Abstract liquid swirls in vibrant yellow and deep black, creating a dynamic and fluid visual effect.

Liquid Cooling Architecture Types

Liquid cooling technologies fall into two primary categories:

  • Direct-to-chip cooling
  • Immersion cooling

Within these categories, there are five main architectures, each with different performance, efficiency, and deployment characteristics.

In practice, many deployments combine liquid cooling with existing air infrastructure, creating hybrid architectures that balance performance and practicality.

Direct-to-Chip Liquid Cooling (Single-Phase)

Overview  

Direct-to-chip liquid cooling (single-phase) delivers coolant directly to the highest heat-generating components – such as CPUs and GPUs – using cold plates. The coolant remains in a liquid state throughout the process, absorbing heat and transferring it away efficiently.

How It Works  

Coolant circulates through cold plates mounted on processors, where it absorbs heat at the source. The warmed liquid is then transported through a closed-loop system to a CDU or heat exchanger, where the heat is removed before the cooled liquid is recirculated.  

Residual heat from other components is managed using reduced airflow within the server.  

Key Characteristics 

  • Liquid coolant remains in a single phase (no boiling or phase change) 
  • Heat is removed directly at the chip level 
  • Closed-loop system connected to a CDU 
  • Compatible with standard rack-based server designs 

Benefits 

  • High efficiency heat removal
    Targets the highest heat loads directly at the source 
  • Supports high-density compute
    Enables significantly higher rack power densities than air cooling 
  • Simplified system design
    More straightforward to design, operate, and maintain compared to two-phase systems 
  • Ease of integration
    Works with existing data centre infrastructure and server form factors 

Considerations 

  • Some airflow is still required for non-liquid-cooled components 
  • Requires integration with a CDU and facility cooling loop 
  • Thermal performance depends on system design and flow management 

Best For 

  • AI and machine learning workloads 
  • High-performance computing (HPC) 
  • Retrofit and hybrid cooling environments 
  • Data centres transitioning from air to liquid cooling 

Key Takeaway  

Direct-to-chip single-phase cooling offers the most practical and widely adopted path to high-density liquid cooling – balancing performance, efficiency, and ease of deployment, particularly in hybrid air + liquid environments.

Direct-to-Chip Liquid Cooling (Two-Phase) 

Overview  

Direct-to-chip liquid cooling (two-phase) delivers coolant directly to high-heat components such as CPUs and GPUs, but unlike single-phase systems, the coolant undergoes a phase change – boiling into vapor as it absorbs heat.  

This phase change enables highly efficient heat removal at very high power densities.  

How It Works  

A refrigerant is circulated to cold plates mounted on processors. As the fluid absorbs heat, it vaporizes at the chip surface, removing large amounts of thermal energy.  

The vapor is then transported to a condenser or heat exchanger, where it is cooled, condensed back into liquid, and returned to the system in a continuous cycle.  

Key Characteristics 

  • Coolant undergoes phase change (liquid to vapor to liquid) 
  • Very high heat transfer capability due to latent heat absorption 
  • Uses engineered refrigerant 
  • Requires tightly controlled system conditions (pressure, temperature) 

Benefits 

  • Exceptional thermal performance
    Removes significantly more heat than single-phase systems 
  • Enables extreme density
    Supports next-generation AI and HPC workloads at very high power levels 
  • Efficient heat transfer
    Phase change allows large amounts of heat to be removed with relatively low flow rates 

Considerations 

  • More complex system design and control requirements 
  • Higher cost due to specialized fluids and components 
  • Requires careful management of pressure, sealing, and condensation 
  • Less common than single-phase in today’s data centre deployments

Best For 

  • Ultra-high-density AI workloads 
  • Advanced HPC environments 
  • Deployments where maximum thermal performance is required 
  • Purpose-built or greenfield facilities 

Key Takeaway  

Two-phase direct-to-chip cooling delivers the highest level of thermal performance, but with increased complexity – making it best suited for specialized, high-density environments where performance outweighs system simplicity.

Immersion Liquid Cooling (Tank, Single-Phase) 

Overview  

Immersion liquid cooling (tank, single-phase) submerges entire servers or IT hardware directly into a bath of dielectric fluid. The fluid absorbs heat from all components simultaneously, providing highly efficient and uniform cooling.  

The coolant remains in a liquid state throughout the process.  

How It Works  

Servers are placed vertically or horizontally within an open or semi-enclosed tank filled with dielectric fluid. As components generate heat, it is transferred directly into the surrounding liquid.  

The warmed fluid is circulated through a heat exchanger, where heat is removed before the cooled liquid is returned to the tank, maintaining a continuous cooling cycle.  

Key Characteristics 

  • IT equipment is fully submerged in dielectric fluid 
  • Fluid is in direct contact with all components 
  • Single-phase system (no boiling or vaporization) 
  • Typically uses open bath or tank-based designs 
  • No need for server-level fans 

Benefits 

  • Highly efficient heat removal
    Direct contact cooling enables strong thermal performance across all components 
  • Uniform temperature distribution
    Eliminates hotspots by cooling all components simultaneously 
  • Reduced airflow requirements
    Removes the need for internal fans and reduces reliance on air cooling 
  • High-density potential
    Supports compact, high-density system configurations 

Considerations 

  • Requires purpose-built infrastructure and tank systems 
  • Serviceability differs significantly from traditional rack environments 
  • Handling and maintenance of fluid must be carefully managed 
  • Integration with standard IT hardware may require modification or validation 

Best For 

  • High-density research-based applications 
  • Controlled environments designed for immersion 
  • Deployments prioritizing thermal efficiency over standardization 
  • Facilities optimized for immersion-based infrastructure 

Key Takeaway  

Single-phase immersion cooling in a tank-based system delivers efficient, uniform cooling across entire servers – but requires a fundamentally different approach to infrastructure, operations, and serviceability.

Immersion Liquid Cooling (Tank, Two-Phase) 

Overview  

Immersion liquid cooling (tank, two-phase) fully submerges IT hardware in a dielectric fluid that boils as it absorbs heat. This phase change enables extremely efficient heat removal, making it one of the most thermally effective cooling methods available.  

How It Works  

IT equipment is submerged in a tank filled with engineered dielectric fluid. As components generate heat, the fluid at their surface boils, converting from liquid to vapor and absorbing large amounts of thermal energy.  

The vapor rises and comes into contact with a condensing coil or heat exchanger, where it cools and condenses back into liquid. The liquid then returns to the tank, creating a continuous, passive cooling cycle.

Key Characteristics 

  • IT hardware is fully submerged in dielectric fluid 
  • Fluid undergoes phase change (liquid to vapor to liquid) 
  • Heat is removed through boiling and condensation 
  • Uses engineered fluids designed for controlled phase change 
  • Minimal or no mechanical pumping required within the tank 

Benefits 

  • Exceptional thermal efficiency
    Phase change enables very high heat transfer capability 
  • Supports extreme densities
    Ideal for ultra-high-performance AI and HPC workloads 
  • Uniform cooling across all components
    Eliminates hotspots through full immersion 
  • Reduced mechanical complexity at IT level
    Little to no need for fans or internal airflow 

Considerations 

  • Requires specialized fluids and containment systems 
  • More complex system design and thermal control 
  • Higher cost due to engineered fluids and infrastructure 
  • Serviceability differs significantly from traditional data centre operations 
  • Fluid management and environmental considerations must be addressed 

Best For 

  • Ultra-high-density AI and HPC deployments 
  • Purpose-built or greenfield environments 
  • Applications requiring maximum thermal performance 
  • Facilities designed specifically for immersion cooling 

Key Takeaway  

Two-phase immersion cooling in a tank-based system offers the highest level of thermal performance – but comes with increased complexity, cost, and operational considerations, making it best suited for specialized, high-density deployments.