> We know Google isn't.

Google isn't internally, so far as we know. Google's hyperscaler products have long offered CUDA options, since the demand isn't limited to AI/tensor applications that cannibalize TPU's value prop: https://cloud.google.com/nvidia