I don’t know of any studies unfortunately, but I did want to point out that training is not quite a one time cost in practice, because training has already been done loads times and is still being done! I’m theory, if we stopped training all AI and just kept the ones we have, then indeed the training cost would be bounded just like you say, I’m just afraid we’re quite far from that.
Just a little note about the word “model”, in the article it’s used in a way that actually includes the weights, and I think this is the usual way of using it! If you change the weights, you get a different model, though the two models will have the same structure.
Anyway, you make good points!