For inference we do not care about training, right? But if we could make cheaper inference machines available, everyone would profit. Isn't it that LLMs use m

For inference we do not care about training, right? But if we could make cheaper inference machines available, everyone would profit. Isn't it that LLMs use more energy in inference than training these days?