Synopsis: In this Techstrong.ai Leadership Insights interview, Daren Shumate, managing principal at Shumate Engineering, explains how innovations in cooling technology are becoming essential to supporting AI-driven data center designs.
Shumate explains that traditional air-cooled chiller designs—long the default for most data centers—are increasingly inefficient at today’s AI power levels. In simple terms, more compute now means a disproportionate rise in cooling demand, turning a “gigawatt build” into a much larger grid and infrastructure requirement once cooling overhead is included. The challenge isn’t only macro (finding enough power supply); it’s also micro: high-performance AI racks run so hot that air, as a heat-transfer medium, is hitting its practical limits.
He then walks through the industry’s shift toward liquid-based strategies such as direct-to-chip cooling and the use of cooling distribution units (CDUs) that connect rack-level loops to building systems. The core engineering difficulty is managing different temperature needs across the facility—warmer loops for direct-to-chip systems versus colder water for air-side cooling—without wasting energy.
Shumate also outlines an emerging “hybrid” approach designed to produce multiple water temperatures within a single loop, improving overall energy efficiency and lowering power usage effectiveness (PUE). The implication: better cooling doesn’t just reduce operating cost—it can change the economics of capacity, shrink required electrical infrastructure, and potentially unlock more revenue-producing compute within the same power envelope.
We close with a look at who is financing today’s gigawatt-scale builds, why “AI bubble” concerns keep surfacing, and why Shumate believes the policy debate should focus less on restricting data centers—and more on modernizing the grid that supports them.
