M2, M3 etc will still need to be ARM compatible chips but Apple can of course add specialized hardware and hide difference through their higher level APIs such are Core ML ARKit, AVFoundation, Core Image etc.

I don't think x86 is prevented from adding more AI into the operating system. But it will naturally be harder to utilize specialized hardware such as Neural Engine on the PC platform until standards emerge for matrix multiplication accelerators akin to Neural Engine.

Because such changes have to be coordinated between numerous companies, this can take a lot of time. Meanwhile Apple has an opening to gain leadership in this area as they can move faster with their full vertical integration.