Skip to content
AI energy · Topic

Training vs inference cost

Understanding AI energy means separating the one-time cost of training a model from the ongoing cost of running it. The balance between them shapes both energy strategy and where the money goes.

Key facts

  • Training is a large, concentrated burst of energy spent once per model version.
  • Inference is a smaller per-event cost paid continuously for the life of the deployed model.
  • For widely used models, lifetime inference energy can exceed the original training energy.
  • Optimizing inference (model choice, hardware, serving) is where most ongoing energy savings live.

Takeaway

For popular models, the ongoing energy of serving them eventually outweighs the one-time cost of training.

Explore other topics