Intel had a project called Larrabee where they put a lot of small but general purpose x86 cores on a GPU card. I've heard that it was a flop because certain GPU operations weren't fast enough on those little cores - you still needed some hardware acceleration. It wasn't a total failure though, Intel managed to pivot it from a GPU project into a moderately successful datacenter product, the Xeon Phi. Of course, time has passed and maybe it's worth revisiting the idea again.
The biggest barrier for anyone making a purchasable GPU is the current patent minefield dominated by Nvidia, Imagination, and others.
The biggest barrier for anyone making a purchasable GPU is the current patent minefield dominated by Nvidia, Imagination, and others.