It has already. Chips that are ostensibly for graphics are containing more and more silicon dedicated to general purpose workloads. Turns out doing it like this this is usually cheaper than manufacturing dedicated parallel processors and GPUs separately, since the requirements have so much overlap.