Air Cooling vs. Liquid Cooling: The Future of Data Center Thermal Management

Why Are Data Centers Turning to Liquid Cooling?

As AI and high-performance computing (HPC) continue to expand, data center infrastructure must keep pace with increasing power demands. Today’s data centers are scaling up rapidly, adopting higher processing speeds and more component miniaturization. Rack power densities that once peaked at 20kW can now exceed 100kW in the largest hyperscale applications. These densities will only continue to grow rapidly; in fact, experts predict 1 megawatt (MW) racks within the next few years.

In the past, standard air-based cooling systems have been the traditional method for keeping Information and Communication Technology (ICT) equipment cool. Air-cooled systems – often configured with hot/cold aisles for extra efficiency – remain a straightforward, cost-effective choice for data centers that use conventional air cooling. But this air-cooling approach is not an optimal choice for many newer applications in the hyperscale and data center facilities of the future. Generative AI and GPU-accelerated servers in HPC applications generate far more heat than traditional computational loads, and air cooling cannot keep up.

As a result, more data centers are turning to liquid-based cooling methods such as direct-to-chip liquid cooling and immersion cooling. These advanced techniques offer significant energy savings, greater processing capacity, and enhanced sustainability.

Regardless of the thermal approach, data center engineers need reliable power delivery. Air-cooled systems demand power supply unit (PSU) and power distribution unit (PDU) connectors that can withstand elevated ambient temperatures in hot-aisle environments, while immersion cooling requires components that can operate safely when submerged in dielectric fluids.

Air Cooling: The Familiar Choice for Lower-Density Data Center Racks

Air cooling remains a common thermal management method for small to midsize data centers with rack densities under around 25 kW. This method uses conventional air conditioning technology and airflow management strategies to regulate ICT equipment temperatures and prevent overheating. A method for improving the energy efficiency of air-cooled data centers involves implementing hot aisle /cold aisle configurations with aisle containment systems.

Introduced by IBM in 1992, the hot/cold aisle configuration is one of the oldest ways to conserve energy within a data center. Server racks in this setup are arranged in alternating rows. Cold aisles are the server rows facing the rack fronts, while hot aisles align with the back of the servers. Cold air from the computer room air conditioner (CRAC) system is delivered via raised floors into the cold aisles, where it is drawn through the front of the servers to cool them. The hot exhaust is then expelled into the hot aisles before returning to the CRAC unit along the ceiling vents to complete the cycle.

To improve the efficiency of this setup, many facilities implement the best practices outlined in the ANSI/TIA-942 standard. The recommendations include:

• Raising the floor by 46 cm (1.5 feet) to improve airflow
• Creating an aisle containment system with permanent, automatic doors to stop cold air from mixing with warm air and improve the efficiency of the system
• Maintaining cold aisle widths of 1.2 meters (4 feet)

Other methods to improve efficiency involve raising air inlet temperatures toward the mid-range of ASHRAE’s 18° to 27°C (64 to 81°F) guidelines and implementing air-side economizers that leverage outside air for cooling.

What Are the Advantages and Downsides of Air-Cooled Data Center Technology?

Air-cooled systems with hot/cold aisles remain a common solution for many smaller data centers, offering a familiar and relatively low-cost option for thermal management. The advantages include:

• Improved Energy Efficiency (Compared to Traditional Layouts): Air-cooled systems with hot/cold aisle layouts can improve cooling efficiency by 10-35% and reduce cooling-related energy costs by up to 30%. In traditional data centers, each subsequent aisle gets progressively warmer as air passes through the racks, meaning the cooling system must work harder. Hot/cold aisle configurations are more efficient because they prevent hot exhaust air and cold air from mixing. When paired with variable-speed fans, U.S. Department of Energy estimates that containment strategies can reduce fan energy use by 20% to 25%.
• Simple and Familiar Maintenance: Air conditioning technology is mature, widely understood, and well-supported. This means data center operators have more options for servicing, quicker turnaround times for repairs, and generally lower maintenance costs compared to newer liquid immersion cooling systems with specialized components.

However, there are significant downsides to the air-cooling approach, including:

• Space Constraints: Air-cooled data centers require large mechanical infrastructure such as CRAC units, raised floors, and ducting. These components occupy valuable floor space in an application where space-savings translates to monetary savings; it costs roughly $600 – $1000 per square foot to build a data center.
• Energy Efficiency (Compared to Liquid Cooling): Despite the efficiency gains generated from implementing hot/cold aisles, air-cooling still requires a significant amount of energy. According to research from 3M, approximately 38% of a traditional air-cooled data center’s total energy consumption is dedicated solely to cooling electronic components. Liquid cooling technologies, in contrast, are much more efficient.
• High Noise Levels: Air cooling equipment generates a substantial amount of noise. Sound levels may exceed 80 decibels in some facilities, requiring hearing protection for personnel exposed over long periods. On the other hand, immersion cooling systems are virtually silent.
• Retrofit Complexity: Designing a new data center facility with hot/cold aisles is relatively straightforward. However, retrofitting a legacy data center with hot/cold aisles can be labor-intensive and may not be worth the cost. Design engineers must consider the effort associated with moving racks, reconfiguring airflow paths, and overhauling heating/AC systems.

For data centers with lower rack densities, air cooling systems remain a viable and cost-effective solution for thermal management. However, when rack densities rise, the limitations of air cooling become harder to justify unless coupled with liquid cooling. The added thermal load requires significant CRAC capacity expansion, driving up utility costs.  

Optimal PDU/PSU Connectors for Air-Cooled Data Centers

Power distribution unit (PDU) and power supply unit (PSU) connectors are important components within ICT design. Design engineers working within air-cooled hot/cold aisle data centers should weigh unique considerations when selecting their PDU and PSU connectors. These connectors and power cords are critical for connecting the PDU to the switch, the PDU to the server, the PDU to storage, and more.

What Is Liquid/Immersion Cooling in Data Centers?

As data center workloads intensify, more hyperscale and data center facilities are turning toward immersion cooling as a high-efficiency alternative or addition to traditional air cooling. Reflecting this trend, the immersion cooling market is forecast to grow from $1B in 2025 to $3.18B in 2030 at an impressive compound annual growth rate of 26.15%.

While immersion cooling might be a novel method in the data center space, the underlying technology is far from new. Various forms of liquid cooling have existed since the late 1800s, when water was used to cool high-voltage transformers. A form of liquid cooling exists in many high-performance desktop PCs as a standard way to prevent overheating. Today, the data centers industry is undergoing a similar transformation.

At its core, liquid cooling involves submerging ICT equipment into a specialized fluid that absorbs and removes the heat via convection. The heat is transferred from the equipment to the fluid, which is then moved away from the equipment and extracted through various methods like heat exchangers or direct liquid-to-liquid cooling. Then, the cooled fluid returns to the immersion enclosure, completing the cycle.

Although deionized water is a potential coolant, it poses leakage risks and often requires a Leak Prevention System (LPS). Instead, modern data centers most often use chemically engineered dielectric fluids. These fluids are inert, non-conductive liquids with excellent thermal conductivity and stability, allowing them to absorb heat more efficiently than air and operate safely around electric equipment. They must be low in viscosity, non-toxic, non-corrosive, and have a high flash point.

When selecting a dielectric fluid, engineers should consider factors such as the fluid’s heat transfer performance, ease of ICT equipment maintenance, material compatibility, environmental impact, and total cost over time. The two main types of dielectric fluids are:

• Fluorochemicals: Fluorochemical fluids are gaining traction due to their exceptional chemical and thermal stability. They have a lower boiling point and can be used for both single-phase and two-phase immersion cooling applications. However, they are relatively expensive and produce greenhouse gas effects.
• Hydrocarbons: Hydrocarbons offer good thermal performance and include options like mineral oils, synthetic oils, and natural oils. They are typically only used in single-phase systems due to their combustibility and flammability, but are more economical than fluorochemicals.

What Are the Types of Liquid Cooling?

Another liquid cooling method is direct-to-chip cooling, where coolant is delivered directly to hotter components, such as GPUs, via cold plates. Rather than immersing the entire server in the dielectric fluid, this method places the cold plate directly on high-heat areas for efficient thermal management. Direct-to-chip systems may be designed as either single phase or two phase systems.

What Are the Advantages and Disadvantages of Liquid Cooling?

Immersion cooling has typically been used in specialized environments, such as HPC and generative AI data centers. However, it is gaining traction in mainstream data centers as operators look for ways to boost sustainability, reduce operational costs, and meet rising performance demands. The benefits of immersion cooling include:

• Increased Processing Power: Liquid cooling enables data centers to support much higher processing power without risk of overheating, including next generation rack densities of 100kW and above. Immersion-cooled data centers can provide up to 120% more processing capacity and up to 40% more energy efficiency compared to traditional air-cooled facilities.
• Reduced Energy Consumption (Cost Savings and Sustainability): By reducing the need for energy-intensive CRAC units, immersion cooling reduces data center energy consumption by over 60% (up to 95% within certain systems). For larger data centers, this amounts to millions of dollars in saved utility costs and operating expenses. In addition, immersion cooling uses less water than air-cooled systems and helps reduce the data center’s overall carbon footprint. As more data center sustainability initiatives like the European Commission’s Energy Efficiency Directive are adopted, these sustainability gains will become more important.
• Enhanced Equipment Lifespan: Immersion cooling minimizes wear and tear on ICT equipment, resulting in extended lifespans and reduced maintenance needs.
• Space Efficiency: Immersion cooling systems do not require extensive airflow infrastructure like ducting or large AC units, meaning more computational power can be packed in smaller physical footprints. This space-saving power advantage is particularly important for certain data center applications in dense urban environments or within developed markets where new data center construction cannot keep up with demand. For example, Northern Virginia’s data center vacancy rate dropped from 1.8% in 2023 to 0.9% in 2024 – despite an 18% increase in inventory over the same period. 
• Scalability: Immersion cooling systems are easier to scale. As data centers are projected to consume an estimated 8% of total U.S. electricity by 2030 (up from 3% in 2022), efficient, scalable infrastructure will be critical to controlling cost and environmental impact.

One drawback of immersion cooling systems involves their higher upfront costs. Compared to traditional air-cooled systems, immersion setups require expensive specialized equipment like tanks and dielectric fluids. However, immersion cooling’s higher energy efficiency translates to lower cooling costs, higher processing power, and reduced energy consumption over time – offsetting the initial investment in the long run.

What PDU/PSU Connectors Are Optimal for Liquid-Cooled Data Centers?

The evolution of immersion and liquid cooling is introducing new demands for connector materials and durability. In addition to withstanding electrical and thermal stress, connectors in immersion applications must also be chemically compatible with the dielectric fluids used. Saf-D-Grid from Anderson Power is built with tough materials that can withstand exposure to certain dielectric fluids. The connector housing resists degradation with many dielectric fluids, ensuring long-term integrity and reliability even when fully or partially immersed. As liquid cooling technologies continue to advance and potential liquid cooling standards emerge from Open Compute Project industry efforts, connector designs like Saf-D-Grid will continue to evolve to support the data centers of the future.

What Are the Types of Saf-D-Grid and Their Power Capacity?

Beyond its durability with immersion cooling applications, Saf-D-Grid delivers additional power density gains to provide a compact, high-performance solution to meet today’s data center challenges. For engineers aiming to optimize power density and save valuable rack space, Saf-D-Grid provides a small but impactful way to meet next-generation power demands. The connector is available in three main configurations, providing engineers with the flexibility to select the sizing, amperage capacity, and voltage support that best align with their specific application:

Additional customizable Saf-D-Grid options are available for data centers with unique form factors or performance parameters, such as low-clearance configurations and an extra PDU key. Design engineers can contact Anderson Power to learn more about Saf-D-Grid solutions.

Air and Liquid Cooling and the Future of Data Centers