I’ve learned a lot in the last few months about the economics of robotaxis, and I’d like to share a bit of it here as a way of accumulating it in one place. A robotaxi is a particular application of a self-driving car. Tesla wants to sell self-driving cars directly to consumers, but there are a few businesses (Waymo, Zoox, until recently Cruise) that are focusing exclusively on robotaxis, meaning that they are in the business of selling rides rather than cars. Although they focus on the same basic technology, these businesses turn out to be extremely different, and that’s what this post focuses on. I’m going to break it into pro-Waymo and pro-Tesla sections to be a little more provocative.
Pro-Waymo
You may have seen a recent video by Mark Rober where he tests a lidar-based Automatic Emergency Braking system against Tesla’s autopilot. Tesla, somewhat infamously, does not use lidar or even radar systems in their vehicles. There are obvious reasons to believe that you don’t need these systems to make a good self-driving car (people don’t need them), but Tesla also doesn’t seem to be that close to full self-driving (Rober’s video definitely doesn’t portray them favorably). The lidar and radar modules were removed (or, in the case of lidar, never installed), because they are expensive. My finger-to-the-wind estimate is that all the cameras in a Tesla combined probably cost the company around \$500; a single lidar puck can cost over $1k (sometimes well over). Expenses of this level would have prevented Tesla from targeting the sub-$50k demographic where they currently do an overwhelming majority of their volume (the sub-$50k models 3 and Y account for more than 95% of vehicles sold).
Waymo (I’ll just use “Waymo” to refer to robotaxi providers generally) can afford to spend far more than Tesla on hardware, because they expect to get much more revenue per vehicle - napkin math says about $100k/year in revenue, which is about the cost of the most expensive Tesla. Tesla has to worry a lot about the cost effectiveness of their hardware, because they sell a vehicle. Waymo buys vehicles and can take advantage of future hardware innovations anywhere in the market. I am confident that they spent many years happily losing money on each marginal vehicle while establishing a brand and collecting data.
And speaking of data: every robotaxi provider besides Tesla creates high-fidelity maps of the cities they operate in. In some ways, this has also turned out to be an advantage. Tesla has struggled for years to ship fully general self-driving, but robotaxi providers don’t need fully general self-driving to capture almost all profitable rideshare markets, because nearly all profitable rideshare markets are in large cities. Tesla’s customers would be understandably upset if the self-driving car software they paid a subscription for kept them geofenced in a small area; Waymo customers can always order an Uber when they have to.
Triangulating a few sources (here, here, here), it looks like the cost of a Waymo vehicle was around $200k last year. Another, more recent and likely more reliable source says the equipment cost is “as much as \$100k per vehicle” - adding on the vehicle itself and integration costs, I think it’s reasonable to take the $200k estimate at face value. Waymo still has a lot of scale efficiencies they can realize as they scale beyond their current network of around 2k cars.
Tesla wants their customers to operate their own small robotaxi businesses by buying Teslas and then making them available for other people to use via a Tesla robotaxi app of some sort. This is a little bit silly. A Tesla costs around \$40k. The $160k difference in up-front cost is not in any way a crippling disadvantage for Waymo, if they can run their robotaxi business better than the typical Tesla owner.
For continuing costs of ownership, I defer to the thorough folks on the Waymo subreddit, who estimate that a robotaxi costs around \$39.5k/vehicle/yr to maintain. This estimate looks very reasonable to me. About half of this, \$20k/vehicle/year, comes from operational support and maintenance - these would both benefit substantially from scale that a Tesla microfleet owner (AV shepherd?) just couldn’t achieve. Another $7k comes from insurance. Waymo has been exhaustive in reporting all of their L4 accident rate data. Tesla doesn’t have L4 accident data. Which one is more insurable? Other nuisances, like potentially having to clean another person’s vomit out of your car, make it seem quite unlikely that a typical person would just farm out their Tesla for a couple of hours while they were at work. Even if Tesla is capable of producing a self-driving car, it’s not certain that anyone will actually be competitive with Waymo at the part of this where you actually have to operate a robotaxi business.
And finally, the Waymo vision of the future seems to involve a much more efficient allocation of resources overall. In order for your car to sit there doing nothing while you’re not driving, we dedicate tremendous amounts of (sometimes extremely valuable) real estate to parking lots, driveways, and garages. Intuitively, the pure robotaxi proposition seems to create a lot of value, and if Waymo is able to capture some of that, they may be very well-positioned.
Pro-Tesla
The pro-Tesla argument is a little less big brained. Tesla has millions of cars on the road. They are all collecting more-or-less expert demonstrations of how to drive. No one has more data on this than Tesla. It’s a little weird that they haven’t already figured out FSD, which suggests there’s something about long-tail problems I’m glossing over here, but still: sooner or later, Tesla will probably figure out FSD.
Waymo makes a \$200k car. Tesla makes a $40k car. If they are both about as good at FSD, someone will notice that it’s not that difficult to make a business that does the same thing as Waymo but cheaper. Operating a fleet of vehicles is something that traditional cab companies have done for a long time; the FSD is the hard part.
Most adults go to work at around the same time of day. They can maybe stagger it a little and carpool, but this makes it hard to get substantially below 0.5 cars per person at best. Waymo has around 2,000 cars. America has 270 million adults.
OK I’ll sneak in one big brain thing: once you factor in the fact that robotaxis need to spend a decent amount of time driving around empty (30-50% of their miles will probably be unoccupied, if they’re like Ubers and non-robo taxis - this number makes a lot of sense if you start thinking about scenarios like rush hour, where there’s a substantial net flow in one direction), they probably won’t be much cheaper to operate per mile driven than a regular car. Maybe robotaxis would let us build walkable paradises that eliminate some of our need to drive, but even then you might imagine it’s pretty close to cost parity, and “robotaxis will eventually drive down their own profits” is not exactly an amazing argument in favor of robotaxis.
Even with a robotaxi option, many people might decide that $40k is not too much money to pay to own their own car. If you drive a lot, or like to go camping, or really value having a mobile place to keep your gym clothes, having your own car is very convenient. Even in cities with excellent public transit that’s probably much cheaper than a robotaxi could ever be, many people still own cars. Tesla is in a good business; they sell a product people will always want, and they can collect data at a scale that Waymo will probably never match. In June, Tesla will start operating robotaxis in Austin, TX. If they demonstrate there that they do have FSD that can match Waymo’s, it’s only a matter of time before Waymo disappears.
Quick closing note: Zoox might be slightly better positioned here because they got acquired by Amazon. Amazon could use self-driving cars for things besides robotaxis, so they’ll likely be willing to operate Zoox at a loss for some time even if it looks like Tesla has them beat in the robotaxi market.