Hey anon, glad to see your project thread, welcome. Seems like you have good insight on costs. This stuff will be expensive, one way or another.
I also like the idea that good AI is a vital initial phase for your project. Sad to say, from my perspective at least, it's also the most difficult
problem (in the long
list of problems) to solve when inventing good robowaifus. But I think with everybody pitching in ideas, then some geniuses among us might pull of a breakthrough or two. We'll see. But as you said, we certainly do not
want botnets infesting our beloved robowaifus.
And yea it's kind of a theme here "Waifubots, not slutbots", crude as it is. Why some big corporations are not already making cute waifubots is quite beyond me tbh. The pent-up demand is very plainly a huge one.
Obviously full-sized will be the most desirable, but I think for the first several years scale models are the only practical approach while keeping the costs relatively modest. Danny Choo's injection-molding line when it's in full swing will probably be a good model to look at for a BJD approach.
There's lots of challenges ahead. Good luck to us all anon.