Inference.

running a trained model to produce an output (as opposed to training it). Most of the day-to-day cost of using AI is inference.