AI turns liquid cooling into a data center must-have solution

Artificial intelligence (AI) has upped the ante across all tech arenas, including one of the most traditional ones: data centers. Modern data centers are running hotter than ever – not just to manage ever-increasing processing demands, but also rising temperatures as the result of AI workloads, which sees no end in sight. Two out of three IT leaders expect AI spending to increase this year, making it a necessity to find new ways to cool data centers.^[1]

Liquid cooling is rapidly gaining traction because it is the future of data center success. A Spring 2024 survey among 812 IT professionals^[2] found that 38.3% expect to use liquid cooling infrastructure in their data centers by 2026. That’s nearly double the percentage from a similar survey conducted at the beginning of 2024, and it is likely to continue this acceleration path over the next few years.

A must-have capability 

Liquid cooling is a simple concept: A circulating liquid removes excess heat from the unit. More specifically, direct-to-chip liquid cooling uses water as the primary coolant, whereas other cooling techniques – such as immersion cooling and traditional air cooling – have limited capabilities. Future-forward data centers have used liquid cooling for decades, but the newest chips – which can reach temperatures as high as 2,000 watts, or approximately 20x what basic computer chips conduct – demand new solutions. Intel, NVIDIA, and other AI chipmakers are focused on increasing processing power, which in turn increases the amount of heat released. Coupled with industry demand for the chips, liquid cooling solutions aren’t just wise, but a necessity for efficient, sustainable, and reliable data centers.

Liquid cooling is a must-have data center capability for three primary reasons, including the need to:

• Address the increasing power densities of modern data centers
• Accommodate the latest and hottest chips 
• Reduce energy costs and achieve greater sustainability

The continuous nature of AI workloads means they typically can’t pause mid-query or save their place during an outage. “Unexpected downtime caused by overheating would be catastrophic,” says Gary Rowe, Executive Chairman at Chilldyne. “The potential for hardware damage from coolant leaks is also enormous. With individual AI chipsets costing over $30,000 each, unreliable liquid cooling methods are both impractical and unaffordable.”

Liquid cooling implementation

These requirements also place importance on having the right partner for a liquid cooling solution. For example, Chilldyne is an innovator in the liquid cooling sector because it uses a proprietary, leak-proof, direct-to-chip system called negative pressure technology. 

Liquid cooling can remove heat 1,000x better than air systems, but negative pressure liquid cooling specifically pulls the coolant through the system rather than pushing it. This mitigates the risk of water leaks and protects chipsets from overheating. 

Chilldyne’s solutions can support both AI and traditional high-performance computing (HPC) chips, managing heat loads of 2,000 watts and beyond per individual processor. In addition, the company has expertise in architecting new AI-driven data centers, as well retrofitting existing ones, to accommodate liquid cooling and provide “unprecedented reliability, efficiency, and sustainability,” Rowe says.

The technology is so vital that the company recently was awarded a grant from the U.S. Department of Energy’s Advanced Research Projects Agency-Energy (ARPA-E) to support its ongoing development of energy-efficient cooling solutions. Chilldyne’s cold plates with turbulators rated at 2,000 watts and automated coolant quality (ACQ) are specifically being supported by the DOE. 

Leadership means laying a foundation not just to handle today’s capacity, but also for the next generation of tech needs. With AI quickly increasing system demands, forward-thinking data center managers are leaning into liquid cooling now.

Visit Chilldyne for more information.

^[1] Foundry, 2023 AI Priorities Study, https://foundryco.com/tools-for-marketers/research-ai-priorities/

^[2] The Register, Spring 2024 survey, https://www.theregister.com/2024/04/22/register_liquid_cooling_survey/