I wonder if it would be possible to have an heterogeneous processor with a mix of arm and x86 cores. (They would probably have to partner with AMD which will produce x86 chiplets for them)
How it'd work - ARM SoC expands on what the T2 does today, except it now runs all system threads, any ARM native user space apps, and maybe specialty fixed function blocks that Apple throws on those chips (e.g. media encode). Meanwhile the Intel chip is a co-processor for any x86 specific user space apps, and it otherwise doesn't touch drivers or any system things.
The market would be segmented with ARM chips at the lower end (MacBooks), and ARM chips with Intel co-processors at the higher end (15" MacBook Pro, iMac, iMac Pro, Mac Pro). This is equivalent to how integrated v discrete GPUs are segmented today.
Over time the ARM chips get faster and faster as Intel inevitably blunders, and more and more code is rewritten as native ARM code, so that after several years the need for x86 is significantly reduced, maybe only existing as thunderbolt eCPU boxes and/or Mac Pro. And by around MacOS ~10.23 in 2026, x86 is sunsetted because remaining use cases are so minimal.
That's not really Apple's style, at least not under Steve Jobs ;). You just throw the old architecture under the bus, provide some emulation to make Microsoft and Adobe catch up, all developers have to adapt and build fat binaries or get left behind.
Apple can do this, since no one has a strong expectation of backwards compatibility anyway.
Of course, ARM might not be competitive enough yet for Mac Pros. So they'll probably start with the MacBook Air or 12", require every developer that submits to the Mac App store to build universal binaries, and have some partners (probably Microsoft) that will provide support on day 1. For a chunk of the MacBook Air audience the architecture will be irrelevant, since they do Office, browse the web, and send e-mails. So, if they get a laptop with significantly better battery life doing that, it will sell. Once the Air has prepared the software market, go up the product line after 1-2 years.
I like the Intel co-processor idea, but Intel CPUs are relatively expensive. So, they'd have to jack up prices, while going ARM directly would increase their margins at the same price points.
fwiw, the reason I think the co-processor idea is likely is that Apple has recently re-committed to the Mac and specifically some niche use cases therein (thus they can't fully ditch Intel), and that they're already shipping an A10 equivalent in Macs today (the T2) so not that hard to go a bit further into the water. Also, in single core perf its possible the ARM chips would exceed most Xeons they ship, which clock lower in favor of more cores.
Agree if Apple had its way and could pick its markets perfectly, they'd have slowly shed niche use cases in favor of the highest margin markets (which is mainstream not pro), and thus at the appropriate moment dumped Intel and gone all-in on ARM. However, they're seeing the traditional pc market only moves so fast, there's many many pro niches they didn't really consider enough, and those niche Mac users really do have a lot of indirect value to Apple's brand. Emulation wouldn't cut it for those users - it'd only work if Apple saw ditching those markets as an acceptable tradeoff.
Intel CPUs are expensive but so are GPUs, and they'd only be bundled in the higher end machines - nothing that's below $2500-3000.
Well, that's my hunch anyway.
In theory it seems you could be able to convert "on the fly" from multiple high level instruction sets to a singl low level microcode.
Not sure it's a good idea tho.