NVidia has a far greater complexity to handle. GPUs do a lot of things and are more generalized with each iteration. In comparison ASICs are super simple chips.

NVidia has a far greater complexity to handle. GPUs do a lot of things and are more generalized with each iteration. In comparison ASICs are super simple chips. If GPUs would still use the fixed rendering pipeline they started with ( only push your triangles and texture to the GPU and you get your image) they could also handle their business with a few hundred employees. Just shrink your chips to the latest fab size, add more computing cores, maybe improve some algorithm if a better solution is found.