Liquid cooling, the hidden engine of artificial intelligence
Can water save data centers from massive overheating?
The explosion of artificial intelligence and high-density computing is pushing technological infrastructure to its limits.
By: Robson Pacheco, Data Center Sales Manager at Vertiv Brazil.
Thanks to advances in high-performance computing (HPC), streaming, machine learning, and AI, we are currently experiencing a moment of profound technological convergence.
A decade ago, IT racks consumed between 5 and 10 kW—some as little as 1 kW.
This range remained relatively stable for years. However, the development of high-performance chips and AI-specific architectures has led to far more powerful and compact systems.
Today, racks can easily exceed 100 kW, with power consumption levels ranging from 400 to 600 watts per processor. This leap forward brings a crucial challenge: dissipating the enormous heat these systems generate.
Modern chips, including AI accelerators such as GPUs and TPUs, cannot effectively transfer heat because they generate extremely high thermal densities and operate at temperatures that surpass the limits of conventional air-cooling methods.
Heat output can reach several hundred watts per square centimeter, making traditional heat sinks and fans insufficient to maintain safe operating temperatures.
As a result, new liquid-cooling technologies—systems that inherently dissipate heat more efficiently than air-based cooling—are increasingly being considered and adopted.
Liquid cooling reduces energy consumption and improves overall system availability by lowering cooling infrastructure power use by up to 30%. It also minimizes thermal throttling—a mechanism that slows processor performance to prevent overheating—while enabling higher-density deployments in data centers and maintaining optimal operating conditions for hardware components.
Why Use Liquid Cooling?
For a long time, air cooling was enough. However, in this new era, liquid cooling has emerged as a necessary solution—with water taking center stage in the next generation of data center cooling.
In the video podcast “What’s Next? with Vertiv“, which you can listen to [here], Diego Corceiro, Mechanical Engineering Manager at LZA, explains that water has a significantly higher thermal conductivity than air, making it a much more efficient medium for heat dissipation. As CPUs and GPUs become denser and pack more computing power per rack, the available space for air intake decreases, and air circulation alone becomes insufficient.
In this context, liquid cooling—particularly direct-to-chip systems, where coolant circulates through cold plates that distribute cooling directly across the chip—offers an effective solution for managing these high-thermal-load components.
Liquid cooling provides several benefits, including superior thermal efficiency due to water’s excellent heat-transfer properties, reduced energy consumption (fewer fans mean less power for air movement), high modular scalability (ideal for phased expansion projects), a smaller physical footprint, and better environmental control (quieter and more energy-efficient systems).
While liquid-cooling technology is advancing rapidly, a hybrid approach combining liquid and air cooling remains the most practical solution for data centers today. In most cases, cooling an entire facility exclusively with water is not yet feasible.
At Vertiv, we often see a mix: some rows equipped with liquid-cooled racks, and others relying on traditional air-based systems such as Computer Room Air Conditioners (CRACs) or Air Handling Units (AHUs)—specialized systems designed to manage temperature and humidity within data centers.
Key Considerations for Your Liquid-Cooling Strategy
If you are an investor planning to build a robust, sophisticated, and secure data center in the region, discover the essential steps and best practices to meet these new demands in our video podcast “What’s Next? with Vertiv”:
Study the market. Who are your clients? What workloads will they run? Without clear answers, your infrastructure could become obsolete within just two years.
Design intelligently. With liquid cooling, you can reduce the number of racks required and consolidate workloads into a smaller space. This gives you the flexibility to start small and scale over time. However, mechanical, electrical, and civil infrastructure must be considered from the outset.
Prepare for the future. If your goal is to build a state-of-the-art facility, integrate liquid cooling from the beginning. This includes understanding thermal requirements, building conditions, access points, and energy-efficiency considerations.
Understand your client. Some customers will only need liquid cooling for specific areas, while others may continue to rely on air cooling. The key is to plan for a realistic payback period that ensures a solid return on investment.
If you already operate a data center, you can still adapt it for high-density computing. It’s not just about adding a chilled-water pipe.
You’ll also need to evaluate your automation system, electrical infrastructure, civil layout, drainage points, and floor load capacity, among other factors. The process is neither simple nor quick—but it is entirely achievable. This guide will help you understand how to make it work.
As AI, HPC, streaming, and other services continue to expand, remember this: demand will keep growing—and liquid cooling is here to stay.
|
|
