Here's the thing, though: You have to assume that they are correct. If one believes that they are incorrect, he is effectively stating that he (someone thinking abstractly in the absence of any hard data) is better at pricing the value of the work than the manufacturer (who is directly affected by the outcome and has a wealth of hard data).

Several commenters have basically claimed that the manufacturers are behaving irrationally; a claim like that requires some powerful arguments to credibly support. I haven't seen much of that yet here.

I suspect that the truth is that manufacturers would be willing to pay to train workers if they could be reasonably sure that the investment would have time to pay off. But thanks to Moore's Law and everything that comes with it, manufacturing technology is changing far more rapidly than it used to when everything was more stable.

It seems likely that it's more cost-effective for the manufacturer to lower that risk by training lower-paid labor elsewhere and letting those workers go (or retraining them, if it's cheap to do so) when they're no longer suitable, than it is to do the same thing in first-world countries for much higher cost.