdoor or rooftop cooling
Closed-loop liquid cooling
Closed-loop liquid cooling acts much like a radiator in your car. Liquid coolant is pumped through cold plates on the components to a radiator inside the server. The radiator is cooled by fans, and the cooled liquid cycles back. It can provide more effective processor cooling from individual components without requiring a facility-water connection. Because the system still rejects heat back into the data center, the data center must retain enough air-cooling capacity for the server and its radiator exhaust. Heat load on the data center is unchanged from standard air cooling.
Heat flow: Component > Radiator > Fan > Hot aisle within the data center
Rear door heat exchangers
A rear door heat exchanger (RDHX) cools server exhaust. It is mounted to the back of the rack. RDHXs contain liquid-cooled heat exchanging coils that neutralize the heat from the server exhaust. They can reduce or neutralize a rack's contribution to room heat without bringing coolant directly to the chips.
This rack-level approach can support existing air-cooled equipment and can also complement direct liquid cooling. Heat load on the data center is reduced when compared to air-cooling alone.
Heat flow: air through servers > fan > RDHX > facility water > outdoor or rooftop cooling
Adaptive rack cooling systems
Adaptive rack cooling systems (ARCS) provide cooled air into a system of up to 4 connected IT racks around a central heat exchanger. The heat exchanger contains flow and control equipment to dose the cooled air across the system. They can add cooling capacity for dense deployments while supporting air-cooled and partially liquid-cooled systems.
This approach can be useful when operators need to isolate rack heat without redesigning the entire data hall. ARCS reduced data center heat load with respect to air-cooling alone.
Heat flow: cool air from ARCS > servers > heat exchanger in ARCS > facility water > outdoor or rooftop cooling
Immersion cooling
Immersion cooling submerges compatible IT equipment in an electrically nonconductive dielectric fluid. In single-phase systems, the fluid remains liquid and transfers heat to a heat exchanger. Tank weight, floor loading, fluid handling, component compatibility, maintenance access, and end-of-life processes can make immersion challenging for many enterprise facilities.
What is the difference between liquid cooling versus air cooling?
The main difference between air cooling and liquid cooling is how they move heat away from data center equipment. Air cooling uses fans that move heat, while liquid cooling circulates coolant closer to heat-generating components to remove concentrated heat more effectively.
Liquid removes heat thousands of times more efficiently than the same volume of air. Also, liquid can be focused directly on the heat sources, instead of relying on fans which may be meters away. While air cooling remains suitable for many general-purpose environments, liquid cooling adoption rates are rapidly increasing to suit high-density AI, HPC, and accelerated computing systems.
| Comparison area | Air cooling | Liquid cooling |
|---|---|---|
| Heat transfer | Moves heat with fans. | Moves heat with liquid coolant, often closer to the heat source. |
| Cooling fluid | Requires high volume and high flow. | Can carry thousand times more heat in the same volume. |
| Rack density | Best suited to lower or moderate heat loads, subject to facility design. | Better suited to higher-performance systems with concentrated heat loads. |
| Cooling efficiency | Requires fans and air-moving equipment; efficiency can decline as airflow demand rises. | Can reduce air-moving energy and transfer heat efficiently, but results depend on the full system and facility. |
| Special facility requirements | Heat capacity. | Facility water. |
| Maintenance | Filter, fan, airflow, and HVAC maintenance. | Inspect connections, coolant quality, filtration, test leak detection system. |
| Typical workloads | General enterprise computing and lowerdensity deployments. | AI, HPC, simulation, and other compute-intensive workloads. |
| Cooling scalability | Can be straightforward at modest densities but may require significant roomlevel expansion as loads grow. | Can scale to high-density and high heat outputs. |
| Capital expense | Often uses familiar infrastructure. | May require higher initial investment and specialized planning. |
| Operating expenses | Cooling costs may be 20-25% higher than liquid cooling. | Can reduce operating expenses by more than 80% |
What are the benefits of liquid cooling?
The main benefits of liquid cooling include more effective heat removal, compatibility with highest performance GPU and CPUs, improved performance and reliability, greater cooling efficiency, and higher rack density to better support modern workloads. It can also help reduce data center space requirements and create opportunities to reuse captured heat.
Consider these benefits:
- Improved performance and reliability—Removing heat effectively helps processors remain within design limits and can reduce thermal throttling, sustaining performance and dependable operation during long-running workloads.
- Greater cooling efficiency—Liquid can transport thousands of times more heat than the same volume of air. By capturing heat near the component, a liquidcooling system can reduce the fan power and air movement needed to maintain safe temperatures.
- Higher rack density—Liquid cooling can support more computing capacity in each rack, allowing organizations to use limited floor space and power more productively. The achievable density depends on the IT equipment, cooling architecture, power distribution, and facility design.
- Support for AI and HPC workloads—High-power CPUs, GPUs, and accelerators often require liquid cooling as the heat loads produced are too high for air cooling alone. Liquid cooling is often the only path to deploy those accelerators.
- Reduced data center footprint—By consolidating liquid-cooled servers into fewer racks than their air-cooled counterparts, organizations will reduce the rack space needed. This can be especially valuable in constrained facilities, colocation environments, and modular deployments.
- Heat reuse opportunities—Liquid systems can produce a concentrated stream of warm water that may be reused for building heating, industrial process water, and aquaculture.
What are the challenges of liquid cooling?
The main challenges of liquid cooling include facility readiness, process changes, and cost differences. When deploying liquid cooling, consider the following:
- Infrastructure planning and facility readiness—Liquid cooling will likely require close coordination across IT, facilities, operations, and service teams. Organizations may need to assess water availability and quality, pipe routes, CDU placement, heat-rejection capacity, floor loading, controls, redundancy, and compatibility with selected servers and racks.
- Process changes—Procedures may need to be updated to include coolant quality, leak detection and response, filter inspection, preventive maintenance, and safe servicing.
- Lifecycle cost and return on investment—Costs should be evaluated across the full lifecycle and space requirements rather than by initial equipment purchase alone. The business case should consider infrastructure investment, energy use, available space, deployment speed, serviceability, workload value, resilience, future expansion, and the cost of not deploying compute-intensive workloads.
How do organizations determine when liquid cooling is needed?
Liquid cooling is most often adopted to accommodate new, high-performance GPUs and CPUs that require liquid cooling. The decision often comes down to, "Can I fit the equipment needed in the space, power, and cooling constraints that I have with the performance that I need?"
Planning a liquid cooling deployment
To plan for liquid cooling, first assess current and future IT loads, along with the expected rack densities, available electrical capacity, cooling capacity, and growth. A new facility can be designed around liquid cooling from the outset, while a retrofit may require phased updates to piping, heat rejection, controls, floor loading, and service access.
Design teams should specify the requirements for the facility water loop and secondary technology IT cooling loop, CDUs placement and redundancy, manifold and connection standards, coolant specifications, monitoring, alarms, and leak-response procedures. Also plan for how air-cooled and liquid-cooled systems will be utilized.
The best practice for deployment planning brings together IT, facilities, mechanical/electrical, and vendors early. Commissioning, documentation, staff training, spare parts, preventive maintenance, and lifecycle support should be addressed before production workloads are moved to the system.
HPE and liquid cooling
HPE brings decades of liquid-cooling innovations and expertise from high-performance computing. Today, HPE applies that experience to AI, HPC, and converged workloads through technologies that span the server, networking, rack, and facility environments.
HPE liquid cooling technologies
HPE's portfolio includes direct liquid cooling, closed-loop liquid cooling for selected servers, rear door heat exchangers, and the HPE Adaptive Rack Cooling System. These solutions address different requirements:
- Direct liquid cooling captures heat at high-power components.
- Closed-loop modules allow for some use of advanced silicon.
- Rear doors and adaptive rack cooling augments air-cooled system remove hot exhaust from the data center.
For the most demanding large-scale AI and HPC systems, HPE also offers fanless direct liquid cooling architecture that cools system components without fans. Other environments may use hybrid designs in which liquid removes most processor heat while air cools remaining components.
Liquid cooling for AI factories and HPC
HPE integrates compute, networking, racks, power, cooling, and services to support highdensity AI factories and HPC environments. The appropriate architecture depends on workload scale, processor and accelerator choices, deployment location, efficiency goals, and facility readiness—not on a single product or cooling method.
HPE services and expertise
Liquid cooling can change both IT and facility requirements. HPE Services can help organizations completely assess sites, plan data centers and cooling designs, integrate and deploy infrastructure, monitor cooling performance, establish maintenance practices, and support systems across their lifecycle.
FAQs
Why is liquid cooling important for AI workloads?
AI training and large-scale inference can keep dense groups of GPUs and other accelerators operating at high utilization. This creates concentrated heat that can exceed the practical limits of air cooling. Liquid cooling captures heat closer to the processors, helping support sustained performance, higher rack density, and efficient operation.
At what rack density does liquid cooling become necessary?
Typically, liquid cooling requirements may start at rack densities as low as 20kW per rack. Additionally, some high-powered chips also require liquid cooling, regardless of rack density. Organizations should evaluate the complete thermal and power environment.
Can an existing data center be retrofitted for liquid cooling?
Yes, many facilities can add liquid cooling in phases. Often, facility water already provides cooling to the computer room for air handling or computer room air conditioning systems (CRAH/CRAC). A retrofit assessment should examine piping routes, CDUs, heat rejection, water quality, floor loading, power, controls, and service access. Rear door heat exchangers, HPE ARCS, and modular capacity may help where a full facility retrofit is not practical.
Does liquid cooling eliminate the need for air cooling?
Not always. Many direct-to-chip systems remove heat from CPUs and GPUs while fans continue to cool memory, storage, networking, and power components. Fully liquid-cooled architectures can remove server fans, but hybrid liquid-and-air designs remain common.
What is the quickest way to deploy liquid cooling?
The fastest option depends on site readiness, thermal capacity, power capacity, and required scale. A liquid-ready colocation facility or prefabricated modular data center may reduce construction and integration work, while rear door or HPE ARCS may add capacity to an existing site more quickly. A site assessment is still needed to verify power, water, controls, network, and service requirements. And liquid-cooled capacity is rapidly growing in colocation sites.
Does liquid cooling improve energy efficiency?
Liquid cooling often improves efficiency, as the cooling components with liquid are more directly efficient than blowing air throughout the data center. Liquid carries heat more effectively and may reduce the energy used by server fans and room-level air movement. Actual savings depend on coolant temperatures, pump energy, heat rejection, climate, workload, and the design of the entire facility. In holistically designed liquid-cooled data centers, operating power costs can be reduced by over 80%
Is liquid cooling safe for enterprise servers?
Purpose-built systems use sealed connections, compatible materials, monitoring, alarms, and leak-response controls to manage risk. Safe operation also depends on correct installation, coolant quality, inspection, maintenance, and trained service procedures. HPE has innovated many aspects of liquid cooling over the last 50 years to optimize cooling efficiency and reliability.
What maintenance does a liquid cooling system require?
Maintenance may include checking coolant chemistry and cleanliness, pumps, filters, heat exchangers, hoses, connectors, manifolds, sensors, and leak-detection systems. Closed-loop modules and facility-connected systems have different service requirements. Operators should define inspection intervals, alert-handling procedures, spare parts, and responsibilities before deployment.