Direct-to-chip cooling

Direct-to-chip (DTC) cooling circulates liquid coolant through cold plates mounted directly on the hottest components — CPUs, GPUs, and memory — removing 70-80% of server heat at the source while fans handle the remainder. Coolant loops connect through manifolds to a coolant distribution unit (CDU) that exchanges heat with the facility water system. Single-phase DTC using water-glycol mixtures is the dominant technology for AI clusters; two-phase variants use dielectric refrigerants that boil on the cold plate. DTC supports rack densities from 50 kW to 150+ kW and is the reference design for NVIDIA GB200/GB300 NVL72 systems. Compared with immersion, DTC preserves standard rack form factors and serviceability, which is why most colocation operators are standardizing on it, typically paired with rear-door heat exchangers or air handling for residual heat.