TDP (Thermal Design Power)

TDP (Thermal Design Power) is the maximum sustained heat, in watts, that a processor or accelerator is designed to dissipate, and the figure cooling systems are sized against. It is not the same as peak electrical draw, which can briefly exceed it. Server CPUs sit in the 150-400 W range, while current AI accelerators are rated at roughly 700-1,200 W each, so an eight-GPU server can exceed 10 kW before networking and fans. Multiplied across a rack, TDP is what pushes density from 5-10 kW per rack in enterprise rooms to 40-130 kW in AI halls, and is the reason direct-to-chip liquid cooling has become standard for new GPU deployments. Operators use vendor TDP, derated by expected utilisation, to size power distribution and contract committed kW.