[Operations Guide] How kt cloud Data Center Chillers Work and How to Optimize Cooling Efficiency

As AI and HPC demand surges and rack power density climbs above 40~100kW, thermal management in data centers has become a core challenge that determines availability. If heat is not removed quickly enough, it leads to performance degradation from throttling, and can even result in equipment shutdown.

Chillers are broadly divided into two types.

• Air-Cooled Chiller: Releases heat into the atmosphere using large fans. Simple to build, but efficiency drops sharply in high outdoor temperatures and there are limits to handling large-scale capacity.
• Water-Cooled Chiller: Removes heat using cooling water via a Cooling Tower. Requires substantial initial piping work, but offers high COP and overwhelming cooling capacity, with 1,000~2,000RT-class turbo chillers used in large data centers.

Inside a chiller, refrigerant repeats a four-stage cycle of compression→condensation→expansion→evaporation. The evaporator absorbs heat from the data center's return water to produce chilled water at around 7°C, and this chilled water passes through CRAC/CRAH units in the server room to absorb heat from the servers before returning to the chiller—a central chilled-water cooling structure.

Key operational points for optimizing cooling efficiency include raising the chilled water supply temperature (CHWS), managing condenser water temperature and cooling tower efficiency, operating multiple chillers with consideration of part-load efficiency (IPLV), and hot aisle/cold aisle separation and airflow management. Continuously monitoring COP and PUE metrics to reduce unnecessary energy consumption forms the foundation for stable service operation.