First, the ISP would need fat stacks of not-mismanaged cash and decide to invest those in huge amounts of hardware that will atrophy in value from before the mo

First, the ISP would need fat stacks of not-mismanaged cash and decide to invest those in huge amounts of hardware that will atrophy in value from before the moment the check clears. Next, the ISP would need the technical chops to set up that hosting themselves (or be able to afford outsourcing the setup and maintenance). All the while, the ISP would need to withstand moving outside their area of expertise and primary business focus and core competence, that being delivering network connectivity (at the lowest possible cost and, most often, at the highest possible price). And finally, if the ISP can do all that, they would need to decide it would be a solid business decision; that the investment would pay off continually, that the customers not only exist but with sufficient quantity and spending capacity to make it worthwhile. And before all that, they need to think of this idea, convince management, executive, board, shareholder, and/or lobbyist levels that it's a good idea and not too risky. *And then* they need to successfully execute the project to bring it to market (including successfully getting their hands on that hardware, which might be tricky in and of itself). All in all, an outfit provisioning AI as a service needs fat stacks of (potentially funny) money, the knowledge and guts to pull it off, and, to have thought of the idea in the first place. And to have convinced the money people that it should be done. Now, make a business leasing these appliances to the ISPs and managing them as a service. That could be a business.