Cooling systems move heat away from chips, racks and rooms using air, liquid or a combination of methods. High-density AI accelerators can produce enough heat that ordinary facility designs no longer provide adequate thermal control.
Cooling affects reliability, performance, water or energy use and how much hardware fits in a facility. The compute and cooling designs must be coordinated because adding electrical power to a rack also increases the heat that must be removed.
ELI5
Data center cooling keeps powerful computers from becoming too hot. AI chips turn much of the electricity they use into heat, and that heat must be carried away continuously.
For example, a dense rack of accelerators may need liquid cooling that moves heat more directly than room air alone. Without enough cooling, the chips may slow down, shut off or suffer reduced reliability.
