“Federico believes that any AI company producing a large volume of tokens with a product to optimize should train its own models.”