I don't think the difference is about the source of the money. It's about whether the returns are, to use terms from machine learning, in exploration versus exp

I don't think the difference is about the source of the money. It's about whether the returns are, to use terms from machine learning, in exploration versus exploitation. If the point of these expensive weapons systems is to inefficiently defeat imaginary enemies, then clearly they are a waste of money and the real motive is corporate welfare. But if the point is 'we don't know what the world will be like in 2050, and the only way to find out and start preparing for it is to build a lot of hardware that doesn't end up getting directly used' then that could be a valid justification. I'm not making any claims about whether that actually is the justification. I'm just saying that if it is, it's the one that would work.