Try Documentalist, my app that offers fast, offline access to 190+ programmer API docs.
Waymo just announced a new trial for their self-driving program. It seems like they are getting closer to providing a commercial self-driving taxi service.
What about the economics of self-driving cars? How much money there is in providing such a service?
My back-of-envelope calculations inevitably lead to at least $100 million/year in profit from San Francisco alone assuming being at least 2x cheaper than taxi or Uber, drinking all of their milkshake.
How big is the market?
In San Francisco alone there are 1.8k taxis and 45k registered Uber and Lyft drivers.
Let's split the difference and say that replacing current usage of taxis and Uber requires 10k cars.
Method 1: taxi drivers salary
According to Wikipedia, in US there are 234k taxi drivers making $22.8k per year on average.
For simple math let's say the salary is $20k. Multiply by 10k cabs, it's $200 million per year. That's profit margin Waymo has to play with it. Let's say they decide to pocket only half of that, using the other half to make their service cheaper than taxis. That's still $100 million from a single city in the US.
Method 2: Waymo cost vs. taxi rates
Let's try a different way of calculating this.
The cheapest UberX fare from San Francisco Airport to San Francisco Downtown is $25. The distance is 13 miles. IRS federal mileage rate is $.54 per mile. This is meant to cover TCO (Total Cost of Ownership) i.e. gas + car depreciation cost. That's $7.02 dollars, let's round it down to $7. Let's say they decide to make $7 profit, so the cost to user is $14. That's $11 less than Uber. Let's say a single car makes the equivalent of 4 such trips (~2hrs of active driving on average per day), making $28 per day of profit. That's around $10k per year. Again we arrive at $100 million per year profit while still offering radically cheaper prices.
Method 3: Waymo cost based on ZipCar cost
ZipCar costs $10/hr and has a very similar cost structure to potential self-driving fleet of taxis. ZipCar pays for the car and gas and manages to make a profit. They have 148 locations, probably 500-1000 cars.
Uber ride sfo downtown -> sfo airport is ~40 minutes and costs $25. So Waymo could ask $37.5 for 1hr. Assuming they can make a profit on $10/hr, they could ask for $18/hr and be half as cheap as Uber. The result is similar to what method 2 gives us.
Let's multiply that by 10 big cities in only sunny U.S. states (California, Nevada, Arizona, Florida, Texas etc.). We're up to 100k cars and $1 billion profit.
I believe those are conservative numbers.
On one hand self-driving hardware will increase the cost of the car which might not be covered by $.55 IRS guidelines.
On the other hand, buying 100k cars means that you can expect at least 20% price reduction compared to what a taxi driver would pay. You save on dealership fees of 4%-15%. Buying in such bulk means the car company will be willing to forfeit a bit of profit margin for large, upfront order. They have 10%-20% to play with.
The same goes for gasoline: Waymo will save 3% profit margins of gas stations and will get cheaper prices by buying in bulk.
What if they go electric? Electric cars might be more expensive to buy but the price of solar energy might be much lower or even 0. Tesla's 85 kWh batter lasts for 300 miles which should last for a full day of driving (10 hrs based on 30 mph average speed of city driving). Waymo can optimize charging times for when energy is cheapest. In California electricity is cheaper during off-peak hours (11pm-6am), even dropping to zero.
They can build their own solar charging stations. Tesla build 842 stations in relatively short time.
Their insurance will be cheaper (dramatically so if we assume that self-driving cars will have close to 0 accidents where self-driving car is at fault and that insurance market is rational).
The cost of servicing the cars in bulk will be lower if Waymo hires its own mechanics.
Average cost of parking will be much smaller. It can be $100-$300 per month in SF. Waymo can build efficient multi-level parking at the outskirts of the city.
What about longer-term savings? When you order 100k cars you can co-design the car. If Waymo notices that a given part of the car breaks down more frequently, they will ask the manufacturer to re-design it for better reliability.
It most extreme case they can completely redesign the car from scratch. The pod that Google has already designed looks like it would be much cheaper to build than a regular car.
This is similar to how Google data-center evolved. At first they used other people's data centers and off-the-shelf computer. Then they started to design custom pc cases. Then they started to build their own datacenters. Then they started to build their own power sources for the datacenters. Then they started to design their own routers and switches and custom cpus optimized for running AI workloads. Each of those activities cut costs. At scale, the cost of hiring people to design those things is dwarfed by savings from being even a bit more efficient.
There are similar opportunities in managing a large fleet of cars. They can redesign cars for reliability. They can reduce fuel cost by picking the cheapest energy source. They can optimize the routes with software. They can encourage people to share rides.
Cheaper prices mean more usage
10 thousand cars is very conservative estimate. We assume merely replacing existing usage of taxis but as Uber has shown that low prices and convenience drive usage. Even the most conservative estimate would be that Uber is 5x of taxi usage.
There are 494k cars registered in just San Francisco. I don't think 50k self-driving cars is unreasonable which would lead to $0.5 billion profit from a single city.
Let's go crazy
A muni ticket is $2.5 per ride. Assuming it's profitable to privde 1hr of driving for $10, this pays for 15 minute ride, alone. That would cover a lot of use cases for muni, but would be much more convenient.
If you double-up, it would be 30 min ride. What if the fleet also has 12-seat Chariot-like mini-buses for most popular routes during rush hours? If they are only half-filled and the cost is $20/hr, you can ride for 45 min for muni-like price of $2.50. That pretty much covers all use cases of using muni for lower price (remember, if you only drive 15 min you would pay $0.9 for the ride).
SF's muni budget is $1 billion per year. This covers more than buses, so let's say it's $0.5 billion for buses and light-rail. As I understand this is what city has to pay on top of ticket revenues.
If Waymo can offer muni prices while making profit and providing better service, it would make sense to disband muni and save the city that $0.5 billion per year.
Initially it would be politically difficult (part of the reason muni is so expensive are very high bus drivers salaries negotatied by unions but at some point muni ridership will die out naturally if there's a cheaper and more convenient alterntive and politicians will have a hard time justifying funding a service nobody uses.
So far we've only talked about replacing taxi-like usage within the city limits.
There are a lot of people who commute 0.5-2 hrs each day to work. Let's say 1hr solo commute twice a day, $20 per day, $400 a month.
Consider that parking alone in SF can cost $100-$300, if commuting is major use of a car, it makes sense to sell the car and pay for it out of pocket.
This is the worst-case scenario. Due to tax law in US, if it's paid by the company as commuter benefit, it's effectively 40% cheaper i.e. $240/month. Many companies will be happy to offer that as a perk, making it free for the individual. $240 isn't much if the worker costs $15k/month.
The first thing Waymo engineers will do is to use past usage data to pack as many people into a single car as possible. If they manage to fill 12-person mini-bus with 10 people on average and the cost of the bus is $20/hr, it's 5x cheaper price i.e. $80 unsubsidized i.e. similar to monthly muni pass.
With prices that cheap the usage will go up alongside Waymo's ability to optimize the cost by packing more people in a single car.
What about meal and package delivery? What if it costs $2.5 per delivery within the city. The customer orders via phone. Restaurant summits a car when the order is ready, puts it in the car which drives to where the customer is. The customer gets an unlock code to open the door and get the order. On average this should be profitable even without tricks like combining multiple orders into a single trip.
Will they sing the body electric?
Why would fleet go electric?
For one, costs are going done. A major pat of the cost of the car is battery. An 85 kWh batter in Tasla used to cost $14.3k but the cost dropped by 35% recently, to $10.6k. Apparently the cost dropped 80% in last 6 years. If it drops additional 50% in next 5 years, the cost of the battery (and therefore of electric car) will drop by $5k.
Operating the car should also be cheaper. Electric cars are said to be more reliable and long-lasting because electric motor is less complicated than gasoline-powered engine.
Electricity should be cheaper than oil, especially in sunny states like California. Solar can already be the cheapest energy source and costs are still dropping. It would make sense for Waymo to build large solar installations nearby, produce their own energy and bypass middle man.
Price of electricity from the grid fluctuates. It's most expensive at peak times (9am-6pm) and cheapest during the night (11pm-5am). A 300 miles range batter is enough for 10 hrs of driving so majority of the fleet could be charged once per day, during the night, when electricity is the cheapest.
What if the fleet goes exclusively electric?
The somewhat optimistic case is 100k cars in San Francisco area alone, easily 1 million cars in US.
So far 570k electric cars were sold in US since 2008, 160k sold in 2016. A fleet like this would greatly speed up adoption of electric cars. It would have a postive knock-off effect on electric car sales to individuals.
For example if you have a fleet of 100k cars in a small area, it would make sense to build a dense network of chargers. If they opened those chargers to public (earning a small profit from charging alone) they would make owning electric car more convenient to people living in San Francisco.
More electric cars means the cost of electric cars would drop even faster, the demand for oil will drop (almost 50% of oil usage is for gas).
Car manufacturers could be hit hard.
Let's say that a car in a fleet gets 8 hr of usage per day on average and a car used by a human 2 hrs on average. Let's say that additional convenience would bump the usage to average 3 hrs per day. Each car in the fleet would thus replace 2.6 owned cars. 100k cars would thus replace 260k cars, reducing total number of cars by 160k. This is a big decline it car sales and a big hit to car manufacturer's stock price.