Then what should the price be?
I hear this a lot about apple products, but when asking what the price should be you usually get some ridicolous answer that is just the sum of the components (oh and when people try to show a similarly speced windows computer it costs about the same)
Of course the cost of a product is more than the BOM. On top of that is R&D, administration and profit margin to name the big ones. (Plus of the margins of the distribution channel (if needed) + taxes and fees.)
Everyone who is not a complete idiot understands that.
So lets try to break it down. Administration is the same. Physical R&D with the notable exception of the CPU/GPU is the same. Lets assume that the profit margin stays the same - which amplifies the final effects of any cost increases or reductions, also since it propagates beyond Apple.
The difficult part to assess is R&D costs for the SoC. But the Very Nice Thing for Apple is that the large base costs associated with that are already paid for in their iPhone(iPad) development. To a large extent, they "simply" need to connect more of the base units, and possibly add/change the external RAM interface, and pull out a number of PCI-e lanes. While it's not trivial, it's comparatively cheap, and includes modelling and most design debugging. Lithografic mask costs are unavoidable for distinct SoCs, and run in the $hundred million range in total for a SoC. Fixed costs can become an issue for limited run silicon, and we don't know Apples volumes per product, but lets treat the big iMac/Mac Pro separately.
Lets assume a die area for a hypothetical iPad Pro/small Mac A14x of 120mm2 (4big4Little8GPU16NPU). That would make roughly 500 dies on a 300mm wafer. Lets assume a good 80% yield (Apple can do as with the A12x and disable functional units to improve yield vs. defects, and use frequency binning to improve yield vs. process variation). That would give us a raw silicon cost of roughly $25 per die. Lets double that to $50 to give ourselves some safe margins. volumes for iPad Pros + small Macs may be, oh, 40 million annually? So if extending the A14 to the A14x cost half a billion USD, thats just over $10 per SoC. Lets call the whole thing $60. So if the BOM previously was $800 and included an intel processor at $300, you have just dropped total BOM to $540. That's a huge difference for those categories of Macs. Of course the cheapest Macs used cheaper intel processors, so savings will be less on the lowest cost items. Plug in your own numbers for the processors you are interested in, Intel supplies reasonably accurate pricing information.
Next example is a more ambitious SoC for MacBook Pros and iMacs. Lets assume (8big4Little16GPU32NPU256-bitLPDDR5). Die size roughly 200mm2, giving 300 raw dies per wafer. Greater probability for defects per die, obviously, but also more room to disable individual units to increase yield. For simple numbers, lets put usable dies per wafer at 67%=200dies => $50 per die. Lets increase that to $75 for extra headroom for errors. Now the volumes for such a chip is way less than for the A14x. MBPs+iMacs may sell 5 million annually (?). This is where the fixed costs start to factor in substantially, and where that uncertainty as to design costs starts to be a real bother. Half a billion dollars is quite high for extending the A14x to A14MBP, but lets keep it. If the design is used for two years (10 million units), it would add $50 to the cost per SoC, bringing it up to $125.
However, this should be compared to paying intel $xxx for the CPU, and AMD $yyy for the GPU+dedicated memory.
Even with hefty error bars, this constitutes
major savings on BOM for Apple.
This is all back of the envelope, but there really is no doubt that Apple will save substantially on their Mac silicon budget.
