Not necessarily, if the training is more efficient, you just throw more data at it for better models which was not possible before