That thing about the self-driving taxi market being 10x the self-driving car market

I came across a chart yesterday on Twitter that suggested self-driving taxi revenues would grow pretty rapidly in the near future, whilst the underlying self-driving car revenues will essentially be flat, with the end point being that the revenues in the taxi market will the 10x the revenues in the underlying car market. I don’t buy that!

The short reason is that today, the cost of a car is only a small component of the cost of providing a taxi service; the major component today tends to be the remuneration of the driver, at least in reasonably deregulated markets. When you have a self-driving car, you no longer need the driver — all you need is a (maintained) car, and fuel. It is not reasonable to assume returns on the car seen as a capital asset being significantly different from those on any other capital assets, which would put it at most into a 5–10% area, probably at the lower end of the spectrum. If you run those number you’ll find you need pretty heroic assumptions especially on the lifetime of a car to make taxi revenues a 10x multiple of the underlying car revenues.

Let’s look at this in more detail: the self-driving taxi value chain has three major components

• producing the self-driving car
• owning / operating the self-driving car
• scheduling services

The first point is obvious. The third one is essentially today’s Uber or Lyft — they try to match the supply and demand of taxi-like services. The second point is about who is owning the physical asset and — importantly — who is putting up the cash for it. Note that the owner also decides with which scheduling services it’ll work, possibly on a real-time basis, just like Uber drivers today might work for Lyft at the same time.

Assuming perfect markets

The 10x claim is comparing the revenues across the entire chain to the revenues at the first level, so let’s dive into some numbers. Let’s look at perfect markets first, ie no participant has a moat and therefore noone can capture a rent.

Producing a self-driving car is intrinsically expensive, but we actually don’t care about this for reasons that will become clear at a moment. It costs what it costs, and it is available in unlimited quantities for this willing and able to come up with the cash.

Now we assume noone will buy a self-driving car for personal use. If anything this of course narrows the gap between car-as-a-service and car revenues, so without this my argument that the 10x is non-sense becomes even stronger. So people or companies buy those cars with the sole purpose of making positive returns out of them. This is not some kind of exotic market with a lot of risk — the demand for rides is highly predictable, especially in the brave new world where people prefer not to own cars. In this case, a return of 5–10% after cost would be plenty. Note that given that this is a stable cash flow it could be levered, and pretty juicy levered returns of 15–30% seem easily achievable.

Now the big cost item is fuel, and this is a bit unpredictable: currently the price of petrol is largely determined by applicable taxes. Self-driving cars would probably be electric, and the power will probably come mostly from renewable sources, so the protect-the-environment argument falls away. Even if the government would like to continue earning this revenue stream, it could not really tax electricity, so chances are that it would tax the cars itself, so I’ll just assume fuel tax is nil. A car today using 5l/100km would use 5000l of fuel over 100,000km. At a fuel price net of tax of €1.40*0.2=€0.28 this is a cost of €1,400, so we are at least one order of magnitude below the cost of the car. Let’s therefore assume cost (fuel and maintenance) will add another 5–10% annually.

Finally let’s get to the scheduling service: in a perfect market place this is actually mostly worthless. There is a fixed cost for developing the software, but in any case this cost is negligible compared to the cost of developing the car and the plant that can build it. There is also some cost of operating this service, but again, compared to the cost of operating and depreciating the cars this is mostly meaningless. I’ll be generous and put their revenue share at 2-5%, but I’d think it is actually too much (as a reference: Uber currently takes 20% but this is mostly not related to operating the system).

Finally let’s assume a car depreciates over 10 years and — for simplicity — let’s assume that customers pay in 10 equal instalments for their car to avoid the timing mismatch we have between the taxi revenues and the car revenues. For the sake of argument, let’s assume the price of a car (aka revenue of the car producer) is 100, leading (at zero interest) to 10 equal installments of 10. This yields the following revenue distribution

• car producer 10
• car operator 10–20 (out of which profit 5–10)
• scheduler 2-5

Using those numbers, the ride market would get revenues 22–35, out of which 10 would go to the producer, so the ride market revenue would be 2–4x of the producer revenue. Now we must keep in mind that this assumes that all self-driving cars become taxis, and I’d also argue that the lower end of the ranges makes more sense than the upper end, so overall I’d think that under this scenario a ratio of 2x is a sensible baseline assumption.

Imperfect markets

There are two market imperfections: firstly, there will be a transition period during which the taxi market will still be dominated by human drivers (and therefore ride prices are likely to stay high). Secondly, some of the players might be able to build moats and appropriate a larger than their perfect-competition share of revenues.

Let’s deal with the transition period first: as long as there are only a few self-driving cars around the marginal producer will still be a human operated car, and therefore ride prices are not expected to move significantly. During this period, the revenues of self-driving taxi operators could indeed be very high, given that they can charge prices as if they had a human operating the car but don’t have to pay a human who does. However, the bottleneck in this case is not so much a shortage of companies operating self-driving cars but rather a shortage of self-driving cars. This of course puts the rent-seeking potential right to the producers of self-driving cars who’d simply jack up their prices during this period, or they’ll only supply operating companies that have a profit sharing agreement first.

After the transition period we have three players, but as we discussed above, the moat of the car producers actually does not matter for what we are looking at. The car operating companies have very little competitive advantage: all they need really is money, and possibly access to a charger network (if cars are electric) and some garages (assuming anyone but the car producer can do much work on those cars). Effectively they are an asset company, so they’ll earn whatever the going rate for a stable income stream of those characteristics is.

The schedulers are slightly more complex: clearly Uber at the moment enjoys some decent advantage over its competitors, and thing like having the biggest customer base and having the biggest base of cars works. However, it is important to understand that in this business it is not that much being the biggest that matters, but it is more about being big enough as to always have a car for customers and vice versa within a reasonable period of time. If the ride market really overtakes the owner-driven market (which is not an unreasonable assumption) then there is place for multiple players who are big enough.

Also the scheduling companies neither really own their customers nor their cars: it is reasonable to assume that customers have all relevant apps installed, or even go through aggregators to find the most attractive ride. Similarly for the car owners: there is no reason to think they’d not just register with all scheduling services and go with the best offers. So even if there is a network advantage it probably won’t allow significantly higher margins as people might well be willing to wait a few minutes for a cheaper fare.

To conclude: even with imperfect markets I don’t see a very large proportion of the ride revenues go to the producers of the cars.