A closed-loop liquid-cooling system is ready to deploy when the IT equipment, CDU, facility heat rejection, and operating procedures have been verified together. If an equipment interface is unresolved or commissioning records are incomplete, hold deployment until both are documented.
What does closed-loop liquid cooling mean in a data center?
A closed-loop liquid-cooling system circulates coolant through a defined circuit, absorbs heat from equipment, and passes that heat to another part of the cooling system before the fluid returns for another pass. The coolant follows a controlled supply and return path. The loop’s boundaries decide which equipment shares the same fluid, pressure conditions, and maintenance needs.
The term covers several physical arrangements. A self-contained liquid-cooling module circulates coolant inside an assembly supplied as part of a server. A facility-integrated direct-to-chip (DLC) installation connects cold plates in IT equipment to external hoses, rack manifolds, and a technology cooling system (TCS). A coolant distribution unit (CDU) can separate that circuit from the facility water system.
HPE’s DL325 Gen12 documentation distinguishes a self-contained liquid-cooling module with a cold plate, pump, tubing, and inlet-air radiator from direct liquid cooling supplied through rack manifolds.
Some data center enclosure literature, including Rittal’s cooling overview, uses closed loop for air circulating within a cabinet. That heat-transfer path differs from coolant flowing through server cold plates. Rear-door heat exchangers also collect heat at a different point in the rack.
How does heat move through a closed-loop liquid-cooling system?
In a common facility-integrated direct-to-chip design, cold plates collect heat from designated processors or accelerators. Coolant carries that heat through server connections and rack distribution to a CDU. A heat exchanger there transfers it to a separate facility-side circuit, which delivers it to the site’s heat-rejection equipment.
The reference heat path has five stages:
- Cold plates: Heat moves from selected electronic components into coolant flowing through the plates.
- Server and rack distribution: Hoses, quick-disconnect couplings, and supply and return manifolds connect the equipment to the technology cooling circuit.
- CDU: A heat exchanger moves energy between the technology and facility circuits, while pumps and controls hold the conditions the design requires.
- Facility water system (FWS): Facility-side fluid carries the transferred heat toward the cooling plant.
- Final rejection: A dry cooler, cooling tower, chiller-based arrangement, or another plant configuration releases the heat.
In this liquid-to-liquid reference design, the technology and facility fluids stay hydraulically separate across the CDU heat exchanger during normal operation. Heat crosses through the exchanger surfaces, and each fluid keeps circulating in its own circuit.
A liquid-to-air CDU works differently. Its air-side heat exchanger and fans move heat from the liquid circuit into the surrounding air. This arrangement can avoid a facility-water connection, but the existing room cooling system still needs capacity to reject that heat. To compare air cooling, evaporative systems, and liquid loops, use the data center cooling method comparison.
Manufacturers sometimes use primary and secondary for opposite sides of a cooling system. Here, TCS always means the IT-side technology cooling circuit, and FWS means the facility-side water circuit in the liquid-to-liquid reference architecture.
A CDU also manages circulation and operating conditions. Eaton’s CDU assembly includes pumps, a reservoir, a control board, a heat exchanger, filters, flow meters, and pressure transducers, though the inventory varies by product.
The US Department of Energy’s data center cooling guidance describes recirculating water arrangements with separate technology and facility cooling responsibilities.
When is closed-loop cooling practical for a data center retrofit?
Retrofit viability depends on the workload and the infrastructure already in the building. A server may support direct-to-chip cooling while the facility lacks water connections, heat-rejection capacity, pipe routes, or controls to run it.
Cold plates cool selected components. Memory, storage, power equipment, and other parts still need airflow, so the facility must handle that remaining load alongside the new liquid circuit.
In an older data center, the building often decides the project. Distribution paths may cross concrete floors, service areas, existing pipework, and cramped maintenance spaces. New monitoring also has to work with the building management system (BMS).
One operator described this difficulty while considering an expansion from four rear-door-cooled racks to additional higher-density equipment:
We have 4 racks running at 45kW each, cooling handled by rear door heat exchangers on a closed loop. Works fine. Planning to add 4 more racks at 60kW and I am not confident the RDHx approach holds up at that density without either a dedicated chiller loop or a fluid management system we don’t currently have. Trying to figure out what people are actually running in production above 50kW before committing to a cooling path. The conference answer is always direct-to-chip liquid cooling. But DLC in an existing facility that was not designed for it is not a minor project. Getting 4 inch supply/return mains through a building with concrete floors and 12 year old raised floor tiles, adding leak detection that the BMS can actually act on, managing the fluid chemistry that is months of work and serious capital, not a swap. Rear door heat exchangers at 60kW feel marginal. Air supplementation at that density is borderline. DLC is the right answer long term but the retrofit complexity is real.
A feasibility review covers five checks:
- Thermal load: Set the intended rack heat load, the liquid-cooled fraction, and the remaining air-cooling demand.
- Facility capacity: Compare the required supply conditions and heat-rejection capacity with the available plant.
- Physical installation: Check distribution routes, equipment space, isolation points, structural limits, and service access.
- Monitoring integration: Confirm that leak detection and cooling alarms reach the BMS and can trigger the protective response.
- Project execution: Account for construction access, maintenance windows, outage effects, capital, and project owners.
A purpose-built facility with the right water and controls infrastructure may need fewer installation changes for the same IT equipment. A legacy room can need heavy modification before the rack becomes the limiting factor.
If the existing cooling still meets documented thermal and reliability needs, keeping rear-door heat exchangers or improving air support remains a viable alternative.
What must be compatible before the liquid loop is connected?
Liquid-cooling components must share one operating envelope across IT equipment, distribution, the CDU, and facility interfaces. Matching couplings cover only the mechanical side. Coolant chemistry, materials, temperature, pressure, and flow also decide whether the assembled system works.
The Open Compute Project Cold Plate Sub-Project covers cold plates, tubing, manifolds, quick-disconnect components, and CDUs. Its interface and standardization work describes how these parts relate.
Interface compatibility matrix
| Parameter | What must match | Requirement owner or evidence | Consequence of a mismatch |
|---|---|---|---|
| Coolant specification | Approved fluid, additives, and chemistry requirements | Server and CDU OEM documentation | Material degradation or unsupported operating conditions |
| Wetted materials | Metals, polymers, hoses, seals, and fluid compatibility | Component specifications and OEM approval | Corrosion, seal deterioration, or leaks |
| Temperature | Permitted supply/return ranges and thermal conditions | Server, CDU, and facility design documents | Inadequate cooling, condensation risk, or operation beyond limits |
| Pressure and flow | Allowable working conditions, pressure loss, and pumping capacity | OEM hydraulic requirements and system design | Insufficient circulation or overstressed components |
| Quick disconnects | Coupling fit, sealing, and approved connection requirements | QD and equipment specifications | Connection failure or unsuitable interface |
| Fluid cleanliness | Filtration, contamination, and maintenance requirements | OEM fluid-quality and maintenance documents | Fouling or reduced heat transfer |
| Monitoring interfaces | Measurement, controls, and alarm compatibility | CDU, IT, and controls documentation | Incomplete fault visibility or coordination |
An unresolved fluid or material requirement can stop equipment acceptance even if the couplings fit. Document each OEM’s limits and reconcile any conflicts before connecting components.
ASHRAE’s facility liquid-supply temperature classes are W17, W27, W32, W40, W45, and W+. The numbers give the upper supply-temperature limit in degrees Celsius for each class, and W+ covers temperatures above 45°C.
Where condensation is possible, the thermal design must account for the dew point of the surrounding air. Vertiv describes controlling TCS supply temperature above the ambient dew point.
IBM’s Power11 water-cooling requirements and HPE’s server-specific guidance each set fluid-quality and service conditions for their own equipment, and suppliers differ. For decisions involving PUE measurements, WUE measurements, and facility meters, consult the data center energy-efficiency measurement guide.
What can fail in a closed loop, and how should the system detect it?
A closed liquid circuit depends on moving coolant, holding suitable thermal conditions, and keeping fluid inside the system. A failed pump, a lost facility heat sink, a leak, or a failed control action can stop heat removal even when most of the circuit is intact.
Each fault shows up differently. Falling flow or changing differential pressure points to a circulation problem. Rising coolant temperature points to inadequate heat removal. Leaks and fluid contamination need other monitoring or inspection.
Failure signals and response categories
| Failure mode | Signal | Response |
|---|---|---|
| Pump or circulation failure | Abnormal flow, pressure, or coolant temperature | Activate the engineered equipment-protection and escalation sequence |
| Facility heat-rejection loss | Elevated return or supply conditions; plant alarms | Coordinate facility and IT protection under the operating procedure |
| Coolant leakage | Leak detection, fluid-level change, or visible evidence | Follow approved containment, isolation, and electrical-safety procedures |
| Fluid contamination or degradation | Out-of-spec sampling results or filtration concerns | Refer the condition to maintenance and OEM teams |
| Power or controls failure | Missing measurements, controller faults, or unavailable actuation | Apply validated fault-response and power-continuity arrangements |
| Condensation risk | Moisture evidence or an unsuitable thermal condition | Apply approved temperature controls and system limits |
Alarm configuration depends on the installed equipment. Each condition needs a defined indication and a protective action.
Redundant pumps cover only a pump failure. Power, controls, valves, pipework, heat exchangers, and the facility heat sink are shared by both pumps, so they decide whether the redundancy actually works.
HPE publishes equipment-specific coolant-leak procedures, and Schneider Electric treats pumps, leaks, fluid quality, and controls as cooling reliability concerns. For wetted electrical equipment, isolation, and return to service, follow the OEM and site safety procedures.
What should commissioning prove before servers go live?
Commissioning has to show that the finished system holds its specified operating conditions and reacts correctly to faults. Finishing the mechanical install or passing a CDU factory test proves only part of that.
Lawrence Berkeley National Laboratory’s liquid-cooled systems commissioning guidance covers pressure integrity, flow, controls, and representative load simulation. The goal is observable operating evidence for the installation being accepted.
Acceptance rests on five types of evidence:
- Factory documentation: Review OEM test reports, component specifications, and accepted operating limits.
- Installed integrity: Record results for pressure integrity, connections, cleanliness, and preparation under the approved procedures.
- Operating performance: Demonstrate flow distribution and required thermal conditions under representative loads, with suitable instruments and recorded results.
- Integrated controls: Test communication among IT equipment, CDU controls, facility systems, and alarms, including responses to fault scenarios.
- Acceptance and handover: Close identified deficiencies, keep test reports and settings, and collect the required approvals before production acceptance.
In a January 12, 2026 public analysis, Uptime Institute described how CDUs complicate commissioning through interactions between IT and facilities systems. Its public material also covers factory witness testing under relevant conditions.
For example, a CDU may be factory-tested before anyone has demonstrated the combined facility water connection and server heat load. Site acceptance must establish the hydraulic conditions, thermal performance, alarm coordination, and fault behavior of the installed system.
After handover, the accepted test records give operators a baseline for judging later changes in thermal performance and control behavior.
Who owns coolant monitoring, service, and recovery after handover?
The server vendor, rack integrator, CDU supplier, facility operator, and IT tenant may each control a different part of the same cooling installation. Who owns what depends on the system design, equipment agreements, and operating contract. For decisions involving rack power, cooling systems, and network availability, consult the data center infrastructure dependency map.
Equinix’s direct-to-chip system discussion describes coordination among customers, equipment manufacturers, and the colocation provider. Temperature, pressure, flow, and physical connections cross organizational boundaries, so ownership becomes a practical operating issue.
Example responsibility matrix
| Equipment or function | Who usually owns it | Responsibility to assign |
|---|---|---|
| Server cold plates and internal components | IT equipment vendor and customer IT team | Equipment limits, approved service, and fault escalation |
| Rack manifolds and quick disconnects | Rack integrator, customer, or service provider | Connections, isolation, and physical maintenance |
| CDU and technology cooling circuit | CDU supplier, facility team, or designated operator | Pump/control monitoring, service, and fluid management |
| Facility water and heat rejection | Facility operator | Supply conditions, plant operation, and recovery |
| BMS and alarm integration | Facilities controls team and IT operations | Alarm routing, protective action, and escalation |
| Fluid testing and maintenance records | Assigned maintenance owner | Sampling, condition review, filtration, and change control |
For each function, distinguish who performs the work, who approves changes, and who is notified when another party’s equipment is affected. This applies most where a colocation tenant and facility operator share physical cooling interfaces.
Routine operation includes condition monitoring, alarm handling, and scheduled manufacturer-approved fluid-quality checks or component inspection. Service arrangements must cover isolation, work authorization, spare parts, fluid management, and restoration of service.
Does a closed liquid loop save water or energy?
A closed IT coolant circuit recirculates its fluid, but the facility’s final heat-rejection equipment determines how much water the site consumes. An evaporative cooling tower can consume water downstream of a closed TCS.
Dry coolers, evaporative towers, and chiller-based arrangements operate differently, so any comparison must cover the complete plant.
The Department of Energy defines annual power usage effectiveness (PUE) as total facility energy use divided by IT equipment energy use, both in kWh. It defines water usage effectiveness (WUE) as annual site water use in liters divided by annual IT energy use in kWh, expressed as L/kWh.
A defensible savings claim requires the actual heat-rejection arrangement, a consistent facility measurement boundary, and an appropriate baseline.
A practical closed-loop cooling readiness checklist
The go-live decision must rest on approved design documents, integrated test results, and named operating responsibilities.
| Readiness gate | Required evidence | Decision if unresolved |
|---|---|---|
| System boundary | Approved architecture drawing identifying IT, TCS, CDU, and facility interfaces | Hold until the circuit boundaries are defined |
| Thermal feasibility | Load assessment covering the cooling plant and remaining air load | Redesign if the facility cannot meet requirements |
| Equipment compatibility | Approved OEM/interface specification matrix | Hold unresolved requirements; redesign a known incompatibility |
| Failure response | Alarm mapping, protective control logic, and approved procedures | Hold until relevant failure behavior is addressed |
| Commissioning | Accepted integrity, thermal, control, and integrated test records | Hold pending completion or correction |
| Operational ownership | Agreed service boundaries, procedures, and responsibility matrix | Hold until responsibility is assigned |
Accept applies when every critical gate has its required evidence and approval. Hold means a material requirement or record is unresolved. Redesign applies when a known physical, thermal, or equipment mismatch keeps the proposed configuration from meeting its requirements.