If we had a vibrant open source TPU interface, inference / training stack... what would be left of Nvidia?
We have that for GPUs and nVidia were doing pretty well before AI hardware started being popular.
We have that for GPUs and nVidia were doing pretty well before AI hardware started being popular.