What Is a CDU (Coolant Distribution Unit)?

A coolant distribution unit (CDU) is the heart of every liquid-cooled data centre. It pumps, filters and controls the liquid that carries heat away from servers. Without it, direct-to-chip cooling simply does not work.

That matters more every year. AI racks now draw far more power than air can cool. As a result, operators are moving to liquid cooling, and the CDU sits at the centre of that shift. In this guide, we explain what a CDU is, how it works, the main types and how to choose the right one.

What is a coolant distribution unit?

A coolant distribution unit is a piece of equipment that links two separate liquid loops. On one side is the facility water system, which carries heat out of the building. On the other side is the technology cooling system, which runs through the servers.

The CDU keeps these two loops apart. In other words, building water never touches the IT hardware. Instead, heat passes from one loop to the other through a heat exchanger inside the CDU. The Open Compute Project describes this isolation as the CDU’s core job.

Think of it like the cooling system in a car. The engine (your GPU) makes heat, the coolant carries it away, and the radiator (the CDU heat exchanger) passes it on. Meanwhile, pumps and controls keep the flow steady.

CDUs also go by other names. You may see “cooling distribution unit” or simply “liquid cooling CDU”. They all describe the same device.

How does a CDU work?

A CDU works by moving heat from a closed server loop into the facility loop. The process runs in a continuous cycle:

  1. Coolant enters the cold plates. Treated coolant flows from the CDU through manifolds to cold plates mounted on GPUs and CPUs.
  2. The coolant absorbs heat. As it passes over the chips, the coolant warms up.
  3. Warm coolant returns to the CDU. The secondary piping carries it back to the unit.
  4. The heat exchanger transfers the heat. Inside the CDU, the warm coolant passes its heat to the cooler facility water.
  5. Facility water carries heat away. Next, dry coolers, cooling towers or chillers reject that heat outdoors.
  6. Cooled coolant goes back out. Finally, the CDU pumps the coolant back to the racks at a set temperature.

Throughout this cycle, sensors track temperature, pressure and flow. The controller then adjusts pump speed and valve positions in real time. As a result, the chips get a steady supply of coolant at the right temperature.

The piping between the CDU and the racks is often called the secondary fluid network (SFN). The piping between the facility plant and the CDU is the primary fluid network (PFN).

Key components of a coolant distribution unit

Every coolant distribution unit is built from the same core parts. However, the quality of each part decides how reliable the whole system is.

  • Heat exchanger: Usually a brazed or gasketed plate heat exchanger. It moves heat between the two loops without mixing them.
  • Pumps: Variable-speed pumps keep coolant moving. Most designs use redundant pumps, so one can fail without stopping the flow.
  • Filtration: Fine filters catch particles before they reach the narrow channels in cold plates.
  • Expansion tank: It absorbs changes in coolant volume as temperatures rise and fall.
  • Control valves: These valves regulate facility water flow to hold the supply temperature steady.
  • Sensors: Temperature, pressure, flow and leak sensors feed live data to the controller.
  • Controller and monitoring: A PLC-based controller runs the unit. It also connects to the building management system (BMS) or DCIM software.
  • Reservoir and fill system: This lets technicians top up and treat the coolant safely.

Types of CDU

CDUs differ in two ways: where they reject heat, and where they sit.

Liquid-to-liquid vs liquid-to-air

A liquid-to-liquid (L2L) CDU passes heat into facility water. It is the standard choice for new AI halls because it handles very high loads efficiently.

A liquid-to-air (L2A) CDU dumps heat into the room air through a radiator and fans. It suits older sites that have no facility water at the rack. However, its capacity is much lower, so it works best for small or pilot deployments.

In-rack, in-row and facility-scale

CDU typeWhere it sitsTypical useExample capacity
In-rackInside the server rack, usually at the bottomOne high-density rack or a small podUp to about 105 kW (Motivair)
In-rowFloor-standing, in the row beside the racksSeveral racks or a full rowUp to 2.5 MW (Cooler Master, OCP 2026)
Facility-scaleIn a plant room or utility corridorWhole halls or AI clusters8 MW-class and above (DCX)

In practice, large AI campuses now favour in-row and facility-scale units. They mean fewer machines to maintain per megawatt. Smaller sites, on the other hand, often start with in-rack units.

Why AI data centres need a coolant distribution unit

AI servers have changed the cooling equation. Modern GPU racks pack so much power into one cabinet that fans and cold aisles can no longer keep up. Liquid carries far more heat than air, so direct-to-chip cooling has become the default for AI clusters.

A coolant distribution unit makes that possible in four ways:

  • It protects the hardware. Cold plates have tiny channels. The CDU keeps the server loop clean, filtered and chemically treated, while building water stays outside it.
  • It holds a precise temperature. GPUs need coolant within a narrow band. The CDU controls supply temperature and flow, even as loads swing.
  • It cuts energy use. Liquid cooling allows warmer water. For example, ASHRAE’s facility water classes run from W17 up to W45 and beyond. Warmer water means fewer chiller hours and a lower PUE.
  • It scales with density. One CDU can serve many racks. Therefore, operators can add capacity row by row as GPU demand grows.

Note that ASHRAE W classes describe the facility water, not the coolant at the chip. The CDU adds an approach temperature between the two, according to GPU Smith’s guide to W classes. So, a good CDU with a small approach lets you run warmer facility water.

How to choose the right CDU

Choosing a CDU is not just about kilowatts. Before you buy, check these seven points:

  1. Cooling capacity. Match the rated load to your racks today, then leave headroom for the next GPU generation.
  2. Flow rate and pressure. Confirm the unit can deliver the flow your cold plates need at the required pressure drop.
  3. Approach temperature. A smaller approach lets you use warmer facility water. In turn, that saves chiller energy.
  4. Redundancy. Look for N+1 pumps, dual power feeds and hot-swappable parts. AI clusters cannot afford downtime.
  5. Filtration and fluid quality. Ask about filter rating, wetted materials and coolant chemistry. Poor fluid quality can foul cold plates over time.
  6. Controls and integration. The CDU should report to your BMS or DCIM over standard protocols such as Modbus, BACnet or SNMP.
  7. Serviceability and support. Check access for maintenance, spare-part lead times and local service teams.

Also, think about the full loop, not just the box. The CDU, the SFN piping and the manifolds must be designed together. Otherwise, flow imbalances and pressure losses can undo a good CDU.

Frequently asked questions about CDUs

What does CDU stand for in a data centre?

CDU stands for coolant distribution unit. It is the device that pumps, filters and controls the liquid used to cool servers.

What does CDU stand for in a data centre?

CDU stands for coolant distribution unit. It is the device that pumps, filters and controls the liquid used to cool servers.

What is the difference between a CDU and a CRAH?

A CRAH (computer room air handler) cools the air in a room. A CDU, by contrast, cools servers directly with liquid. Many AI halls use both: a CDU for the GPUs and air handling for the remaining heat.

Do I need a CDU for direct-to-chip cooling?

Yes, in almost every case. Direct-to-chip cooling needs a clean, controlled loop. The CDU creates that loop and keeps it separate from facility water.

Can a CDU work in an existing air-cooled data center?

Yes. A liquid-to-air CDU needs no facility water, so it can add liquid cooling to an existing hall. For larger loads, however, a liquid-to-liquid CDU is usually the better long-term option.

How long does a coolant distribution unit last?

A well-maintained CDU can run for many years. Regular filter changes, coolant testing and pump servicing keep it reliable.

Conclusion: the CDU is the core of liquid cooling.

In short, a coolant distribution unit is what makes liquid cooling safe, efficient, and scalable. It isolates the server loop, controls temperature and flow, and moves heat out of the building. As AI racks grow denser, the CDU has become as critical as the power train feeding them.

At RexEdge, we design and build liquid cooling infrastructure for high-density data centres. Our work covers CDUs and the secondary and primary fluid networks (SFN and PFN) that connect them to racks and plant. Because we engineer the whole loop, every part works together from day one.

Planning a liquid-cooled AI deployment? Talk to the RexEdge team to size the right CDU and piping for your site.

Home » Blogs & Articles » What Is a CDU (Coolant Distribution Unit)?

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top