August 8 was the day Tesla said it would reveal its robotaxi, though this has been delayed until October, and the world has been puzzling over just what they plan to show. What is what they plan to eventually ship isn’t a robotaxi at all, but a new, cheaper Uber-style service based on Tesla’s Supervised FSD system and low-cost workers?
When Tesla delayed the reveal, it caused a fairly large hit to the stock price, suggesting investors put a lot of stock in it. So does Elon Musk, who has repeatedly stated that Tesla should no longer be viewed and an electric car company but as an AI, self-driving and robotics company, and that this is what will justify its high valuation.
The problem is that Tesla isn’t remotely close to putting a robotaxi into operation. Their FSD software still has trouble doing a dozen trips in a row without human assistance, while other robotaxi firms, like Waymo, are doing 100,000 trips/week without significant incident—and still do not feel themselves ready for production. It’s a long way from dozens of trips to tens of thousands. In addition, operating a robotaxi service requires much more than having a self-driving system and vehicle, there’s a long list of infrastructure and logistics that need to be set up in each service area, and Tesla has shown very little progress on these.
Tesla is not without assets, however, in its “former” role as an electric car company. Might those assets be exploited to make a very different, and potentially low cost Uber-style service using human operators and fleet-owned EVs with very low cost of operation?
Electric cars have a reputation for being more expensive to buy, but they are also much cheaper to operate and maintain than typical gasoline cars. This makes them cheaper than gasoline cars in total cost of ownership. If you drive a lot, as Uber drivers do, this factor is even stronger, which has prompted the question of why more Ubers aren’t electric. Factors include the higher purchase cost, and the fact that many Uber drivers don’t own their home and thus can’t charge it there, forcing them to use expensive and time-consuming fast chargers.
A Taxi-Network (TNC) run by a company like Tesla would solve these problems. For them, the cars are cheaper to buy (not just because they make them, but because they get them very cheaply when they come off-lease or in trade-in) and raising the money is easy. Because Tesla owns (by far) the largest and best charging network, they also can charge at lower cost. By directing drivers to chargers with spare capacity, they can charge for the wholesale cost of electricity. They also have good economies of scale and OEM costs of maintenance and service, not that EVs need much of that. They also operate an insurance company, and claim their cars have fewer crashes. (More on that, below.)
Uber Flipped Upside-Down
Uber’s initial attraction was they were a software company. Cars were bought and maintained by drivers, who also provided the labor as gig workers. It took a long time and heavy losses, but Uber just announced decent profits. The losses came in part because of costs recruiting drivers. Uber exploits the fact that their drivers own and maintain their cars privately, and mistakenly consider their only operating costs to be fuel and a few other items, giving them a false impression of their profit. (Some Uber drivers lease/rent cars just for Uber driving, and indeed many of Hertz’s Teslas were rented in this fashion, though not with great success.)
Tesla could make an interesting proposition: Drive as a gig-worker in our service, and Tesla provides the car. On top of that, you can use the car as your personal car during your off-time. Now, people who don’t own a car can be a TNC driver, and even get a (free or discounted) quality car for themselves as a bonus. This has echoes of the old taxi industry, where taxi fleets owned cars and hired drivers or leased the cabs out to drivers, but that industry was warped by the medallion-style systems in most towns.
There are three levels at which this could take place:
- The person does all the driving, like any TNC. The ownership model is different, and the car is much cheaper to operate.
- The person supervises Tesla Supervised FSD. To do so, they must pass certain tests for quality of driving and safety. Their high-quality driving quickly generates tons of high-quality training data to improve the system.
- Tesla FSD graduates (some years from now) to not needing supervision on some decent subset of roads, including most freeways. The person takes over on difficult streets, construction zones, pick-up/drop-off, bad weather, locations outside the service area etc. If the vehicle takes all duties for a long enough duration, the driver clocks out, and gets reduced pay. They are free to read, watch videos, do other gig work online, and in some situations, sleep. They handle charging, cleaning etc.
- Eventually, if the FSD system becomes generally capable, and Tesla has built the necessary logistics in a service area, the car runs empty and Tesla has a robotaxi service. It does not have any strong advantages over the competitors, but it was able to expand first into more cities, gather training data at no cost and develop all the tools and learnings needed.
It’s worth noting that even the basic first level would be an attractive service. Integrating the ride-hail driver app into the car and its screen can improve the experience, and the ride will be low emissions, quiet and reasonably luxurious.
All these stages might be able to find drivers willing to work for less, both because they need not own a car, and in stage 3 because the job is much easier and includes long breaks where their time is mostly (though not entirely) their own. Tesla could even try to set up a laptop gigwork market for the drivers to make money in these down-times, though only a few tasks fit that description if the driver has to be called to attention suddenly. If the car is good enough to only need the driver at planned times, more is possible.
There are a number of markets where minimum wage is only $7/hour. The combination of a car that costs 20 cents/mile to operate with a human attendant who is $10/hour all-in could generate a very cost-effective TNC service. A 10 mile Uber ride is usually >$25. This ride might have a COGS of $3 for the vehicle and $3-$6 for the driver. If the driver can do other work on the drive, it could drop even lower. Even doubling this cost could be a great value service that would quickly take over Uber’s business.
The drivers might be gig-workers or they might even be employees, something Uber and Lyft have fought. Employees don’t control their hours but would need to be paid for idle time. It’s unclear if utilization could be kept high enough to occupy full-time workers, but the gig option is always present.
As noted, there’s a ton of logistics Tesla has yet to set up, but with drivers, it would be pretty similar to what Uber needed to start, as the drivers would solve many of the problems. The service would also be a fairly nice car, quiet and environmentally friendly. For some riders it would quickly be the obvious choice, and it’s a good stepping stone to an actual robotaxi service should Tesla manage to pull one off—with its current hardware or more likely with a new hardware generation.
