What Is Hybrid Cooling?_
Hybrid cooling uses cold plates attached to high-heat components (GPUs, CPUs) to remove the majority of heat via liquid, while remaining lower-power components (VRMs, memory, networking) are cooled by rack fans or facility air. This approach reduces reliance on full-immersion or single-phase liquid systems and allows gradual adoption of liquid cooling in existing data centers. It requires careful airflow management to avoid hot spots where air-cooled zones overlap with liquid-cooled zones.
Technical Details
Direct-to-chip cold plates circulate coolant (typically water or dielectric fluid) through a closed loop, with heat rejected via a CDU or facility water loop. Air-cooled components must still meet their OEM-specified inlet temperature limits, so rack airflow paths are often partitioned with blanking panels and ducting. The liquid loop operates at low pressure (below typical limits, e.g., ASHRAE or vendor specifications) and uses quick-disconnect fittings for serviceability. Hybrid cooling is common in H100 and B200 racks where not all components are liquid-ready, but becomes less common in GB300 NVL72 designs where nearly all heat is liquid-rejected.
How Leviathan Systems Works with Hybrid Cooling
In field deployments, hybrid cooling appears when we retrofit air-cooled racks with liquid cooling plates for GPUs only, leaving networking switches and power supplies on air. We must verify that the rack's airflow direction (front-to-rear) and fan speeds are adjusted to compensate for the reduced air flow from the GPU zone.
Related service
Liquid Cooling Services →