Liquid cooling, in data centers, is the use of water or another coolant to carry heat away from servers instead of relying only on fans and chilled air. The two main approaches are direct-to-chip cooling, in which cold plates sit on processors and memory with coolant pumped through them, and immersion cooling, in which whole servers are submerged in a non-conductive fluid. Heat is then passed through coolant distribution units to cooling towers or chillers.
Liquid carries heat far more effectively than air, which matters as racks of AI accelerators such as GPUs draw much more power than traditional servers. Air cooling struggles at those densities, so many new AI data centers are designed around liquid systems. Liquid cooling can also improve energy efficiency and reduce water use, depending on the design. Suppliers of pumps, cold plates, heat exchangers and thermal management equipment are often grouped as an AI infrastructure theme in thematic investing.
Ref 1300000266