Look at the architecture of AMD SoCs in the PS5 and Xbox. Obviously Apple cannot get away with unified memory in the higher-end machines, but what would make them prefer an AMD dGPU when it seems they have everything they need internally. Again, assuming the GPU scales well.
Apple could absolutely choose to make better cooling, but they've demonstrated time and time again that they'd rather just let thermal throttling happen and trust that their users will continue to buy a product that doesn't live up to its spec sheet's promises.
Any cooling that could support dedicated-GPU-level performance would require making the chassis dramatically thicker - take a look at some of the gaming laptops out there today with desktop-class GPUs, they can get be like 2 inches thick in pursuit of cooling good enough to support a full-power GPU without massive throttling.
The Radeon Pro 5600M with 8GB of HBM2 has a pretty significant power envelope, and is coupled to 6 or 8 core Intel H series that are equally power hungry.
The fact 16" MBPs are undercooled is beyond the point. Apple ships laptops with a ~95W TDP. Why would they not ship a SoC with 16/24/32 graphics units for high-end laptops and iMacs?
In that case, you're not talking about a Radeon Pro 5600M but rather a desktop-class RTX 3090 with 24GB of high-performance RAM and a TDP of 350W. A MBP couldn't hope to power or cool one of those with a built-in GPU, never mind come close to matching the performance or supporting CUDA without nVidia sueing Apple out of existence.