Who are you, who is so wise in the ways of science?

And they would’ve gotten away with it if it wasn’t for you meddling kid!

Google

Out of curiosity, why? They have their own TPU which they claim to be quite efficient. Is it because they can’t produce enough? Or because they have to resell NVIDIA for their own cloud, Google Cloud, to customers because they prefer to stick to CUDA? Or something else?

Dell, HPe and Supermicro. System integrators are buying shitloads to resell.