[thatguysaguy]

TPUs predate LLMs by a long time. They were already being used for all the other internal ML work needed for search, youtube, etc.