How to Optimize Cooling for AI-Powered Data Centers

Data center equipment density is higher than ever, a trend sure to continue as demand for artificial intelligence (AI) and high-performance computing (HPC) soars. This trend is forcing data center operators to rethink their infrastructure and cooling systems at a time when energy optimization and meeting sustainability goals are top priorities.

Traditionally, data centers used air-based cooling methods such as in-row cooling and fan walls to optimize temperatures. However, as average cabinet power densities exceed 15 kW, legacy methods typically don’t provide sufficient heat dissipation. Thankfully, there are intermediate steps data center operators can take to improve cooling efficiency and ease the transition from air to liquid cooling.

In a recent webinar, AI-Ready Data Centers: Uncovering the Essential Infrastructure Upgrades, Legrand’s Saman Berookhim, Senior Product Manager, and Klaus Dafinger, Cooling Marketing Manager, discussed how AI workloads drive innovation in containment and cooling.

Evaluating hybrid cooling technologies

To support AI workloads, operators should consider different cooling technologies, including hybrid approaches combining air and liquid cooling. A phased approach makes the most sense. Operators should first focus on managing airflow at the cabinet level through measures such as installing blanking panels, perforated doors, air dams, floor seals, and skirts.

Next, operators should focus on mitigating thermal hotspots to improve efficiency. If left unaddressed, hotspots can damage equipment. One way to address them is through cabinet placement. For instance, placing too many high-consumption cabinets in close proximity could obstruct airflow and lead to uneven cooling. 

Optimizing airflow with sensors

Sensor technology is essential to maximizing airflow efficiency with cabinet placement. Sensors monitor and collect data on conditions such as temperature, airflow, and humidity, enabling operators to identify hotspots, improve efficiencies, and avoid mixing cold and hot air. As such, data insights support containment strategies to optimize energy consumption and heat dissipation as data-intensive workloads increase.

Rear door heat exchangers

Rear door heat exchangers, which use air-assisted liquid cooling to dissipate heat from cabinets, also can help with the transition from air cooling. When added to cabinets to absorb heat from air exiting servers, rear door heat exchangers can eliminate hotspots in almost any scenario. 

Rear door heat exchangers enable data centers to support the next generation of AI advancements while maintaining high performance and efficiency standards. Exchangers deliver multiple benefits, including enhanced cooling efficiency, lower energy consumption, scalability, space optimization, reliability, and cost-effectiveness.

Meeting decarbonization goals

No matter what combination of cooling systems they use, data center operators must also comply with decarbonization mandates as governments and organizations seek to mitigate climate change. Cooling can account for up to 40% of a data center's total energy consumption, highlighting its significant impact on operational efficiency and energy costs. To achieve sustainability goals while supporting data-intensive technologies, operators must optimize power usage effectiveness (PUE) by leveraging cooling methods that operate more efficiently and require less energy. 

In addition, decarbonization regulations require operators to navigate complex requirements that differ from country to country and state to state. Data center teams must design and implement solutions that comply with or surpass local regulatory requirements. However, there are rewards for meeting efficiency targets, such as utilities in locations like Chicago, Virginia, and California, for example, starting to dole out credits to organizations that employ responsible energy practices.

Preparing for now and the future

As AI technology continues to evolve, the demands on data center infrastructure will intensify. Now and in the future, data center cooling will require adaptability and the adoption of cooling solutions such as rear door heat exchangers and immersion cooling. To prepare for AI, operators should evaluate current capacity against anticipated future growth in order to plan a phased implementation that gradually integrates liquid cooling. Measures like direct-to-chip liquid cooling and liquid-to-air CDUs can provide intermediate, hybrid options.

Learn More

With AI and HPC gaining traction, data centers need a comprehensive, forward-looking approach to cooling. Gain additional insights by watching the webinar AI-Ready Data Centers: Uncovering the Essential Infrastructure Upgrades. This on-demand webinar delves into additional detail on cooling and highlights critical infrastructure changes that support high-density computing. Need additional help with getting your data center AI-ready? Legrand brings extensive expertise in delivering innovative data center cooling solutions, offering tailored strategies to help future-proof your infrastructure and meet evolving demands.