I would not expect Tesla selling their processor on the market to be a thing for a coupe of reasons.
Primarily, today, this kind of software is written specifically to a particular chip—you cannot simply swap out on processor for the next, so for a company to pick the Tesla processor, they are essentially handcuffing themselves to Tesla and Tesla’s direction and developing what Tesla may eventually consider to be a competitive product. Regardless of how advanced the Tesla chip is today (and bear in mind that folks constantly leapfrog each other), its a risky move to give up architectural control like that. Folks like NVDIA are a safer choice in that they don’t have skin in the final product but do have incentive to make sure you are successful.
The second issues is that selling your chips to external customers is a lot of work—its not just popping your chips in a box and strolling to FedEx. You need to clean up your SW libraries so they a suitable for external consumption, you need to create docs and training, you need to hire engineers who will work with your customers to help them adapt their SW to your HW and you need a support organization. You need to do this if you want to be a successful chip purveyor because one big thing potential customers will evaluate is how easy is it to develop for you chip because that represents opex and time to market for them. Doing this well will burn opex at Tesla that could probably be put to better use somewhere else.
