The number of liquid-cooled data centers is expected to grow from less than 1% of the market to around 30% of all installations in the next two years.

The reason behind this staggering growth is obvious - processor power is increasing rapidly every 6 months, and so is the power per rack. As a result, data center temperatures are rising, and air cooling has reached its limit. Hyperscale data center operators are no longer wondering if they should use liquid cooling, but rather which technology they should use and how quickly they can be up and running. This article will take a deep dive into the various options available today so that you can make the best decision for your specific application.

As shown in the figure below, liquid cooling is mainly divided into two categories: immersion and direct-to-chip, and each cooling method has both single-phase and two-phase options. People often call direct-to-chip cooling "cold plate cooling" because it uses a cold plate placed directly on top of the CPU or GPU. Immersion cooling, on the other hand, uses large, heavy water tanks instead of standard racks, and the tanks are filled with liquid. Servers and equipment are directly immersed in the liquid.

Immersion Cooling Overview

Liquid-Filled Container

There are two types of immersion liquid cooling. Single-phase immersion cooling uses an oily liquid in a tank to absorb heat. The heated liquid rises to the top of the tank and is pumped to a heat exchange unit that cools the liquid and returns it to the tank that holds the hardware.

Here are the pros and cons of single-phase immersion:

Pros: 

Able to remove 100% of the heat from the server. 

Cons: 

Because the oil rises slowly to the top of the tank and is pumped to cool it, it can only cool processors and chips with a maximum power consumption of 500 watts. Large, bulky tanks require a large data center infrastructure investment, are expensive to maintain, and often require downtime. Oil can also be flammable at high temperatures, requiring extensive maintenance due to the large volume of the tank. Since all equipment is directly immersed in the liquid, there is also a risk of equipment damage.

Two-phase immersion cooling uses a low-boiling-point dielectric fluid instead of oil. The heat boils the fluid, creating steam, which rises to the top of the tank, where a network of pipes provides flowing cooling water. The steam condenses and drips back into the tank.

Example of a bidirectional immersion tank filled with dielectric fluid

The pros and cons of dual-phase immersion are as follows:

Pros: 

This method effectively removes 100% of the heat, and the use of a dielectric fluid ensures the safety of IT equipment. 

Cons: 

Large, bulky tanks require a significant data center infrastructure investment, are expensive to maintain, and often require downtime. Not only do the components inside the tanks need to be compatible with the dielectric fluid to prevent damage, but the boiling process inside the tanks often boils off material from motherboards, solder joints, and other equipment. This process can harm the life of the equipment, and as material is shed, it needs to be constantly filtered, which requires large filters. The tanks require regular maintenance, and when the tanks are opened, vapors from the dielectric fluid are released into the atmosphere.

Direct Chip Liquid Cooling Overview

Cold Plate on Top of GPU/CPU

Direct chip cooling consists of a cold plate that sits on top of the CPU or GPU. Unlike immersion cooling, where the device is submerged in liquid in a large tank, the liquid used in direct chip cooling is contained within a compact cold plate.

Direct chip liquid cooling includes single-phase or two-phase processes. Single-phase direct chip cooling uses water or a water-glycol mixture as the coolant in the cold plate. The water remains in liquid form, and the heat dissipation capacity of this method depends on the water flow rate.

Here are the pros and cons of single-phase direct chip cooling:

Pros: 

Cold plates eliminate the need for large, expensive water tanks required for immersion cooling. 

Cons: 

Using water in cold plates poses a risk of corrosion if the water is not continually filtered and maintained, and a single leak could be catastrophic to a server.Since cooling relies on a constant flow of water, the higher the chip temperature, the greater the water flow required. This design requires high pressure and flow, larger pipes, tubes and connectors, and power-hungry pumps to continuously deliver water.

Unlike single-phase direct chip cooling, dual-phase direct chip cooling does not use water in the cold plate, but uses thermal transfer fluid, which is 100% safe for IT equipment. The heat generated by the GPU and CPU causes the thermal transfer fluid to boil at a low temperature, absorbing the heat and keeping the chip at a constant temperature. This process is similar to boiling water to keep the bottom of a pot at 100⁰C, but at a lower temperature. When the liquid inside the cold plate boils, the liquid in the cold plate will never exceed the boiling point. This effect makes the technology highly scalable and can be used to cool future chips with higher temperatures.

Here are the pros and cons of two-phase direct chip cooling:

Pros: This approach requires almost no changes to data center infrastructure, and the thermal fluid does not need to be filtered, balanced, or replaced. The liquid is in a closed system, which means it is not released into the atmosphere. No water is used in the cold plate and server, eliminating the risk of leaks, and because the temperature of the liquid never exceeds the boiling point, it can operate seamlessly even if the chip temperature increases in the future. 

Cons: Liquid cooling is only used for CPU/GPU cooling. Other components (such as memory and I/O) still need air cooling.

Liquid cooling market is getting as hot as chips

According to a recent research report, the liquid cooling market is expected to grow from $5.65 billion in 2024 to $48.42 billion in 2034. The reason for this huge growth is obvious: as heat inside AI factories and data centers rises to unprecedented levels, there is simply no other way to support AI buildouts.

Fortunately, as this article highlights, there are many different types of liquid cooling technologies that can be used to dissipate heat, each with its own pros and cons. As chip power continues to increase, it is critical to ensure that the selected cooling solution can adapt to the next generation of chips. Data centers, AI factories, and hyperscale computing centers need to decide which solution is better suited to their current needs and meets their near-term requirements. Ultimately, this choice will be based on performance, cost, power consumption, ease of use, scalability, and sustainability.

Source: This article is compiled from hpcwired

Reference link: https://www.hpcwire.com/2025/05/06/which-liquid-cooling-is-right-for-you-immersion-and-direct-to-chip-explained/

View more at EASELINK