Latency
The time something takes to happen: a network message to arrive, or a model to answer.
Also: delay
Latency is delay, measured in microseconds for networks and in milliseconds or seconds for model replies. In training clusters, low network latency matters because GPUs often wait on each other to exchange results; InfiniBand's main advantage has been its low latency. In inference, latency is what the user feels: time to first token and time between tokens. Physical distance adds latency too, which is why inference providers spread capacity across regions while training can sit wherever power is cheapest.