GPUs are parallel compute engines. The difference between a high performance CPU core design from Intel/AMD/Apple and the low end stuff is a bunch of fancy branch prediction, out of order execution, cache hierarchies, etc. all designed to improve single-thread performance. The primary difference between a large GPU and a small GPU is that a large GPU has more cores.

Sufficiently far in the past you might have been able to get away with an integrated GPU that didn't even have meaningful 3D acceleration etc., but those days are gone. Even web browsers are leaning on the GPU to render content, which matters for iGPUs for battery life, which makes performance per watt the name of the game. And that's the same thing that matters most for large GPUs because the constraint on performance is power and thermals.

Which is to say, if you're already doing the work to make a competitive iGPU, you've done most of the work to make a competitive discrete GPU.

The thing Intel really needs to do is to get the power efficiency of their own process on par with TSMC. Without that they're dead; they can't even fab their own GPUs.