If Apple were to start doing their own fab they would likely start with some of the simpler support chips and work their way up in complexity through networking before getting to the CPU. This fits their historical pattern of quiet experimentation with new technologies well ahead of large investments. Bringing PMIC and Bluetooth fully in-house, for example, could yield benefits in supply predictability and give them more design freedom for low power wearables. Later they might do the cell modem. Main CPU is a harder sell, letting an external partner like TSMC handle it allows them to move on quickly while the partner sells the older process capacity to AMD, Intel, etc.

Not the most likely scenario in my opinion, but it's how they would do it.