So you mean that the whole "Intel keeps the foundry margin" is a myth. Then it should be backed up by numbers. I have some, and they say quite the contrary: we're at the end of Moore's law for foundries, not for Intel:
Intel And TSMC: The Unwinnable Struggle.
Now that I think longer about it, however, I do think you have a point, but it simply doesn't match reality. In any case, Intel uses its process lead to be able to manufacture cutting edge processors no one else can build, so it generates more money, on top of the higher margins of the already cheaper transistors of smaller nodes. That money is then put into the process R&D and capex to extend their lead.
A foundry like TSMC, on the other hand, generates money by selling wafers. ARM is able to build a widely used ISA and architectures although its revenue is minuscule compared to what you need for a bleeding edge fab. Qualcomm builds similar parts with probably a relatively similar budget, but generates loads of money because it has most of the mobile market share. A huge amount of the money goes to the company that doesn't need is as much as TSMC, who built the thing. So there are huge inefficiencies. Image if TSMC could use its own money and that of Qualcomm.
Just like selling the separate parts of a car don't really cost much, but the car manufacturer adds added value to the parts by making a car out of it. Intel, unlike TSMC, not only builds the car, it also sells the car, so it gets access to the money of the added value the car is extrinsically worth.
It turns out, in the semiconductor industry, IDMs are much more efficient and thus much better able to compete.
And since Qualcomm isn't one, it goes with its wafers to TSMC and pays the foundry toll, which effectively increases Qualcomm's price/transistor and thus decreases margins.