I'm not talking about net loss, I'm talking about operating costs.
The maps they make for self-driving cars are made constantly - it's not something they make every year. They're updated weekly, if not daily - and they use a separate much more expensive car to do it. There's a whole separate team just for dealing with this. The amount of data the taxis collect is insane. Terabytes of data per car per hour, which needs to be stored and processed and transported (the data is so much it often makes more sense to physically ship hard drives than to send it over the internet). The "phone home" team is paid employees, and in order to have cellular service to the cars the companies pay a much more expensive plan with cellular providers than you're used to with your cellphone. The engineers who are maintaining the software and making fixes cost money. The people that clean and maintain the cars cost money. Not to mention the wear and tear on the vehicle (which includes the extremely expensive sensor suite) and the electricity to move the vehicle.
I am saying the unit economics, which has nothing to do with the previously spent billions of dollars or efforts of expansion, is currently dogshit for self-driving cars. Theoretically it will improve with scale, but Waymo isn't at scale yet. I think people that haven't worked in robotics have this notion that robots are wildly profitable but this is a common fallacy. Robots almost always lose money, and usually a shitload.