drag the divider inside the fixed envelope to reallocate power
CPO frees network power in a fixed facility power budget, reallocated to GPU compute fixed facility power budget network share compute share per-switch power: high back-end fabric switch east-west switch east-west switch east-west east-west AI traffic makes the back-end fabric the dominant network power draw GPU capacity freed power reallocated to compute added GPU capacity tokens-per-watt same envelope — freed kW becomes GPU capacity and tokens