A bit more detail
Inference happens every time a user prompts a model or an application queries it. It differs from training, which is the initial phase where a model learns from data. As adoption grows, inference typically accounts for the majority of ongoing computing costs.