A premier electronic component distributor, focus on supplying and solving for new energy vehicle, motorcycle and BESS industries.
info@eshine-cd.com+86 18848211277
Liquid-Cooled AI Data Centers: Power Component Selection Under Higher Thermal Loads
Liquid cooling is changing how high-density AI data centers manage heat. As GPU clusters and server racks demand more power, cooling systems are evolving alongside power distribution and conversion architectures.
However, the adoption of liquid cooling does not automatically mean that every electrical component operates at a lower temperature. The actual thermal conditions depend on where a component is installed, how heat is transferred, and whether the component is directly connected to a liquid-cooling system or remains within an air-cooled enclosure.
For power electronics designers, the important question is not simply whether a data center uses liquid cooling. It is how the cooling architecture changes the operating conditions of the components within the power delivery system.
Different Liquid-Cooling Architectures Create Different Thermal Conditions
Liquid cooling can be implemented at different levels of an AI data center. Each approach affects the surrounding electrical equipment differently.
Direct-to-chip cooling circulates coolant through cold plates attached to heat-generating components such as GPUs and CPUs. It removes heat close to the source, but power supplies and other electrical components may still rely on separate airflow or cooling arrangements.
Rear-door heat exchangers remove heat from the air leaving the server rack. They can reduce the heat load released into the data center space, while the internal cooling conditions of individual power modules still depend on their own design.
Immersion cooling places compatible IT equipment in a dielectric fluid. This approach creates a different thermal and material environment, making fluid compatibility an important consideration for equipment that is directly exposed to the coolant.
These architectures should not be treated as interchangeable. A power component located inside an air-cooled power shelf may experience very different conditions from one integrated into a dedicated liquid-cooled power unit.
The first step in component selection is therefore to understand the cooling boundary: which equipment is liquid-cooled, which remains air-cooled, and how heat moves between the two.
Power Semiconductors: Junction Temperature and Operating Margin
Power semiconductors are essential in the conversion stages of AI data center power systems. MOSFETs, IGBTs, SiC devices and GaN devices may be used depending on the voltage level, switching frequency, efficiency target and converter topology.
Their thermal performance is closely related to electrical operating conditions.
Conduction losses and switching losses generate heat during operation. The resulting junction temperature depends on the power dissipated, the thermal resistance of the device and package, the cooling interface and the surrounding thermal environment.
When evaluating semiconductor devices, engineers should consider:
Maximum junction temperature and recommended operating range
Conduction and switching losses at the intended operating point
Thermal resistance from junction to case and from case to ambient or coolant interface
Switching frequency and load profile
Transient thermal behavior during rapid load changes
Required electrical isolation and mounting conditions
A liquid-cooled system may improve heat removal at a particular interface, but this benefit depends on the actual thermal path. If the semiconductor is not connected to the liquid-cooling loop, its operating temperature may still be governed by airflow, heatsink design and enclosure conditions.
The selected device should therefore be evaluated using the complete thermal path rather than the cooling method used elsewhere in the rack.
Inductors and Magnetic Components: Losses Still Matter
Magnetic components remain important in power conversion and voltage regulation circuits, including power supplies and DC-DC converter stages.
Inductor losses generally include winding losses and core losses. Both are affected by the electrical operating point and component construction.
At high current, winding resistance contributes to copper losses. Core losses depend on factors such as switching frequency, flux density and core material. These losses contribute to the component's temperature rise and can affect its electrical characteristics over time.
For AI server power applications, engineers should review:
Rated and saturation current
DC resistance and winding losses
Core material and operating frequency
Inductance variation under DC bias
Temperature rise at the expected load
Thermal characteristics of the package and mounting arrangement
Liquid cooling elsewhere in the system does not eliminate these considerations. If an inductor remains in an air-cooled section, its thermal performance still depends on local airflow and heat transfer.
For compact power conversion assemblies, the spacing between magnetic components, switching devices and other heat sources can also influence the final temperature distribution.
Capacitors: Temperature, Ripple Current and Service Life
Capacitors are used throughout power conversion and distribution circuits for filtering, energy storage and voltage stabilization.
Their operating conditions can vary significantly depending on their position in the power architecture.
DC-link capacitors, for example, may experience ripple current and electrical stress associated with converter operation. Output capacitors may be exposed to different ripple profiles and transient conditions.
When selecting capacitors for high-density power systems, engineers should evaluate:
Rated voltage and capacitance
Ripple current capability
Equivalent series resistance (ESR)
Operating temperature range
Expected lifetime under actual electrical and thermal conditions
Mechanical dimensions and mounting requirements
Temperature is an important factor in capacitor service life, but it should not be assessed independently of ripple current and voltage stress.
A cooling system that reduces the temperature of nearby equipment does not necessarily reduce the internal heating of a capacitor caused by ripple current. The component's own electrical losses and thermal path still need to be considered.
Connectors and Interconnects: Local Temperature Rise
High-power AI systems require compact electrical connections across power shelves, server racks and distribution assemblies.
Connectors, terminals and busbar interfaces must carry the intended current while maintaining acceptable contact resistance and mechanical stability.
Even when the overall enclosure temperature is controlled, local heating can occur at electrical connection points. Contact resistance, current loading, conductor size, mating quality and installation conditions all influence temperature rise.
Component selection should account for:
Rated current under the actual installation conditions
Contact resistance and associated power losses
Temperature rise at the intended current
Wire or busbar compatibility
Vibration and mechanical retention
Insulation and creepage requirements
Environmental protection and maintenance conditions
In liquid-cooled equipment, the position of connectors relative to cold plates, coolant lines and other thermal interfaces also deserves attention.
The presence of coolant pipes or manifolds near electrical connections makes installation layout and service access important parts of the design review.
Protection Components Must Match Their Actual Operating Conditions
Fuses and other protection devices are selected according to electrical fault requirements, not simply the nominal power rating of a system.
For high-power DC applications, engineers need to verify rated voltage, continuous current, breaking capacity, time-current characteristics and coordination with the wider protection system.
Thermal conditions also affect the way protection components operate. Ambient temperature, enclosure temperature, mounting arrangement and heat generated by adjacent components may influence the applicable current-carrying capability.
A fuse selected using its nominal current rating alone may not be suitable for the actual installation conditions. Manufacturer derating guidance and the intended operating environment should be reviewed.
The same principle applies to switching devices and other protection-related components: electrical ratings, thermal conditions and installation requirements need to be considered together.
Cooling Infrastructure Also Creates Electrical Design Considerations
Liquid cooling introduces additional equipment and interfaces into the data center environment.
Depending on the architecture, the cooling system may include coolant distribution units (CDUs), pumps, valves, sensors, monitoring devices and control circuits.
These devices create auxiliary electrical loads and introduce additional requirements for power supply, switching, sensing and communication.
For power electronics and control circuit designers, several details deserve attention:
Power requirements of pumps, CDUs and control equipment
Supply stability for monitoring and control devices
Temperature and flow-rate sensing interfaces
Communication between cooling controls and facility monitoring systems
Electrical protection for auxiliary loads
Equipment placement and access for maintenance
Cooling performance and electrical performance are closely connected. A cooling system requires reliable power and control, while the power equipment depends on suitable thermal conditions to operate within its specified limits.
This relationship is particularly important in high-density AI facilities, where power and cooling infrastructure must be planned together rather than developed as completely separate systems.
A Practical Review Before Finalizing Component Selection
When evaluating power components for a liquid-cooled AI data center, engineers should establish the actual operating conditions before comparing individual part numbers.
A useful review can include the following points:
Identify the cooling architecture. Determine whether the system uses direct-to-chip cooling, rear-door heat exchangers, immersion cooling or a combination of approaches.
Map component locations. Identify which power components are directly liquid-cooled, indirectly affected by the cooling system or still dependent on local airflow.
Define electrical operating conditions. Confirm voltage, continuous current, peak current, switching frequency and expected load variation.
Evaluate thermal limits. Review junction temperature, temperature rise, thermal resistance and manufacturer derating requirements.
Check material and environmental compatibility. For equipment exposed to coolant or dielectric fluids, verify compatibility with the relevant materials and operating conditions.
Review electrical connections. Check connector ratings, contact resistance, insulation, installation space and temperature rise.
Consider auxiliary equipment. Include the power, control, sensing and communication requirements of cooling infrastructure.
Verify the complete assembly. Component specifications should be assessed within the final enclosure and cooling arrangement, not only under isolated datasheet conditions.
This process helps prevent a common design mistake: assuming that the presence of liquid cooling automatically resolves thermal limitations across the entire power system.
Component Selection Within an Integrated Power and Cooling Design
Liquid cooling is becoming an important part of high-density AI data center infrastructure, but it does not replace the need for careful electrical and thermal engineering.
Semiconductors, inductors, capacitors, connectors, switching devices and protection components each have different operating characteristics. Their suitability depends on the electrical duty, local thermal conditions, installation environment and interaction with surrounding equipment.
As power conversion and cooling systems become more closely integrated, component selection needs to reflect the actual conditions at each stage of the power delivery chain.
ESHINE supports electronic component sourcing and technical selection for power electronics and energy-related applications. For AI server and data center power projects, component requirements can be evaluated according to electrical parameters, thermal conditions, package constraints and application needs.
Sharing the intended circuit function, operating parameters and installation environment helps narrow down suitable component options for further technical review.
+86 28 86519933