Nvidia makes most of their money selling GPUs as computational accelerators-- both generic CUDA and neural network applications. They don't _need_ to profit of

Nvidia makes most of their money selling GPUs as computational accelerators-- both generic CUDA and neural network applications. They don't _need_ to profit off their ML models. It's a value added service. Ensuring their hardware has marketshare at the leading edge of the ecosystem is the main point.