Data Center Cooling: Air vs Liquid vs Immersion

Data center cooling systems fall into three main categories: air cooling, liquid cooling (direct-to-chip or rear-door heat exchangers), and immersion cooling (full or two-phase). Air cooling is the traditional baseline, liquid cooling is the current upgrade path for high-density racks, and immersion cooling offers the highest heat removal per volume but with significant infrastructure changes.

Formula

This calculator works out two key metrics for comparing cooling systems: Power Usage Effectiveness (PUE) and cooling efficiency. PUE is the total facility power (IT load plus cooling and overhead) divided by the IT load alone. Cooling efficiency is simply the IT heat load divided by the cooling system’s power draw.

PUE = (IT load + Cooling power) / IT load

Cooling efficiency = IT load / Cooling power

Where IT load is the heat generated by servers (kW), and Cooling power is the electricity consumed by the cooling system (kW). A lower PUE means more of the facility’s power goes to computing, not cooling. Higher cooling efficiency means each kW of cooling power removes more kW of heat.

Worked Example

Using the default values: IT load = 10 kW, cooling power = 3 kW.

PUE = (10 + 3) / 10 = 1.30

Cooling efficiency = 10 / 3 = 3.33 kW heat per kW cooling

This means for every 1 kW of cooling electricity, the system removes 3.33 kW of heat, and the facility uses 1.30 kW total for every 1 kW of IT load.

Background: Why Cooling Matters

Data centers generate enormous amounts of heat. A single rack of modern servers can dissipate 10–30 kW, and high-density AI clusters can exceed 100 kW per rack. Without cooling, chips quickly exceed their safe operating temperature (usually 80–90°C junction temperature) and either throttle performance or fail. Cooling accounts for roughly 20–40% of a data center’s electricity bill, so choosing the right system is both a technical and financial decision.

Air cooling has been the standard for decades, using computer room air conditioners (CRACs) or air handlers (CRAHs) to push chilled air through raised floors or overhead ducts. Liquid cooling emerged in the 2010s for high-performance computing, and immersion cooling has gained traction recently for extreme densities and efficiency gains.

How Air Cooling Works

Air cooling relies on moving cool air across server components and exhausting warm air. Typical configurations include:

• Room-level cooling: CRAC/CRAH units blow cold air under a raised floor; it enters through perforated tiles, passes through racks, and returns warm to the unit.
• Row-level cooling: In-row or row-based units sit between racks, shortening the air path and allowing higher densities.
• Rack-level cooling: Rear-door heat exchangers or top-mounted units capture heat at the rack exhaust.

Air cooling is simple, cheap to install, and easy to maintain. However, it becomes inefficient at high densities because air has low heat capacity and requires large volumes of airflow. Typical air-cooled facilities achieve PUE around 1.3–1.5, with cooling efficiency around 1–2 kW heat per kW cooling.

How Liquid Cooling Works

Liquid cooling uses water or a coolant to remove heat directly from components. The two main approaches:

• Direct-to-chip (cold plate): A metal plate with internal channels sits on top of CPUs/GPUs. Coolant flows through, absorbing heat, then goes to a heat exchanger or cooling tower.
• Rear-door heat exchangers: A coil mounted on the back of the rack; warm air from servers passes through the coil, transferring heat to liquid.

Liquid cooling can handle 30–100 kW per rack with ease, and PUE can drop to 1.1–1.2. Cooling efficiency is typically 3–5 kW heat per kW cooling. The main challenges are leak risk, higher upfront cost, and the need for a liquid loop infrastructure (pumps, pipes, heat exchangers).

How Immersion Cooling Works

Immersion cooling submerges servers entirely in a dielectric (non-conductive) liquid. There are two types:

• Single-phase: Servers sit in a tank filled with a dielectric fluid (e.g., mineral oil or synthetic ester). The fluid absorbs heat and is pumped to a heat exchanger.
• Two-phase: The fluid boils at a low temperature, and the vapor condenses on a coil above the tank, releasing heat. This allows extremely high heat transfer but requires a sealed system and specialized fluids.

Immersion cooling can remove 100+ kW per tank, achieve PUE as low as 1.02–1.05, and cooling efficiency of 5–10 kW heat per kW cooling. It also eliminates fans and air filters, reducing maintenance. Drawbacks include high capital cost, the need to redesign server hardware for submersion, and the difficulty of servicing components (you must pull the whole server out and drain it).

Choosing the Right System

Selecting a cooling system depends on several factors:

• Rack density: Air works up to ~20 kW/rack; liquid for 20–100 kW; immersion for >100 kW.
• Existing infrastructure: Retrofitting liquid or immersion into an air-cooled facility is costly and may require downtime.
• Climate: Cooler climates allow free-air cooling (using outside air) with less mechanical cooling, improving PUE for air systems.
• Water availability: Liquid cooling often requires a water loop and cooling tower, which may not be feasible in water-scarce regions.
• Maintenance and serviceability: Air is easiest to service; immersion is the hardest.
• Total cost of ownership: Consider not just installation but energy savings, maintenance, and downtime costs over 5–10 years.

For most enterprise data centers, air cooling remains sufficient. Liquid cooling is becoming the norm for AI and HPC workloads. Immersion is still niche, mainly used by cryptocurrency miners and a few hyperscale operators.

Common Problems and Misconceptions

One common misconception is that liquid cooling means water near electronics, which is dangerous. In practice, direct-to-chip systems use sealed, non-conductive coolants or carefully managed water loops with leak detection. Immersion fluids are specifically chosen for their dielectric properties.

Another issue is that PUE alone doesn’t tell the whole story. A facility with PUE 1.1 but high cooling power per kW of heat may still be less efficient than one with PUE 1.2 but lower total energy use. Always compare cooling efficiency and total energy consumption.

Air cooling also suffers from hot spots and uneven airflow. Liquid cooling eliminates most of these issues but introduces pump failures and potential leaks. Immersion cooling has the issue of fluid degradation over time and the need for periodic fluid replacement.

Costs and Safety Considerations

Installation costs vary widely by region and scale, but generally:

• Air cooling: Lowest upfront cost, highest operating cost at high densities.
• Liquid cooling: Moderate upfront cost (piping, pumps, heat exchangers), lower operating cost.
• Immersion cooling: Highest upfront cost (tanks, fluid, facility modifications), lowest operating cost.

Safety is a critical factor. Liquid cooling systems must prevent leaks near electrical components; use non-conductive coolants and robust containment. Immersion systems must handle fluid spills, vapor emissions (for two-phase), and ensure proper grounding. Air systems need adequate airflow and fire suppression, but have fewer fluid hazards.

Always consult with qualified engineers and follow local building and electrical codes. The calculator above provides only a general efficiency comparison, not a full cost analysis.

Frequently Asked Questions

Is liquid cooling safe for electronics?

Yes, when properly installed. Direct-to-chip systems use sealed loops with non-conductive coolants or dielectric fluids. The risk of short-circuit is minimal because the coolant doesn’t conduct electricity, and systems include leak detection and shutoff valves.

Can I retrofit my existing air-cooled data center to liquid cooling?

Yes, but it’s not trivial. You need to add piping, pumps, heat exchangers, and possibly modify racks. Retrofitting is often more expensive than building new, and requires careful planning to avoid downtime. Some facilities start with rear-door heat exchangers as a less invasive step.

What is the typical PUE for each cooling type?

Air-cooled facilities typically achieve PUE 1.3–1.5. Liquid-cooled facilities can reach 1.1–1.2. Immersion cooling can achieve 1.02–1.05. However, PUE varies with climate, IT load, and design efficiency.

Does immersion cooling require special servers?

Yes. Standard servers have fans, air ducts, and components that may not be rated for submersion. Immersion-ready servers remove fans, use sealed connectors, and have heat-dissipating designs. Many major server vendors now offer immersion-ready models.

Which cooling method is best for small data centers?

For small or low-density facilities, air cooling is usually the most cost-effective and easiest to maintain. Liquid or immersion only becomes worthwhile when rack densities exceed ~20 kW/rack or when energy costs are extremely high.

What is two-phase immersion cooling?

Two-phase immersion uses a fluid that boils at a low temperature (around 50–60°C). As components heat up, the fluid boils and rises as vapor, then condenses on a cooled coil, releasing heat. This phase change transfers heat extremely efficiently, allowing higher densities than single-phase.

Related

• Apple Music Autoplay: What Happens After Your Queue Ends?
• Primary vs Secondary Phone: Which Setup Fits You?
• HomePod: Who Should Buy Apple’s Siri Smart Speaker?

sel mode | Cooling mode | Air=1, Liquid=2, Immersion=3
in heat | Server heat load | kW | 10 | 0 | 1000
in watts | Cooling system power | kW | 3 | 0 | 1000
out pue | Power usage effectiveness (PUE) | - | (heat + watts) / heat | 2
out eff | Cooling efficiency | kW heat / kW cooling | heat / watts | 2