Coolant Distribution Units (CDU) Market (2026-2036)

This report addresses coolant distribution units across all end-use industries, including data centres and artificial intelligence infrastructure, semiconductor manufacturing, medical imaging equipment, industrial lasers and machine tools, automotive and battery test, power electronics, scientific instruments and professional audiovisual applications. Data centre applications constitute the largest end-use segment and are covered in greater depth in dedicated coverage of that segment, so the two should not be read as additive.

The growth of this market is driven by the transition of computing from air to liquid cooling, by rising thermal density in power electronics and test equipment, and by the extension of precision temperature control into processes that previously tolerated wider variation. The more consequential development is that the product is being pulled apart by two opposite forces. In data centre applications the buyer has published both the design and the method of scoring it, which turns the unit into a build-to-print commodity. In process and precision applications the unit is qualified against a specific item of capital equipment and is effectively unswitchable once designed in. These are two businesses with opposite competitive dynamics operating under one product name. A supplier strong in one is not necessarily credible in the other, and the strategic question is which set of conditions a firm is equipped to survive rather than which segment grows faster.

A coolant distribution unit sits between two fluid circuits and manages the transfer of heat from one to the other while isolating them hydraulically. In its data centre form it separates the facility water system from the technology cooling system, so that treated coolant of known chemistry and pressure can be delivered to cold plates without exposing them to the facility loop. In its process form it holds a tool, a laser, a magnet or a test article at a specified temperature within a tolerance the process requires. The underlying components are similar in both cases, consisting of pumps, heat exchangers, filtration, instrumentation and controls, which is why one product name covers both. The commercial realities could hardly be more different.

The data centre half of this market has been deliberately commoditised by its largest buyers, and the evidence is now published rather than inferred. OCP Project Deschutes is a data centre facilities specification for a 2 MW next-generation coolant distribution unit for high-power IT pods including artificial intelligence and machine learning ASICs. Google is the sole listed contributor, NVIDIA is acknowledged for feedback, and the document describes itself as the first CDU specification submitted to the Open Compute Project. Version 1.0.0 took effect on 26 February 2026, having evolved from an initial 0.75 release in July 2025. The specification sets a thermal load of 2,000 kW at a 3 degree Celsius approach temperature, an IT flow of 500 GPM with 80 to 90 psi available IT pressure, a facility flow of 500 GPM at 0.25 GPM per kW with a 15 to 30 psi facility pressure drop, an operating pressure range of 0 to 130 psig, power consumption of 74 kW and pump voltage of 380 to 416 Vac. It specifies pumps rated at 500 GPM at 112 psi, three dual-pass heat exchangers in parallel, dimensions of 65.25 by 91.44 by 42.29 inches, a wet weight of 6,900 lbs and a dry weight of approximately 5,300 lbs, and deionised water and PG25 as coolants with approved chemistries stated.

A specification of that resolution is not a requirement document, it is a manufacturing drawing. When the buyer states the pump duty, the heat exchanger architecture, the external dimensions to a quarter of an inch, the wet and dry weights, the power draw and the approved coolant chemistries, the supplier is building to print. On interoperability the specification requires controls to support both IPv4 and IPv6, with the programmable logic controller reading and writing Modbus over TCP across either, and requires units to operate in parallel on a common header under individual or group control. Pumps and accessories follow N plus one operation with each pump having its own variable frequency drive, input power requires at least 1N feed redundancy with 2N desired, and redundant components must be replaceable during continuous operation with safety interlocks. Pumps are serviceable from front and rear with front preferred, filters and the expansion tank are front-serviceable, and vendors must supply maintenance procedures for field-replaceable units. Those provisions complete the picture, because parallel operation on a common header under group control, with front-serviceable components and vendor-supplied maintenance procedures, describes a unit designed to be interchangeable within an installation and therefore between suppliers.

The second instrument removes the remaining room for differentiation by defining how units are to be measured. The Open Compute Project has separately published Liquid to Liquid CDU Test Methodology and Performance Rating at Revision 1.0, dated August 2024 with an initial release recorded as 1 November 2024, authored by contributors from Intel, CoolIT Systems and Meta Green Cooling with input from further liquid cooling vendors and integrators. The document sets design guidelines, qualification tests and a performance rating format for liquid-to-liquid units separating the facility water system from the technology cooling system. It standardises cooling capacity in kW, facility-side approach temperature, flow rate and pressure drop, and technology-side pressure head and flow rate, with water on the facility side and propylene glycol on the technology side. Test conditions specify an operating range of 17 to 65 degrees Celsius, a non-operating range of minus 40 to 85 degrees Celsius, relative humidity of 10 to 90 percent, a service life of 10 years for pump-based units, and filtration at 200 microns or finer for the primary filter and 50 microns or finer for the secondary, with reliability testing covering hydrostatic pressure, chemical compatibility, shock and vibration and thermal cycling.

The rating method is the decisive part. For rating, the capacity of an in-rack unit is the highest value in kW at which it meets a technology-side pressure head of at least 25 psi, a facility-side pressure drop of no more than 75 psi and an approach temperature difference of 5 degrees Celsius at a flow of 1.5 litres per minute per kW, with in-row units held to a technology-side pressure head of at least 40 psi on the same basis. Units are compared through a capacity matrix of approach temperature difference against flow rate per kW, in which lower approach temperature difference and lower flow indicate better performance, improvable through cold plate optimisation and through unit or system design. A supplier can therefore no longer claim superior performance on its own terms, because capacity is defined as the point at which stated pressure and approach temperature conditions are met at a stated flow per kW, and comparison is made on a published matrix. What survives as differentiation is thermal efficiency per unit of flow, since the document is explicit that lower approach temperature difference and lower flow indicate better performance. That is a narrow and demanding basis on which to compete, and it is the basis the customer has chosen.

The consequence for the data centre segment is a shift of value away from product design and towards manufacturing cost, delivery capacity and field service. A build-to-print unit of 6,900 lbs wet weight containing three heat exchangers and pumps rated at 500 GPM is an exercise in supply chain and assembly, and the firms that win it are those able to produce at volume, deliver against an energisation date and support a fleet. This is why the segment is attracting entrants from power electronics, thermal module and contract manufacturing backgrounds rather than only from precision cooling, and why incumbents whose advantage rested on thermal engineering find that advantage neutralised.

The process and precision half of the market operates on inverted principles. A unit cooling a lithography or etch tool, a medical imaging magnet, an industrial laser or a battery test cell is specified on temperature stability rather than on capacity, is qualified against a particular equipment platform, and is frequently named in the equipment documentation. Nobody publishes those designs, because the equipment maker regards the thermal behaviour of its tool as proprietary and the end user regards requalification as a cost to be avoided. Volumes are low, unit prices are high, product lives are long and switching is rare. The supplier holds the position in this half, which is the opposite of the data centre situation.

Form factor and capacity trends follow from the data centre specification directly. A 2 MW unit of the dimensions published is facility-scale plant rather than rack or row equipment, which is why the floor-standing category and the above 1 MW capacity band grow fastest even though in-row units and the 100 to 500 kW band remain the larger shares. The published requirement for parallel operation on a common header reinforces this, since it describes an installation built from several large units operating as a group rather than many small ones distributed through the hall.

A caution belongs in any assessment of this market. The commoditisation of the data centre segment is a benefit to buyers and a hazard to suppliers, and it is proceeding faster than the segment is growing. A vendor whose plan assumes that liquid cooling demand will support historical margins is likely to find volume rising while price per kW falls, and should model the two separately. The process and precision segment offers the inverse profile, with defensible positions and modest growth, so the strategic question for most suppliers is not which segment grows faster but which set of competitive conditions they are equipped to survive.

Qualification restrains the process and precision half of this market because a unit is accepted against a specific equipment platform rather than against a general standard. An equipment maker will not revalidate thermal behaviour to admit an alternative supplier without commercial reason, and an end user treats requalification as a cost and a risk to a production process, so designed-in positions persist for the life of the platform. The restraint protects incumbents and simultaneously caps the rate at which the segment can grow or reallocate, which means suppliers cannot convert data centre volume capability into process market share. The barrier is commercial and procedural rather than technical, so it cannot be overcome by building a better unit.

Open specification restrains supplier economics even as it expands demand, because a unit built to a published design with a published rating method competes on cost rather than on capability.

A specification that states pump duty, heat exchanger architecture, external dimensions, weights, power draw and approved coolant chemistries leaves the supplier assembling to a drawing, and a rating method that defines capacity as the point at which stated conditions are met removes the ability to claim superior performance on alternative terms.

The restraint falls on margin rather than on volume, and it is proceeding faster than the segment is growing, so a supplier modelling liquid cooling demand at historical margins will overstate its own prospects.

The effect is visible first in tender pricing and only later in reported margin, which delays recognition inside supplier organisations.