UC Davis

The rapid expansion of artificial intelligence, machine learning, and cloud computing is driving unprecedented growth in datacenter energy demand. Traditional air-cooling systems, which can consume up to 40% of facility electricity, are increasingly inadequate for high density racks exceeding 100 kW. Liquid cooling technologies present a significant opportunity to slash datacenter electricity consumption and peak demand, but their performance has not been systematically evaluated. This paper presents insights from a market characterization and energy modeling study of liquid cooling technologies, comparing single-mode and hybrid approaches to baseline air-cooled datacenters. The energy modeling framework integrates component-level simulations with room scale computational fluid dynamics. Server fans, heat sinks, Direct to Chip (D2C) cold plates, and rear door heat exchangers were modeled to quantify airflow, pressure drop, and pumping power requirements. Each cooling strategy was evaluated for the percentage of rack power consumed by the secondary cooling loop while maintaining chip case temperatures below 70°C for the same facility water temperature of 30°C, keeping the primary side energy consumption percentage similar across the various strategies. Results show that D2C provides the lowest percentage of secondary loop energy use (~ 93% lower than baseline air-cooled strategy) and the highest compute density( ~ 15 times of highest density baseline air-cooled strategy), although it comes with higher cost, greater maintenance complexity, and limited retrofitability. Hybrid liquid‑cooling strategies offer a more balanced approach that improves retrofit feasibility while moderating cost and efficiency trade‑offs. Market analysis highlights that commercial D2C systems are deployment‑ready, while immersion technologies remain niche and 2-phase systems are emerging. The study demonstrates how advanced liquid cooling technologies can meet rising compute demands while impr30oving energy efficiency.