It's interesting how we can see in real time the commodification of ML hardware. From my understanding, prior to Tensorflow, most serious ML projects involved c

It's interesting how we can see in real time the commodification of ML hardware. From my understanding, prior to Tensorflow, most serious ML projects involved clusters of NVIDIA GPUs, and production ML software was tightly coupled to CUDA. Google shifted the software landscape by pushing a multi-target ML platform. Now, all of a sudden, NVIDIA is under intense pressure to keep up with new compute hardware looking to eat their lunch in the AI sector.