TSMC cant afford to let Intel steal Apple’s foundry business
it really isn't about that . Most likely Apple 'asked' for this N3B variant that is a bit too aggressive. Intel Foundary could not have delivered something equivalent by Feb-June this year at all. Intel wasn't even an option.
'25-27 maybe. But for the next two years, nope. intel doesn't even have the basic EUV fab equipment to do the volume job more so then their fab process design.
Decent chance the problem is more so that Intel dropped out of
using N3B. ( Along with every other major player, but Apple). Apple isn't big enough to carry a whole process node variant by themselves for the normal process lifecycle length. As much as folks like to spin 'Apple gets fab nodes blocked from other players' , TSMC can't really run a business like that. The other folks buy up early wafers also and the 'bad die' costs are spread out over several players. The problem this time is that everyone else 'ran away'. N3B 'bakes' very slow ( need to put in relativelly precise orders well over a quarter in advance to have the right amount of product) , and most vendors don't have the inventory control options and relatively steady/predictable demand that Apple has. It is also relatively expensive ( and folks don't have margins that Apple has , or the elasticity to command higher prices from their customers. especially with a demand bubble bust on the horizon back in 2020-22. )
Because TSMC couldn't find someone else to 'sharet he pain' Apple is making TSMC soak much of it up. In turn , TSMC is probably going to shift some 'pain' to Apple later (2024-2025). Apple is probably just shifting when the cost rolls in over time.
Pretty good chance Intel hiccupping CPU/GPU design process over last 2-3 years caused them to shift to N3E (which costs incrementally less (both in raw wafer costs and 'early' defect rates') and not quite as long 'bake' time (still longer than N5 family but much closer) . Same with AMD , Nvidia , etc.
