No need for AI chips
Job one is getting cars and perhaps later, trucks off of the end of an assembly line in sufficient quantities and at a profit. No body I know is all that keen on an autonomous car. Sure, they'd like a direct ride home from the local or a party, but beyond a simple taxi, they aren't all that interested. A shared car service is even further down the list. When it's time to pick up the kids from football practice, who wants to have to wait 20 minutes for the car to show up? If practice ended early due to rain, that 20 minutes is even more of a nuisance.
Knock, knock. Who's there? The competition. The Chevy Bolt is available right from dealer lots in the US. Somebody in line for a M3 seeing neighbors with the Bolt might just decide to cancel their reservation and go for the Bolt instead. Not every one, of course, but the Bolt is only one model that directly competes with the Model 3 today. Tomorrow is another day. By the summer of 2018, there could be many more 300 mile range (no HVAC on) cars ready to purchase.
There was a great 3-part series presented by James May, Kate Humble and ? on the Mini plant in the UK where they try to follow a car all the way through the plant. Anybody with a engineering background will be able to get a whole bunch of information on modern auto production out of what is shown and talked about. The factory produces a car every 68 seconds and given similar circumstances at the Tesla plant, which is giving Tesla a big load of optimism, to meet the goal of 500,0000 cars per year would mean a car leaving the line, complete and ready to sell, every 37 seconds. Having worked in production planning, I know how hard that is. Is BMW making their own silicon? Likely not. It's too far outside their core competency and they'd have to be able to keep a team busy all of the time. No, the best approach is to outsource those sorts of items to the company that is in the best position to deliver.
Another series to watch is on the Elizabeth Line (Cross rail). A great look into what it takes to put in a passenger underground line. After seeing those shows, I take Elon's statements about being able to bore 4x faster with a 2x smaller machine with a grain of salt. The larger machines can work very quickly when there aren't issues. It's when they come to dodgy ground that they have to grout and take special care so it doesn't collapse that takes the most time. That's not going to change. What is important is the passenger throughput of the system. 1-2 people in a car carried on a "skateboard" that is lowered into the system via dozens or hundreds of elevators isn't going to be able to move enough people during rush out to make any sort of difference. Los Angeles would already have loads more underground lines except for the cost and the "Not Under My Feet" of everybody that would be above the tunnels. A city doesn't have to make money from the till if the city as a whole brings in more business as a result of the transportation system. A privately build system would have to be able to charge enough to earn a profit from fares. What could happen if Elon puts a system in and the company goes BK, they city would have to buy them out (at fat profit) and either run it themselves or fill it in to avoid liabilities.