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.
These models can also be available to countless devices in the house or delivery vehicles of the future (drones asking to share an AI token so they can land properly??).
I was just thinking on the lines of what would the world look like when AI is in abundance, which there will be for sure, how would we use it etc.
I do find these kinds of exercises useful(_sometimes_) to imagine, what if we have a lot of something or we don't have it at all. I might be totally off here though.
Local offices will likely be slower than the large regional data centers, unless that is offset by having fewer users. But does it scale? If we take this a few steps further, why not have the LLM running locally in your home?
The idea is to let customers choose which content creators they want to reward, and not have to set up 100 potentially hackable accounts with small creators without payment processing.
User gets to choose, from a vast array of creators, not just inane "bundles" and from excess ISP profits.
Unlike the internet for decades, quality gets a chance versus quantity.
Mom and Pop thank you.
Your ISP doesn't own rackspace?