Earlier quoted context omitted.
M2 seems very unlikely to me because that will create product misinformation easily. Imagine the following M1 -- 4 A14 cores M2 -- 8 A14 cores M3 -- 4 A15 cores That "third generation" sounds better, but is actually inferior to the second generation. X1 or M1x seem much more likely. It's the same reason why they made an A12x instead of calling it A13. They probably need 3 designs, but economies of scale begin to be a…
Another possibility is that they skip M1x altogether; if the Pro machines are coming in the autumn and not for WWDC, then the frame will be closer to the iPhones and it would make sense for them to use that later tech. M1 (winter 2020) — 4x A14 cores M2x (autumn 2021) — 8x A15 cores M2 (winter/spring 2022) — 4x A15 cores etc. There’s really no reason for the naming and timing of Pro machines to lag behind consumer ma…
With new nodes, it takes time to get things right. A smaller chip is generally preferred as easier to work with and test. It also gives the node more time to stabilize so your big chips get better yields.
Addressable market is another factor. There are over a billion iPhone users replacing devices every 1-3 years on average. There’s around 100 million Mac users as of 2018 replacing their computers every 5-8 years and that being split between 2 and 3 major market segments.
That means making over 250 million A14 chips, but only around 15 million M1 chips (aside from iPads which are there to improve economies of scale by reducing two designs to one). The M1x probably addresses around 4.5 million computers and the M1p or whatever they call it probably addresses under a half million machines.
A lot of people wanting the performance boost or fearing outdated, unsupported x86 machines will probably cause an increase in M1 demand, but that will be followed by a decreased demand for M2 and maybe even as far as M3 or M4 while demand catches up with the altered purchase cycle. Apple is no doubt well aware of this.
In any case, it’s best to try a new node with a smaller part where you know you need a million wafers anyway. Once you’ve optimized and refined a couple steppings, you move to the larger chip with smaller demand where yields are worse in general.
Personally, I think they would greatly benefit from a unified chiplet design starting with something close to the A14 then scaling up.