The process of increasing AI-driven densification is turning thermal management into a more intricate endeavor, transforming it from a simple engineering task into a sophisticated system-wide optimization process. With the rising density of infrastructure, cooling has evolved beyond just eliminating heat.
On the contrary, every choice made throughout the thermal chain has consequences for efficiency, adaptability, and potential growth. Prior to the advent of AI and increased data density, those engaged in the data center industry were contemplating the same density per rack. Maurizio Frizziero from Vertiv says that with AI, densities are shifting from 10 and 15 kilowatts to 100, 200 kilowatts, and even closer to one megawatt per data center.
In the current DCD>Talks podcast, Maurizio Frizziero, who holds the position of vice president for chilled water solutions at Vertiv, discusses the compelling reasons why AI is leading operators to reconsider the design, operation, and optimization of cooling infrastructure, covering the entire spectrum from deployment to optimization. For a long time, cooling methods were based on fairly consistent rack densities.
Operators can choose and set up cooling systems by considering factors like environmental conditions, redundancy needs, and anticipated workloads. Frizziero claims that AI has significantly altered this equation. He explains that prior to AI and increased data center density, everyone in the industry was considering similar rack densities. “Thanks to AI, the power densities are shifting from 10 and 15 kilowatts to higher levels such as 100, 200 kilowatts, and even approaching one megawatt per data center.”
