It includes underfloor and overhead integrated liquid cooling designs, hybrid air cooling and liquid cooling, and advanced schemes for heat recovery and energy storage.
Today’s AI data centers face unprecedented cooling challenges. The latest GPUs and CPUs generate significantly more heat than before, while rack densities continue to climb as more processing units are packed into tighter spaces. With rack power consumption exceeding 30 kW, traditional air cooling alone cannot dissipate heat efficiently enough to maintain safe operating temperatures in these environments.
EPG, a Singapore-based manufacturer, supplies liquid cooling for GPU clusters across APAC — direct-to-chip stations, CDUs and pipeline modules as prefabricated PODs.
Direct-to-chip targets the hottest components while keeping servers serviceable — the dominant choice for GB200-class racks. Immersion reaches high densities but adds fluid-handling complexity. EPG is built around direct-to-chip cold plate with hybrid air–liquid support.
Above ~30 kW/rack, use direct-to-chip liquid cooling for 75–80% of the heat plus air cooling for the rest — supplied by EPG as modular PODs.