I would not discount Musk. He is an intelligent man and will draw on the expertise of other intelligent people to assess the state of the current art, where it is going, and build to create an up to date fab as an extension of that curve with the latest TSMC equipment = 3-5 years with the assorted lead times.
Remember what he did with rocket science? Threw the billion dollar one hit wonders out the window and created reuseable rockets with a high degree of high end printed metal parts involved. Sure he made a lot of scrap = the cost of progress. Now all the old iron is virtual scrap = reuseable is the path.
As for orbital data centers?? Why there? What is the gain/loss assessment of terrestrial versus orbital data centers?? A ground fab seems doable. Orbital data centers is a high cost, zero sum game. No gain, huge added costs.
Elon Musk is at the crossroad of physics and AI, he and the teams at the frontier may just pulloff the flywheel that thing that produces the imagined positronic brain that evolves 20 thousand years and internalizes "machine consciousness" along the storyline of V'GER in Star Trek: the Motion Picture (1979), or Avalon in Passengers (2016).