Paying for LLM inference by the kilowatt-hour instead of per token

AI is an energy problem.

Marketplace success will likely be determined by energy costs as much if not more so than the technical aspects of model development.