2. Those Fabs are not leading edge and not target for the general consumer brands. The current target of 5nm in ~2024, a time when TSMC should be on 2nm GAAFET in their main Fab in Taiwan.
3. The only real reason is National Security, which is now being handled by Intel. These chip were used used and Fabbed by IBM and GF before GF's exit on leading edge.
4. 60% of TSMC's customers are US companies, but these US companies 's customer aren't 60% Americans. And not all Americans chips needs to be Fabbed locally.
I don't think the DoD is willing to put all of it's eggs in that one basket that hasn't been performing lately. Particularly when the DoD sees massive amounts of edge inference being a key weapons component going forward and they need newer nodes than 14nm+++++++.
Hard to blame them.
That is of course, as you said unless both Intel and Samsung seriously messed up their plan.