Machine learning is now driving AI, and the first word in Machine Learning is machine. Thus, we need even faster computers to enhance AI, but the slowing of Moore’s Law means conventional machines are barely improving. In order to expand the impact of AI, we are forced to design machines specifically for AI. Google is a leader on this new architectural path with the announcement of three generations of Tensor Processing Units (TPUs), with the first deployed for the cloud in 2015. The co-design of machine learning specific software and hardware let production machine learning applications scale to 1000 chips with almost 1000 times the performance and deliver 10 times the performance/Watt of the most efficient general-purpose supercomputers.