I agree with everything you said but the TSMC bit. They are quite literally competing with NVidia et al for fab space for customers chips. Sure they get the AI bits built-in to existing products but surely they’re bigger/more expensive to manufacture and commit from TSMC because of it.
I was trying to say that there was still a cost to using their own chips for server AI because they still had to pay to have them made so they weren’t “free” because they’re Apple products as opposed to buying nVidia parts.
You’re right, there is a cost to them to put the AI stuff on end user chips too since die space isn’t free and extra circuits mean fewer chips fit per wafer.
