The Verge published an interesting piece this week on why Apple (at least as the rumor goes) will replace the use of Intel processors with its homegrown ARM-based processors. The author asserts that it is because Intel is standing still. That is arguably the case, but I see this as part of a bigger trend. We’ve reached an inflection point where it broadly makes more sense to do custom chip designs than use off the shelf components.
The use of custom chips (ASICs) for very specialized purposes has been with us for decades. For example in the 1980s the use of a custom ECC chip allowed DEC to go from a virtual unknown in disk drives to an industry leader. But in general Moore’s Law made custom chip design a losing proposition. The rule of thumb was that any performance advantage you could get from the custom design would be exceeded by the next generation of general purpose processors. The timeline was simple, a new manufacturing process generation would be introduced simultaneously with a new general purpose processor. Availability of that process generation for custom designs would follow sometime later. You would get your custom chip designed and in production in that process just months before the semiconductor companies would introduce a new process and processor. Your custom chip that looked so good on paper suddenly was expensive and offered little benefit over the newly introduced general purpose processor. Your project would either not make it out the door, or the next generation of it would forgo the use of custom silicon. I saw many attempts at custom chip designs fall to this cycle.
Many things have changed over the decades. The gap in when process technology leaders like Intel had a generation available for its own designs and when that generation was available for custom designs shrunk (or disappeared). The rise of chip manufacturing Foundries (also key to the rise of ARM), related by their going from being generationally behind on processes to being competitive with Intel. Indeed Intel itself has gotten into the Foundry business. The availability of licensed architectures, licensed core designs and components, and the design tools to use them lowered the engineering costs of custom chips. Etc. But most importantly, the rise of users with sufficient volumes to justify custom chip designs.
Apple sells enough devices to justify custom designs and gain early access to the latest process technology. AWS and Azure also have that kind of buying power and business justification. AWS Nitro uses a custom chip (done by its in-house design team) to great advantage, including allowing a “bare metal” general purpose computer to operate in, and take advantage of all the benefits of, the AWS infrastructure. Apple uses its chip design capabilities to get higher performance than off the shelf ARM chips, and to provide features specific to its unique user experiences.
Once you have a great in-house design team, and know how to get the best out of partner foundries, the question of where else you can get advantages out of custom chips is on the table. Are all the ideas we had back in the 90s for using custom chips to speed up databases, which fell victim to the Moore’s Law/General Purpose Processor cycle, now back on the table? They (and newer ideas of course, like using GPUs to speed query optimization) should be.
By the way, I’m a little skeptical on the rumor about Apple fully moving from Intel to its own ARM-based design for the Mac. It makes sense for the MacBook, but not the iMac/Mac and particularly the Pro versions of them. It doesn’t matter if I’m right or wrong in the short-term, any move away from Intel x86s to Apple custom ARM-based processors in the Mac line foretells the day is coming when custom chips power the entire lineup. And that is something that could easily spread to Windows-based PCs as well.
Hal's (Im)Perfect Vision 
It is possible that Apple will add an ARM as the primary chip, and have a second Intel chip for when heavy lifting (big x86 binaries) is needed. The kernel + OS would primarily run on the ARM processor, so the Intel chip would only be running some programs, and not doing housekeeping like USB, interrupts, display driving, networking etc. That would let them get better performance on that same Intel chip compared to other operating systems (which also have to run their housekeeping on it incurring an overhead) and would also let them use a smaller cheaper Intel chip since they won’t need so many cores.
On the high end they would still need a high-end x86 to get the right performance, so the costs would be out of control. On the low end x86 emulation (ala what Microsoft is doing with ARM) is probably acceptable, so you don’t need an x86 “co-processor”. The real key becomes how quickly they can get partners to create ARM native versions of applications, and I would guess their community will move quickly. Emulation is just about the long-tail, such as apps no longer actively being developed. My best guess would be they maintain dual architectures for a while (like years), continuing on with x86 on the high end while replacing the x86 with a custom ARM-based processor on the low end.