When a hyperscale data center operator in Northern Virginia commissioned its latest 50-megawatt facility, the thermal design team encountered an issue that air could not solve alone. The cutting-edge AI accelerators and modern high-speed CPUs were dissipating more than 1,000 watts per chip, resulting in a heat flow higher than a hot plate. Traditional air cooling using hot aisle containment and energy-efficient CRAC units was unable to cool down the chips fast enough. The answer was the direct liquid cooling (DLC) that incorporates cold plates, coolant distribution units (CDUs), precision-engineered connectors, and stainless-steel manifolds that circulate a dielectric or treated water coolants through cold plates of each chip to cool it down. The power usage effectiveness (PUE) for that data center declined from 1.4 to under 1.1, whereas the rack that used traditional air cooling could handle only half of the capacity, which is now twice as much.
Summary: The main elements of the DLC apparatus comprise cold plates, which are heat exchangers mounted on microprocessors, coolant distribution systems (CDUs), which include pumps, heat exchangers, and controls maintaining the temperature and flow of the coolant, piping and connectors that form the liquid route to the rack and each server, and manifolds and dripless disconnects or couplings that allow one server to be connected and cut off from the system without turning off the whole loop. Selecting, placing, and maintaining the components is a field of competence that is based on the practices of the sanitary and high-purity industries. Any leakage from the liquid cooling system would cost millions of dollars in IT equipment destruction; thus, the components responsible for this leakage must be manufactured with the same levels of material traceability, surface finish, and pressure integrity as industry sectors such as pharmaceutical and semiconductor industries.
The Importance of Liquid Cooling for Modern Data Centers
The switch from air cooling to liquid cooling in data centers is not a fad; rather, it is due to thermodynamic principles surrounding the thermal design power (TDP) of processors. A decade ago, the average server CPU generated 150 watts, while today’s AI processor can dissipate heat in the range of 700-1,200 watts, and fully-equipped GPU servers consume more than 10,000 watts of power for each rack unit. Air is not suitable for cooling the heating requirements since it has low thermal capacity and thermal conductivity. To remove such amount of heat from such tiny area, facilities have to use an unbelievable amount of air power, namely fan speed, airflow volumes, and temperature delta, which is not possible in traditional facilities. Liquid has four times better thermal capacity and over twenty times higher thermal conductivity. The path of liquid cooling allows to remove the heat from chips and deliver it to the place where it can be disposed of, such as a cooling tower, chiller, or heat exchanger with much less energy consumption than it takes to operate any air cooling system.
The second factor influencing the rise of liquid cooling in modern data centers is compute density and the cost of real estate. A liquid cooling system can provide 50kW per rack, while air cooling system provides only up to 10 kW per rack. Therefore, much more processing power is available for the same space when one is using liquid water for cooling. In regions where rents for data centers are high and energy prices are high, availability of such technology gives a tremendous competitive advantage to companies using it. The third factor is water consumption. One might think that liquid cooling means larger consumption of water, but in reality, it is often the opposite.
The Two Main Types of Liquid Cooling: Direct Liquid Cooling vs. Immersion Cooling
The table below summarises the two primary liquid cooling architectures and where each is best applied.
| Architecture | How It Works | Best For | Key Considerations |
|---|---|---|---|
| Direct Liquid Cooling (DLC) | A cold plate is mounted directly on the chip (CPU, GPU, or AI accelerator). Coolant — typically treated, deionised water or a water‑glycol mixture — is circulated through the cold plate via a network of tubing, quick‑disconnect couplings, and stainless steel manifolds. The coolant absorbs the chip's heat and carries it to a coolant distribution unit (CDU), where it is rejected to the facility's primary cooling loop. The rest of the server's components — memory, storage, power supplies — are cooled by air. | Retrofitting existing air‑cooled data centres with high‑density racks, new AI and HPC deployments, any scenario where only specific high‑power components require liquid cooling and the rest of the rack can remain air‑cooled. | DLC captures 60–85% of the server's total heat at the cold plate. The remaining heat must still be managed by the room's air‑conditioning system. The cold plates, the tubing, and the quick‑disconnect couplings must be 100% leak‑free, because a single drip can destroy a server. The coolant must be treated and filtered to prevent corrosion, scaling, and biological growth in the loop. |
| Immersion Cooling (Single‑Phase or Two‑Phase) | The entire server — motherboard, chips, power supplies, cabling — is submerged in a dielectric fluid that is electrically non‑conductive. In single‑phase immersion, the fluid is circulated through a heat exchanger. In two‑phase immersion, the fluid boils at a low temperature, and the vapour is condensed and returned to the tank. | Greenfield data centres designed from the ground up for immersion, extreme‑density deployments, edge computing sites, and applications where water near electronics is unacceptable and a dielectric fluid is required. | Immersion cooling captures nearly 100% of the server's heat. However, it requires specialised tanks, server chassis, and maintenance procedures. The dielectric fluid is expensive, and the server warranty may be affected. Immersion is less common than DLC in existing data centres because of the infrastructure changes required. |
The Critical Fluid Components of a Direct Liquid Cooling System
A DLC system is a liquid handling network, and its performance is also measured by the effectiveness of its parts in transmitting, regulating, and connecting the coolant.The key aspects of the fluid handling industry that make good use of the precision efficiency of the material and sophistication of production are the stainless steel manifolds (which are designed to allow the flow of coolant between the main feed lines of the system and the cooling stations, manufactured of 316L stainless steel with electropolished surfaces so as not to rust and release any foreign substances into the coolant — if it corrodes, the components become a source of contamination of the cold plates and the efficiency of the cooling will be lowered), the quenched couplings (the couplings need to allow a signle server to be connected or disconnected without introducing any air into the system and spilling the coolant on the machine); these are the most important mechanical part of the DLC system and must be well-protected.The commonly used types of hoses include EPDM, PTFE-coated, and corrugated stainless steel hoses. The quality standards governing these components require them to be produced in accordance with regular traceability of materials, surface performance specifications, and pressure integrity. The members of ASHRAE Technical Committee 9.9 provide guidelines for the requirements applicable to a liquid cooling system and the quality of coolant needed for each of the components involved.
Installation and Maintenance: The Techniques Used to Ensure a Leak-Free Operation
Setting up a liquid cooling system in a data center mainly refers to the piping and fluid-handling tasks. The requirements and methods applied to the pharmaceutical water delivery systems are valid for the light cooling project as well.The major installation practices consist of clean assembly (every tube, every connection, and every manifold has to be washed, sealed and remain free from debris at the moment of installation — a piece of dirt or a metal scrap getting into the coolant circuit will eventually block the cold plate or control valve), pressure testing (the whole installed circuit shall be pressure tested with nitrogen or with the system coolant at a pressure above the maximum operating one — usually 1.5 times higher than the design pressure — and this pressure should be upheld for a fixed period to prove that there are no leaks at any joint, connection, or valve), passivation and flushing (a new stainless steel system needs to be passivated — chemically treated in order to remove the free iron from the surface and to form the chromium oxide layer protecting the surface — and followed by careful flushing of the whole system with the coolant in order to wash out all chemical and solid residues before connecting the IT equipment), and commissioning and flow balancing (as soon as the system is filled with coolant and starts operating, the flow in each row and to each cold plate has to be monitored and balanced in order to make sure that every server receives the flow design rate).
The maintenance of a direct liquid cooling (DLC) system is represented by quality assurance of the coolant (the pH, conductivity and level of inhibitors should be controlled on a regular basis since an acidic coolant or a coolant devoid of its inhibitors will start corroding the stainless steel, cold plate interiors, and seals of the pump), leak detection (each rack manifold and cold plate should have built-in leak detection system — a simple conductivity sensor installed at a drip pan under the connections — which will inform the building management system about the leak before the damage is inflicted on the IT equipment), and scheduled replacement of operation-elapsed components (quick-connect couplings and flexible hoses in a pipe have a certain service life defined by the manufacturer of the parts which is why their replacement must be scheduled to avoid any functional failure).
How Eagle Fittings Supports Data Centre Liquid Cooling
Eagle Fittings is a leader in stainless steel manufacturing, as it builds on decades of experience gained from working in the sanitary and high-purity fluid control industries. The company's essential skills in the production of 316L stainless steel manfolds as well as electro polished tubing and several types of fittings including plastic tube fitting, quick connect fittings, ball valves and check valves can lay the groundwork for creating DLC systems utilized in cooling the most powerful AI and HPC.
Eagle Fittings uses the same manufacturing processes like material traceability, surface finish treatment and pressure testing, which are in demand in pharma industries (e.g. used for WFI distribution systems). Our article on what a sanitary fitting is explains the design and manufacturing principles that ensure a leak‑free, clean, and reliable fluid connection — principles that apply equally to a dairy processing line and to a direct‑to‑chip cooling loop. For data centre operators, colocation providers, and liquid cooling system integrators, Eagle Fittings provides the stainless steel fluid handling components — the manifolds, the connectors, the valves — that a reliable, maintainable, and leak‑free liquid cooling installation demands.
Frequently Asked Questions
What is direct liquid cooling (DLC) in a data centre?
Direct liquid cooling (DLC) refers to a process whereby a cold plate is installed directly onto the power chip to allow a liquid coolant to flow through it and thus extract heat from the chip. The liquid coolant is then sent to a heat exchanger, where its heat will be rejected to the main cooling system of the facility. This kind of cooling system is widely used today for converting air-cooled data centres into ones that can accommodate high-density racks.
Is liquid cooling better than air cooling for data centres?
In case of the high-density racks that have a power rating of more than 20 kW per rack, it is safe to say that liquid cooling is significantly better than air cooling in that it removes heat more effectively than the latter, uses less energy, allows for greater rack density and lowers PUE from about 1.4-1.6 to below 1.1. In case of low-density racks, however, it is still more reasonable to choose air cooling system than liquid ones, and most modern data centres use the mixed type of cooling: air for the less power demanding components of their data centre and liquid cooling for the chips with high performance.
What coolant is used in data centre liquid cooling?
The most widely used coolant for their cooling systems is deionized and treated water that contains an anti-corrosion agent and biocides, as well as water-propylene glycol mixture which is used for freeze prevention in colder climates. The coolant must be non-corrosive to the polymer materials, copper and steel that are parts of the cooling circuit. Moreover, it must have a low electrical conductivity level which is important in case of a leak.
How do you prevent leaks in a liquid‑cooled data centre?
The leak prevention in liquid-cooled data centres is achieved through use of precise manufactured components, in following establishment methods ensuring cleaner assembling of the components of the cold plates and in constant monitoring of the quality of the components used in the liquid cooling process.
References
- ASHRAE TC 9.9 — Liquid Cooling Guidelines for Data Centers. The industry standard for liquid cooling design, coolant quality, and operational guidance, published by the American Society of Heating, Refrigerating and Air‑Conditioning Engineers.
- Uptime Institute — Data Center Cooling and PUE Trends. Annual research on data centre cooling technologies, PUE benchmarks, and the adoption rates of liquid cooling in hyperscale and colocation facilities.
- McKinsey & Company — The Future of Data Center Infrastructure. Research on the impact of AI and high‑performance computing on data centre power and cooling requirements.
- Open Compute Project (OCP) — Advanced Cooling Solutions. The industry collaboration that publishes open‑source specifications for liquid cooling components, including cold plates, manifolds, and quick‑disconnect couplings, to promote interoperability and reduce cost.
Liquid cooling solutions for data centers are not a temporary response to a hot chip. These are the permanent thermal management systems of today’s AI and high-performance computing age. The age when the chips are too hot for any air-cooling systems, since the racks contain too highest number of heat-generating components that no elevated floor cooling techniques can help, and energy efficiency demands are too great to allow any kind of mechanical refrigeration systems in the past. Any of the devices functioning impeccably as part of liquid cooling system comprise stainless steel manifolds, non-leak couplings, and well-engineered valves that all require the same level of quality, accuracy, and maintenance as it was the case with pharmaceutical water systems.
