Coolant Distribution Units in Liquid-Cooled Data Centers: A Technical Deep Dive

Table of Contents
1. Role and Operational Mechanism of the CDU
2. System Architecture and Key Thermal Parameters
3. Internal Structure and Core Components
4. Automatic Control System and Sensor Configuration
5. Structural Design and Patented Technologies
6. Design Standards and Regulatory Framework
7. Piping Systems and Quick-Connect Fitting Technology
8. Key Manufacturers and Competitive Landscape
9. Product Pain Points and Technical Bottlenecks
10. Engineering and Installation Guidelines

 

 

1.What a CDU Does and How It Works

The Role of the CDU

A Coolant Distribution Unit (CDU) is the thermal management hub of any liquid-cooled data center. Think of it as the system that sits between your building's chilled-water plant and the cold plates bolted to your servers — quietly doing the hard work of moving heat from silicon to sky.

CDUs typically account for 30–40% of the total cost of a liquid cooling system, which makes them the single biggest lever in optimizing lifecycle TCO. That dollar weight reflects both hardware complexity and the fact that a CDU failure is never a minor inconvenience: it puts racks offline.

Functionally, every CDU delivers four things: heat exchange between the facility side and the IT side, hydraulic isolation between those two loops, precise supply-temperature control, and leak detection with automatic shutdown. The industry shorthand — "the heart and brain of a liquid cooling system" — isn't wrong.

 

How the Heat Actually Moves

The core physics is straightforward: a brazed-plate heat exchanger couples two independent fluid loops without letting them mix.

The primary loop connects to the building's cooling plant — a cooling tower, dry cooler, or chiller. Chilled water or glycol solution enters the "hot side" of the heat exchanger, picks up heat from the IT loop, and returns to the plant to reject that heat. Typical parameters: supply 7–12 °C / return 12–18 °C for chilled-water plants; 30–40 °C / 35–45 °C for dry-cooler setups.

The secondary loop circulates deionized water or a dielectric fluid through server cold plates, absorbing heat from CPUs and GPUs. The warmed fluid returns to the "cold side" of the heat exchanger, gives up that heat to the primary loop, and heads back to the racks. Typical parameters: supply 18–25 °C / return 30–45 °C.

Why physical separation matters: Keeping the two loops hydraulically isolated is non-negotiable. The ultra-pure DI water in the secondary loop must never contact the facility's industrial water — contamination would degrade electrical resistivity and risk equipment damage. Isolation also lets each loop operate at its own pressure and temperature without affecting the other.

 

The Full Thermal Cycle, Step by Step

A variable-frequency pump drives secondary coolant (roughly 22 °C) through rack cold plates, where it absorbs chip heat and exits at around 40 °C.

That warm return flow passes through the secondary side of the plate heat exchanger, transferring heat to the primary-loop water.

Primary-loop water — arriving from the chiller or dry cooler — absorbs that heat and returns to the outdoor plant for rejection.

The cooled secondary fluid loops back to the racks, completing one closed thermal cycle.

Throughout this process, the PLC controller samples temperature and flow data continuously, running PID algorithms to trim pump speed and modulating valve position to hold the system at its target operating point.

 

 

2.System Architecture and Thermal Parameters

A Three-Tier Architecture

Liquid-cooled CDU systems organize naturally into three tiers:

Cooling source tier — outdoor cooling towers, dry coolers, or chillers that supply the primary-loop cooling capacity.

CDU distribution tier — the plate heat exchanger, variable-speed pumps, valve manifolds, sensors, and PLC that form the CDU itself.

IT load tier — server cold plates and the piping connecting them inside each rack.

Flow paths are straightforward: the primary side forms a closed loop between the outdoor plant and the CDU heat exchanger; the secondary side forms a closed loop from the CDU through distribution manifolds, through each rack's cold plates, and back.

 

Thermal Operating Parameters

 

Key Performance Metrics

 

 

3.Inside a CDU: Components and ConnectionCore Components and Selection Criteria

Every CDU is built from a predictable set of components. Getting the selection right for each one is where engineering decisions have the longest-lasting consequences.

 

 

Piping Connection Methods

How components are joined inside a CDU directly determines long-term reliability and how easy the system is to maintain.

 

Electrical Labeling Standards

Control panel components follow IEC labeling conventions throughout: -QF (circuit breaker), -KA (intermediate relay), -KM (contactor), -HL (indicator light), -S (pushbutton or selector switch), -V (variable frequency drive), -FU (fuse), -KT (timing relay).

 

 

4.Automation, Controls, and Instrumentation

The Four-Layer Control Architecture

CDU control systems follow a well-established four-layer hierarchy:

Communications: Between layers 1 and 2, Modbus TCP or BACnet/IP over Ethernet is standard. Between layers 2 and 3/4, the mix is typically 4–20 mA analog, RS-485 Modbus RTU, or IO-Link digital.

 

Instrumentation Layout

 

Control Strategies

Temperature PID control targets secondary supply temperature at 22 °C ± 0.5 °C. The PLC computes the error between the PT100 reading and setpoint, then outputs a 4–20 mA signal to modulate the primary-side control valve or primary pump speed. Proportional action dominates when error is large; integral action eliminates steady-state offset as the system converges.

Flow PI control maintains secondary-loop flow at setpoint through pump speed adjustment. This is typically cascaded with temperature control — the temperature PID output serves as the flow PI setpoint, and the flow PI output drives the VFD directly.

Differential pressure control in multi-rack parallel configurations uses pressure-independent control valves (PICVs) on each branch, or a CDU bypass valve, to hold branch differential pressure steady at 0.1–0.3 MPa regardless of load variation.

Safety interlocks are non-negotiable. Every CDU should implement these at minimum:

 

 

5.Form Factors and Recent Patent Activity

Three Common Form Factors

In-Rack CDU

Dimensions: 2U–4U height, standard 19-inch rack width

Capacity: 20–80 kW

Best for: small data centers, edge sites, lab deployments

Representative products: Vertiv CoolChip series, Envicool XD series,Lianli Liquid Cooling's Air-Liquid CDU and Liquid-Liquid CDU series.

Trade-off: highly space-efficient and fast to deploy, but limited single-unit capacity and lower redundancy

 

In-Row CDU (Floor-standing)

Dimensions: 600–1,200 mm wide, 1,800–2,200 mm tall standalone enclosure

Capacity: 100–500 kW

Best for: mid-scale data centers, racks running 30–80 kW

Representative products: Envicool Coolinside series, ZTE modular CDU,Lianli Liquid Cooling 500–2000 kW In-Row CDU

Trade-off: supports dual-pump, dual-HEX redundancy and easier servicing, but occupies significant floor space and can weigh 300–600 kg fully charged

 

Modular / Stackable CDU

Dimensions: standardized modules, 2–5 units stackable

Capacity: 200–400 kW per module, 2 MW+ combined

Best for: hyperscale and AI cluster data centers

Representative products: ZTE 400 kW–2 MW modular CDU, Vertiv large-capacity modules

Trade-off: elastic capacity expansion, N+1 redundancy, fault isolation — at higher upfront cost and dedicated power feed requirements

 

Notable Patents (2025–2026)

 

 

6.Standards and Regulatory Landscape

Applicable Standards

 

The Standardization Push

Interoperability is the industry's most urgent unsolved problem. Today, a CDU from vendor A often cannot connect natively to racks or couplings from vendor B — and that lock-in is expensive for operators.

The bottom line: standards are the mechanism that converts liquid cooling from a proprietary, high-TCO specialty to a commodity infrastructure layer. Projects specifying OCP-conformant interfaces and UQD-compliant couplings today are buying themselves optionality tomorrow.

 

 

7.Piping Systems and Quick-Disconnect Technology

Quick-Disconnect Coupling Types

Universal Quick Disconnect (UQD) Intel-led, OCP-standardized male/female coupling pair with integrated dual self-sealing valves. When disconnected, both ends seal automatically — maximum leakage under 0.05 mL. Bore sizes 5–12 mm; rated for 0–1.0 MPa and -10 °C to +70 °C service.

Blind Mate Quick Coupling (BMQC) Designed specifically for high-density rack environments where technicians cannot visually align connections. Tolerates ±3 mm positional misalignment and ±2° angular offset — critical for GPU server clusters where rack density leaves little room for precise hand-connection.

Dry Break Coupling Maximum leakage under 0.1 mL when disconnected. Used at critical connection points where even momentary exposure to coolant is unacceptable.

 

Material Selection Guide

 

 Coupling Supplier Landscape

Practical guidance: For HPC, supercomputing, and NVIDIA GPU clusters, Stäubli and CEJN remain the default choice where reliability matters most. Large-scale internet data centers deploying at cost can look to JONHON's UQD lineup. BEISHIDE and Dingtong offer strong value propositions for new-build projects evaluating domestic alternatives.

 

Piping Installation Best Practices

Pipe material: 316L stainless steel throughout, both primary and secondary loops. Copper is off the table — copper ions contaminate DI water and raise conductivity.

Welding: TIG process (ER316L filler). Follow with acid pickling and passivation. Orbital welding where access allows.

Pressure testing: Hydrostatic at 1.5× working pressure for 30 minutes, then pneumatic leak test at 1.1× working pressure.

Pre-commissioning flush: Circulate at > 2 m/s until strainer screens come up clean.

Insulation: Primary chilled-water lines need closed-cell foam insulation, minimum 25 mm. Secondary lines (18–45 °C) generally don't require insulation.

 

 

8.Market Landscape and Competitive Dynamics

 

CDUs Manufacturer Overview

 

Competitive Tiers

Tier 1 — Full-stack leaders: Envicool in China and Vertiv globally. Envicool holds 42–50% of the domestic cold-plate market and is the only Chinese liquid cooling vendor formally embedded in NVIDIA's MGX partner ecosystem.

Tier 2 — ICT platform vendors: Huawei and ZTE leverage captive AI compute ecosystems to push CDU adoption. ZTE's MWC 2026 modular CDU — supporting 400 kW to 2 MW through stackable modules — is a meaningfully differentiated play.

Tier 3 — Specialized players: Gaolan (extending from power electronics cooling), Shenling (Huawei ecosystem), and Yinlun (heat exchanger background) each bring focused technical depth and established customer bases in their respective niches.

Tier 4 — Innovators: Liquid Cooling Era, Chunjun New Materials, and peers are pursuing structural innovation through patent portfolios, concentrating on modular and stackable architectures.

 

Pump Supply Chain

 

 

9.Pain Points and the Technology Roadmap

 

Five Problems That Keep Operators Up at Night

Problem 1: Leak risk

A coolant leak into a server chassis is a bad day. It means short circuits, potential total-loss hardware damage, and cleanup costs that can easily run into the millions. Leaks most often originate at quick-disconnect seals as they age or at flanged joint faces.

Mitigations: (1) Negative-pressure liquid cooling — liquid ring vacuum CDUs maintain secondary-loop pressure below atmospheric, so a pipe breach draws air in rather than pushing liquid out (commercial systems are now available). (2) Three-stage response — leak detection cable triggers automatic pump shutdown and alarm before coolant spreads. (3) Dual self-sealing couplings — UQD-style connections seal both sides on disconnection, limiting any release to under 0.05 mL.

Problem 2: Fragmented standards and vendor lock-in

Incompatible coupling sizes, non-standard manifold interfaces, and proprietary control protocols regularly force data center operators into single-vendor dependency. The OCP Foundation, UQD Alliance, and China's TC28 are all working to fix this, but the honest assessment as of mid-2026 is that the specifications exist but field adoption lags. Specifying OCP/UQD-conformant equipment today is the most practical hedge.

Problem 3: Pump reliability and efficiency

Pumps are the only rotating component in a CDU, and they carry the system's reliability on their shoulders. Traditional mechanical seals can leak over time, canned-motor pumps run less efficiently, and magnetic-levitation pumps cost significantly more. Oversized pump selections — common when engineers apply conservative safety factors — waste energy continuously.

Mitigation: IE4/IE5 efficiency-class variable-speed pumps sized precisely to the actual system curve, with active/standby redundancy.

Problem 4: Water quality management

DI water isn't a set-and-forget medium. Secondary-loop conductivity climbing above 1 µS/cm raises electrical fault risk. Biological growth in stagnant sections causes fouling that degrades both heat transfer and flow. Both problems are insidious — they develop gradually and often go undetected until something more serious happens.

Mitigation: Always-on conductivity monitoring feeding an automatic bypass ion exchange system, with periodic bacteriological testing and a documented water quality program.

Problem 5: Weight and access constraints

A fully charged cabinet CDU can exceed 500 kg — enough to be a structural concern on older raised-floor data centers. Rack-mounted CDUs avoid the weight problem but often sacrifice maintainability, with complex piping competing for technician access in tight aisles.

Mitigation: Front-of-unit service clearance of at least 800 mm is mandatory. Design modular sub-assemblies to slide out for replacement without requiring the full CDU to be drained and moved.

 

 

Where the Technology Is Heading

 

10.Installation, Commissioning, and Operations

Site Prerequisites

Before a CDU arrives on site, the following needs to be confirmed:

 

Piping Construction Standards

Material: 316L stainless pipe throughout, wall thickness per pressure class (Sch 10S or Sch 40S). No copper.

Welding: TIG process, ER316L filler wire. 100% visual inspection of welds plus 10% RT sampling.

Passivation: Acid pickling and passivation of all welds before system fill.

Pressure testing: Hydrostatic at 1.5× design pressure (30-minute hold) followed by pneumatic at 1.1× design pressure (15-minute hold, leakage < 0.5%).

System flush: Pre-commissioning flush at > 2 m/s until strainers are clean.

Insulation: Closed-cell foam on primary chilled-water lines, minimum 25 mm, flame class B1.

Pipe supports: Stainless hangers at DN50: 2.5 m, DN80: 3.0 m, DN100: 3.5 m intervals.

Quick-connect installation: Mark male/female pairs to prevent mismatch; PTFE tape on all threaded connections.

 

Commissioning Sequence

Positioning: Level installation with service clearances of 800 mm front, 600 mm sides, 500 mm rear. Drip pan or floor drain directly below the unit.

Fill and deaeration: Fill secondary loop with DI water, open automatic air vents at CDU high points and system high points. Let system stand 24 hours, then verify conductivity < 1 µS/cm.

PID tuning: Step-response test to establish initial PID parameters (Ziegler-Nichols or auto-tune). Target: secondary supply temperature within ± 0.5 °C of setpoint under steady-state load.

Interlock verification: Test every safety interlock individually — leak detection to pump trip, over-temperature to maximum cooling response, over-pressure to bypass opening.

Redundancy test: Force primary pump failure, verify standby switchover completes in < 5 seconds with < 10% flow deviation.

72-hour burn-in: Full-load continuous operation for 72 hours before formal acceptance.

 

Acceptance Criteria

 

Preventive Maintenance Schedule

 

 Troubleshooting Common Faults

 

Quick Tip
Lianli Liquid Cooling offers customizable CDU solutions for data centers, including rack-mount and cabinet-style units available in both air-to-liquid and liquid-to-liquid configurations. We provide comprehensive, one-stop liquid cooling solutions. Operating our own R&D and manufacturing facility in Longgang, Shenzhen, we produce CDUs compatible with NVIDIA, Intel, and AMD servers. Our products hold authoritative international certifications such as RoHS, CE, and UL; they achieve a PUE of less than 1.15, reduce cooling energy consumption by over 40%, and significantly lower long-term operating costs.

 

Conclusion

Coolant distribution units sit at the intersection of two converging forces: the relentless climb in rack power density driven by AI accelerators, and the data center industry's hard push to squeeze more out of every watt it consumes. CDUs are no longer optional infrastructure for niche HPC deployments — they're becoming a baseline requirement for any facility running dense GPU workloads.

 

The near-term technology trajectory is clear: negative-pressure CDUs to solve the leak problem at its source, AI-augmented control systems to handle the volatile demand profiles of modern AI training and inference, and modular designs capable of scaling past 2 MW per system as GB200- and B300-class deployments become the norm. Waste heat recovery at 45–60 °C return temperatures is moving from experimental to economically interesting.

For the engineering and operations teams responsible for these systems, four areas deserve sustained attention above all others: piping integrity and seal quality, coupling selection and standardization, PID control commissioning, and ongoing water quality management. Getting these right at installation protects the investment over a system lifetime measured in decades.

 

Standardization remains the industry's most important unfinished business. As OCP V3 and UQD specifications gain adoption and national standards solidify around 2026–2027, the interoperability gaps that currently force vendor lock-in will narrow. When that happens, liquid cooling's cost and complexity barriers will drop — and the technology's shift from specialist preference to infrastructure default will accelerate sharply.

 

FAQ:

1.How significant are the actual energy savings of a CDU compared to traditional air-cooling systems?
The energy-saving advantages of CDU-based liquid cooling systems are primarily reflected in the PUE (Power Usage Effectiveness) metric. While traditional air-cooled data centers typically have a PUE between 1.4 and 1.6, adopting a CDU liquid cooling solution can lower the system PUE to below 1.10—with some highly optimized projects reaching as low as 1.05. Taking a data center with a 1 MW IT load as an example: reducing the PUE from 1.5 to 1.1 saves approximately 3.5 million kWh of electricity annually. Based on an industrial electricity rate of 0.7 RMB/kWh, this results in annual electricity cost savings exceeding 2.4 million RMB. The more significant the energy savings, the shorter the CDU's return on investment (ROI) period; the incremental costs can typically be recouped within 3 to 5 years.

 

2.How do I determine if the coolant on the CDU's secondary side needs replacement or treatment?
Water quality management is crucial for the long-term, reliable operation of a CDU. Key indicators to monitor include:
① Conductivity: The normal value should be below 1 μS/cm. If it exceeds 2 μS/cm, a coolant replenishment or replacement process must be initiated immediately to avoid increasing the risk of electrical leakage in servers;
② pH Value: The normal range is 6.5–8.0; values ​​that are too acidic or too alkaline accelerate corrosion in piping and heat exchangers;
③ Particulates and Microorganisms: Regular visual inspections of samples are required; sterilization and flushing are necessary if turbidity, discoloration, or unusual odors are detected. It is recommended to conduct comprehensive water quality testing quarterly and replace the coolant annually, while also ensuring that online conductivity sensors are properly calibrated to enable real-time alerts.

 

3.Can CDUs and quick-connect couplings from different manufacturers be used together? What is the current state of compatibility?
This is one of the most common concerns regarding equipment selection in the industry today. Compatibility currently falls into two categories: standardized interfaces allow for mixed use—meaning that quick-disconnects (UQDs) compliant with OCP UQD specifications (such as those from brands like Stäubli, CEJN, and AVIC-JONHON) are theoretically interchangeable between male and female connectors. Conversely, non-standard interfaces do not allow for mixed use; some manufacturers still employ proprietary connector specifications and tubing sizes. If open standards are not specified during procurement, there is a risk of vendor lock-in when expanding capacity or replacing equipment later. To mitigate future compatibility issues, it is recommended to explicitly require compliance with OCP Open Rack V3 and UQD connector standards during the project tendering phase and to stipulate interface interoperability verification tests within the contract.