I have noted two AI companies going belly up with earnings in a year matching costs per month. So I assumed that was around the worse case scenario, and for not yet bankrupt AI companies earnings were probably a bit better, perhaps just losing ten times their earnings.
I now see the flaw of my reasoning. Capital isn't allocated on profits, it's allocated on hype. Having profits draws the company down because it's no longer pure hype, and thus doesn't contribute to the hype bubble the same way.
So existing, not yet bankrupt, AI companies probably has significantly worse cost to income ratio than twelve.