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
Let's get to work
Want to build with AI for real?
Beyond the explainer, we design, secure, build and run production AI. Tell us what you have in mind.