While I've yet to get back to you on your IDM is a doomed business model thesis, we can simplify your observations to two things:
Intel's only very long term success is the line of CPUs that started with the 1971 4004. Even though it was founded to do memory because of the coming disparate volume of that vs. logic, and reached HVM with the first major commercially available 1103 1K bit DRAM also in 1971, it wasn't super competitive in the market in the 1970s and dropped out in the 1980s. It did do other types of memory which it often developed, and as you note that included flash recently, but nothing super long lived.
Every other CPU it did either failed, or it eventually quit the market. A possible exception is microcontrollers from 1976 through 2007, but I'm not sure how significant they were to the bottom line after the IBM PC started a new phase in the industry with the 8088.
Touching on our IDM discussion:
Intel didn't adopt EUV technology until 2023 (4 year after TSMC) because EUV didn't match Intel's product roadmap.
It would be more accurate to say Intel's roadmap got stuck in a huge pothole with 10 nm.
Although I agree in part, unlike TSMC N7 which in due course had a N7+ variant with some EUV use that started their production experience, 10 nm was more aggressive than N7 and never planned on adding EUV.
But they were planning on EUV for their next node, which ended up being the Intel 4/3 family. But didn't that get delayed by the 10 nm debacle because they depended on technologies developed for 10 nm, which took so long to get out of the lab?
We could also discuss the misses of one sort or another of its successful line of CPUs:
Like the 8085, 286, Netburst, FSB and other memory madness, having to adopt AMD's 64 bit macroarchitecture, problems all the way to catastrophes stemming from verification and simulation after Brian Krzanich nuked that, and I infer having to mark time by spending too much energy on switching from their own bespoke CAD to industry standard (Arrow Lake for example), and/or in that generation maybe failing to do well in switching to tile based designs.
We could further simplify this to how one man, Brian Krzanich, between 10 nm and verification and simulation, turned Intel the IDM from being generally successful to generally failing. He doesn't bear all the blame, he inherited, grew up in an increasingly dysfunctional corporate culture, didn't e.g. have much power and I assume influence when Craig Barrett took over from Andy Grove, but he's certainly an inflection point.
