I would think that the biggest obstacles for small startups in the AI space is the expense and availability of suitable hardware. Then the expense of the elec

I would think that the biggest obstacles for small startups in the AI space is the expense and availability of suitable hardware. Then the expense of the electricity itself. Big Cloud companies, who are also AI companies, with huge wallets have a huge advantage to start with. The regular answer is of course go to the cloud, which is by far the easiest approach. I would think the cost of intense "AI compute" over a long timespan in the cloud is really expensive as well. And you are paying your competitors for your work. If any of that is correct, which it might not be, then building the initial models would cost more than a startup couldl afford. You can also go to the cloud and used "AI APIs" usually proprietary in one way or another to the cloud vendors themselves and limiting the scope of what can be accomplished. (Plus paying the competing in order to compete)