>The _vast majority_ of AI spend is wasted

I don't think this part is correct. It is spent on things. There will be a quantity that will go training models that do not contribute research data or active use that you could say that the resources were wasted. If the money is spent on infrastructure or the resulting models found active use (or provided research insights) then I don't think you can say that the money has been wasted.

I think a large proportion of the advances in AI that have come in the last year have come from being able to see the models from the previous year in action.

That does not guarantee that money will be returned to investors, Just that it wasn't wasted.