The rise of Generative AI, from tools like ChatGPT to a new wave of AI tools, is reshaping technology demands, driving a surge in the need for advanced CPUs and GPUs. These powerful processors, with ever-increasing thermal design power (TDP)—like Nvidia’s GB200 Grace Blackwell superchip consuming 1,200W per GPU with racks powering 120kW—are pushing power and cooling requirements to unprecedented levels, making traditional air cooling solutions insufficient.
Enter liquid cooling. With adoption projected to grow at a 45% compound annual growth rate (CAGR) from 2023 to 2028, direct-to-chip liquid cooling solutions, powered by coolant distribution units (CDUs), are transforming how data centers handle the heat generated by high-performance AI workloads. Cooling efficiency isn’t just a consideration—it’s a necessity as AI processors continue to push power and density boundaries.
This article examines how CDUs are supporting thermal heat management, enabling operators to meet the relentless cooling demands of AI and high-performance computing while paving the way for the next generation of data center efficiency.
CEO and Founder, JetCool Technologies.
What is a Coolant Distribution Unit (CDU)?
A CDU is the heartbeat of a liquid cooling system, carefully regulating coolant temperature and flow rates to maintain optimal cooling efficiency. By managing the coolant flow to IT equipment and returning the IT heat to the facility’s water for re-cooling, CDUs help stabilize temperatures and minimize the risk of overheating. In direct liquid cooling, the CDU plays a vital role by allowing for temperature conditioning to prevent condensation and by isolating the IT equipment from harsher facility water, which may contain mineral deposits, particulates, and other impurities that could damage cooling systems or reduce efficiency.
CDUs are also pivotal for increasing system longevity. According to a recent Uptime Institute study, over 70% of unplanned downtime in data centers is linked to power and cooling system failures. With advanced CDUs, operators can reduce such risks, ensuring reliable operations even under heavy workloads.
Different Types of CDUs: Liquid-to-Liquid vs Liquid-to-Air
Liquid-to-Liquid and Liquid-to-Air CDUs both serve the essential function of cooling IT equipment but are suited for different environments based on their cooling mechanisms and efficiency levels. Liquid-to-Liquid CDUs are ideal for facilities with access to facility water, offering high cooling capacity and efficiency due to the superior thermal conductivity of water.
These systems transfer heat from the IT equipment’s coolant loop to the facility’s water loop using a heat exchanger, which is well-suited for high-density environments. In contrast, Liquid-to-Air CDUs use air-cooled radiators and fan systems to dissipate heat into the surrounding air, making them a better choice for locations without access to facility water, though they typically offer lower overall cooling capacity and efficiency.
While both systems include similar components needed to run (such as pumps, temperature control systems, and filtration), the key difference lies in how they transfer heat. Liquid-to-liquid CDUs rely on primary and secondary pumps to circulate coolant and water, while Liquid-to-Air CDUs depend on fans to move air over radiators. Maintenance for both systems is essential, though Liquid-to-Air CDUs may require more frequent attention to air filters and fan components.
In contrast, liquid-to-liquid systems require monitoring the water loop and its cleanliness. Ultimately, the choice between a Liquid-to-Liquid or Liquid-to-Air CDU depends on the available infrastructure and specific cooling and efficiency goals, with Liquid-to-Liquid systems excelling in high-density environments and Liquid-to-Air systems offering flexibility where water access is limited.
Industry Trends and Future Directions
The adoption of liquid cooling is on a sharp upward trajectory. In 2023, only 10% of data centers used liquid cooling, but this figure is expected to reach 50% by 2030. Driving this shift are the increasing thermal demands of AI and HPC and environmental considerations. Traditional air cooling can consume up to 40% of a data center’s energy, with goals for advanced cooling to reduce energy expenditures to as little as 5% of total IT load. This is a significant factor in reducing overall Power Usage Effectiveness (PUE).
CDUs are also addressing water conservation challenges. Records obtained by the Financial Times reveal that data center water consumption has spiked by nearly two-thirds since 2019, with over 1.85 billion gallons consumed in 2023 compared to 1.13 billion gallons in 2019. Technologies that eliminate reliance on evaporative cooling are becoming indispensable. Eco-friendly coolants and closed-loop systems are gaining traction, helping operators reduce their environmental footprint.
Conclusion
As AI and HPC workloads intensify, data centers must adopt cutting-edge solutions like CDUs to stay ahead. By enabling efficient, scalable cooling, CDUs not only support operational demands but also align with sustainability goals. With liquid cooling expected to dominate the market by 2030 and the data center cooling market set to exceed $20 billion by 2028, the time to embrace these technologies is now. CDUs are not just a tool for cooling—they are the foundation for a sustainable, high-performance future.
We list the best web hosting services.
This article was produced as part of TechRadarPro’s Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro
Services Marketplace – Listings, Bookings & Reviews