They would have to give up on the “unified memory architecture” (their terms) and sacrifice some of the simplicity they get by being able to not care about anything important Being outside the one big pool of fast addressable memory. This would have major impact on their GPU firmware/drivers as it would now need to handle data in the CXL hardware… all if this is extremely normal in the PC / wider hardware world, but Apple are going down the hardware optimisation path like consoles used to, and this overhead can be cut quickly to get the kind of gains that have made the M1 generation seem so much better than the previous intel hardware.

In essence, if they aren’t going to enable high power GPU compute (external AMD/Nvidia cards) for things like machine learning and parallel physics simulations in many industries various analysis software suites… then there’s little to be gained from anything remotely like this, CXL or otherwise.

I have a friend who a Apple Silicon iMac who bought the whole memory tale and it hasn't worked out well for him, so thats why I have an /s