The thing is, though, that tying design and manufacturing together has been really good for Intel. Historically Intel's performance in terms of driver current and so forth at a given process node has been better than its competitors and it has reached those nodes more quickly. Intel manufacturing paid for that lead partially by accepting more restrictive design rules for transistor layout than a merchant silicon shop could ever get away with because they knew they only had one customer and one design they had to make work. With Intel's numerous missteps at 10nm that lead is over and the manufacturing debacle is taking down the design too but there's no reason to think Intel would be starting from such a dominant place if they had split up. If I were to go back to 2013 I'd tell them "10 nm is too early for cobalt interconnets!" rather than "Split up your design and manufacturing".

And really you could also look at the problem as the design team being too wedded to x86 and unwilling to consider other ISAs just as much as the manufacturing problem. Intel could have done an ARM processor for the iPhone but they wanted to make a go at an x86 mobile chip instead. And while the x86 tax is small potatoes when you're paying for a deep OoO pipeline it's pretty noticeable in the world of in order processors or graphics processing units.