How CDUs are Powering the Much-Needed Transition to Liquid Cooling
Data centers are facing an unprecedented transformation. The rise of AI and high-performance computing (HPC) has pushed traditional air cooling to its breaking point. According to Uptime Institute, air cooling becomes inefficient and costly beyond about 20 kW per rack, yet AI servers today often exceed 100 kW, and emerging GPU systems are expected to top 600 kW per rack by 2027.
This performance explosion has made liquid cooling not just a high-tech alternative, but a necessity. Compared to air cooling, liquid systems can transfer up to 3,000 times more heat, cut total energy use by about 30%, and reduce greenhouse gas emissions by up to 21%. They
At the core of every liquid cooling system is the coolant distribution unit (CDU). CDUs are what manages the flow of coolant between the cooling source (like a chiller) and the equipment being cooled. It regulates temperature, pressure, and flow rate to ensure efficient heat removal and system stability, often including pumps, valves, sensors, and heat exchangers for precise thermal management.
As a central component of any liquid cooling system, it is mission critical that data center operators select the right CDU for their deployment. For a comprehensive, step-by-step guide on how to select a CDU for data center liquid cooling deployment, check out the complete whitepaper here.
Or, if you’re just looking to get your toes wet, here’s a quick, abbreviated guide on how to select the right CDU for your liquid cooling deployment.
First Things First: Evaluating Environment & Workloads
Before looking at vendors or comparing features, start with fundamentals. Every successful CDU deployment begins with a clear understanding of the environment it will be deployed in and the types of workloads it will be handling. What that ultimately entails is understanding four core components: total design power / thermal capacity (in kW), flow requirements, head pressure, and facility integration matters.
- Thermal Capacity: This is the cornerstone of CDU selection. Your CDU must be capable of handling the total design power of the system, or the amount of heat your servers and GPUs generate. As a rule of thumb, plan for the heat load of your most power-dense racks and the total load per compute pod.
- Flow Rate: Flow rate is directly tied to cooling performance. Modern AI and HPC clusters demand higher flow rates than ever—often 1.5 to 2 liters per minute per kilowatt of heat.
- Head Pressure: The CDU’s pumps must generate enough pressure to push coolant through every loop, bend, and cold plate in your system. Calculate distances to and from the CDU as they can influence ultimate head pressure
- Facility Integration: Your CDU doesn’t operate in a vacuum—it has to fit into your data center’s chilled water system, power infrastructure, and floor layout. Before selecting a CDU, ask yourself the following: Is your chilled water constant-flow or variable-flow? Do you have sufficient floor space and power feeds? Will the CDU be in-row, end-of-aisle, or in the gray space?
Sometimes, More is Better: Build in Redundancy and Reliability
In the age of AI, where uptime equals revenue, outages and downtime are no longer acceptable. A single cooling failure can jeopardize millions of dollars of hardware and undermine countless hours of training or other computational workloads. That’s why redundancy isn’t a luxury, it’s common sense.
When considering your own CDU selection, look for certain safeguards, such as: :
- N+1 pump redundancy (an extra pump that kicks in automatically)
- Dual power feeds for resilience
- Bypass filtration for maintenance without downtime
- Even complete CDU redundancy for critical or high-value deployments
Balancing cost and reliability is key. A more expensive, redundant system will pay for itself many times over the first time it successfully avoids devastating outage.
Cost Considerations: Think Beyond the Initial Pricetag
While CAPEX (upfront cost) matters, it’s only part of the equation. OPEX (i.e. operating expenses like energy, maintenance, and potential downtime) can far outweigh the initial purchase price of a system. When cost is considered in relation to return on investment (ROI), CDUs often demonstrate rapid payback periods.
By enabling higher-density deployments, reducing the need for overprovisioned cooling infrastructure, and improving system efficiency, many CDUs pay for themselves in under a year. Still, operators should build financial models that account for local utility rates, redundancy strategies, and anticipated expansion plans.
A low-cost CDU that lacks essential features may ultimately cost far more in downtime, service interruptions, and retrofit expenses. While overprovisioning is a waste of capital, underprovisioning is a recipe for disaster.
Beyond the Basics: A Brief Look at Advanced Features and Scalability
Once the essentials are covered, shift your focus to features that future-proof your deployment. Although they may not be absolutely necessary today, with the rate of advancements we’re currently seeing in hardware, what is “good enough” today might fall short tomorrow..
- Monitoring & Control: Modern CDUs come with advanced telemetry and group control capabilities. These allow multiple CDUs to coordinate efficiently, balancing flow and preventing unnecessary energy consumption. Real-time visibility into flow rate, pressure, and temperature enables proactive maintenance and performance optimization.
- Scalability & Modularity: Your data center will grow, and your CDUs should too. Modular designs let you expand capacity incrementally as workloads increase. With rack densities rising from 10 kW to 80 kW (and heading toward 600 kW), scalability isn’t optional.
These are just a couple of the types of advanced capabilities worth considering when selecting a CDU. While they may seem like extras, they can prove essential for long-term efficiency and resilience.
Avoiding Pitfalls and Common Mistakes
The CDU market is crowded and complex. Beyond that, not all vendor claims hold up under close scrutiny. By staying vigilant and avoiding some of these common pitfalls, you can ensure your deployment won’t fall flat.
- Overfiltration: Filters finer than 25 microns may sound better but can actually hurt efficiency by stripping coolant additives and increasing pressure drop.
- Overprovisioning: Bigger isn’t always better. A three-pump system isn’t necessarily more effective than two if designed well. Oversizing increases both energy use and costs.
- Misleading Specs: Vendor claims often reflect ideal lab conditions, not real-world performance. Always request independently verified test data under realistic temperatures and redundancy configurations.
- Ignoring Coolant Chemistry: Different coolants—like PG-25 or pure water—affect heat rejection and component compatibility. Make sure your CDU is designed for your fluid choice.
A Look Behind the CDU: Evaluate Vendors Like Partners, Not Suppliers
The best CDU is only as good as the company behind it. Vendor experience, manufacturing capability, and post-sale support play a major role in long-term success. Look for partners that offer proven, validated performance data and certifications (UL, CE, etc.), strong manufacturing capacity and supply chain resilience, and responsive service and field support.
Remember, CDUs are complex machines with lots of moving parts. Purchasing one is also investing in the supplier itself. Make trust and reliability top priorities when determining who to buy your CDU from. Additionally, take seriously things like regional production, sustainable sourcing practices, and regulatory compliance matters specific to your specific geographic location. It is imperative that your CDU partner is capable of keeping up with global supply matters and is cognizant and in compliance with your region’s local regulations.
The Bottom Line
As the computing world races toward ever-denser, AI-driven architectures, data centers must evolve to keep up. Air cooling simply can’t sustain the power and thermal demands of modern workloads. Liquid cooling is the path forward, and the CDU is its driving force..
Selecting the right CDU means balancing performance, reliability, cost, and scalability. By understanding your requirements, embracing redundancy, avoiding marketing hype, and partnering with an experienced vendor, you can build a cooling infrastructure ready for the future of high-density computing.
Want to learn more about the CDU selection process? Read the full whitepaper
Ready to take the next step? LiquidStack’s experts can help you align CDU selection with your facility’s unique demands.
Contact us today to start your liquid cooling journey.
