Comment by sbarre

A question I don't see addressed in all these articles: what prevents Nvidia from doing the same thing and iterating on their more general-purpose GPU towards a more focused TPU-like chip as well, if that turns out to be what the market really wants.