[streetcat1]

The end game for tensor flow is auto ml not cloud computing (which is a commodity by now).

At this point most of the model training is commodity. The game will shift to auto ml services.