Everything they showed today, from the vision models to the path planning to labeling to simulation was bleeding edge. And they seem to have managed to put everything together in what appears to be a tremendous engineering effort. I can't see why they can't use a similar playbook for the humanoid robot and increase value over time. The vision framework they used for self-driving is fairly general and can probably adapted to humanoids. The action policy space is less re-usable but I suspect they have clever ideas on how to tackle that.

Musk may be bad at forecasting but he's darn good at selling stuff that half-works and then assembling a team that can execute on his vision.