Unlike earlier reports, this research provides the first openly available reference design for two-phase cooling systems. By grounding its findings in performance and cost data, the study lays out a clear case for operators considering the move away from single-phase setups.
The benefits go beyond incremental savings. Two-phase systems are shown to handle higher compute densities, enabling operators to run more GPUs within the same power envelope—a critical factor as demand for AI-ready infrastructure accelerates.
The financial impact is substantial. A 10MW reference design modeled across multiple climates found annual operating expenses up to 35% lower than with single-phase systems, and a 12% drop in total cost of ownership over five years—without requiring significantly higher upfront capital.
When scaled, the numbers become hard to ignore. In Texas’s Austin-San Antonio corridor, converting new facilities to two-phase cooling could trim more than $50 million in electricity costs each year, enough to supply power to roughly 330,000 homes. Extrapolated across North America’s 52.4GW pipeline, the savings approach $10 billion annually.
The technical findings are equally significant. Two-phase cooling can operate effectively at water temperatures around 8°C higher than competing designs, opening the door to more free cooling and reduced reliance on energy-intensive chillers. The reference design also showed that these systems need less heat rejection infrastructure—12 fans per chiller instead of 16—and consume no active water for the cooling process, easing concerns in regions facing water scarcity.
The study addresses long-held assumptions that two-phase cooling is overly complex or impractical. It demonstrates that facilities already planning single-phase direct-to-chip systems can transition without large-scale redesigns. This practical pathway may lower the barrier for adoption as operators search for solutions that meet AI’s growing energy appetite without overwhelming local grids.
