Cooling methods that use liquid (direct-to-chip cold plates or immersion) instead of air alone to remove heat from high-density compute. Above roughly 50 kW per rack, liquid or hybrid cooling becomes necessary for AI hardware.