You are correct that Tesla takes on the risk, and that risk is immense, but that doesn't mean it can't be profitable.

Keeping capital expenditures to build the CyberCab line separate for a moment.

The cars are going to cost Tesla somewhere in the 20k-30k range to build with 0 profit. We know it'll be cheaper to build at scale than a Model 3 which they can sell for a profit at 37k USD so lets just say 30k.

The average Lyft/Uber rides are somewhere between $15-$20, so given no driver, lets say the average autonomous ride at scale, is $10. Waymo's is currently almost $20, but scale is small so that should come down.

That's 3000 rides to make up the base cost of the car, which means it's 8.22 rides a day for a year to cover the capital costs of the car, and Uber drivers typically do 15-30 rides a day, so the car could pay for itself in 100-200 days at $10 average rides.

Beyond that, you need to cover all the additional costs like charging (they own the infrastructure so its industrial rate cost), service, operations management (e.g if a car gets stuck who goes to retrieve it in person like Waymo does), customer support etc etc.

So if we say 30 rides a day because it can run 24/7, that's $300 a day per car in revenue to cover all the extra costs after the car is paid off in 100 days and start paying off the production lines.

That seems doable to me, and this is probably pretty conservative (edit on ride count per day per car.)