I absolutely believe inference is cheap and efficient, but the net impact of training the model should be also added to the equation. Not just the cost of infer

I absolutely believe inference is cheap and efficient, but the net impact of training the model should be also added to the equation. Not just the cost of inference. Is this information readily available somewhere?