Isn’t that already the case, with the capacity required to train these models?