Originally, this post was supposed to be a place for me to talk about my long position in Tesla and how that might evolve over time in a changing market environment; talking about the competition was only supposed to be a very brief section. For better or worse, the competition was all I could write about, and it’s not like I knew enough about world politics to say anything worthwhile on macro trends and the trade war. So, I want to talk about Tesla’s competition in the autonomous driving space – who it is, the current state of affairs, and why I’m not worried just yet.
The Current State of Affairs
As a bit of background, it helps to understand who’s where in the self-driving car race today. This graph is probably the clearest depiction.
The further right a circle is, the better the technology is. Grey dots are from 2015, blue dots are from 2016, and black dots are from 2017. A few important takeaways are:
A better plot to visualize progress would have plotted the log of miles between disengagements instead.
The two furthest-along companies are Waymo and Cruise.
In 2017, Cruise managed to reach where Google was 2 years prior – but they’ve caught up remarkably fast.
Virtually everyone else is behind. I enjoy pointing out that Mercedes-Benz truly is the best or, in this case, nothing [1].
Just this month announced and released Waymo One. I cannot state how important this is: Waymo is the first company to release a product. Granted, Waymo One is only available in Phoenix, and it isn’t available to just everyone currently, but the only firm today generating revenue instead of concepts or promises is Waymo. It’s likely that Waymo will “crack” autonomy first, by which I mean develop an algorithm capable of driving autonomously anywhere in the nation. The historical data points to it, and Google is easily one of the best companies on the planet when it comes to applying ML and AI to problems (e.g. the Pixel’s Night Sight mode). I’m not saying that building a great phone camera is equivalent to building an autonomous vehicle, but Google has showed time and time again that they know how to do AI.
That said, I don’t think Tesla is out of the race just yet for two reasons: data and hardware.
Data
At the core of the autonomous driving problem is the desire for data – just as human drivers gain experience over time, the algorithms that enable autonomous driving require insane amounts of data for validation. Waymo claims on their website to have driven “over 10 million miles on public roads and 7 billion miles in simulation”. Cruise’s website notes that they “collect and analyze petabytes of data”. But Tesla has hundreds of thousands of Autopilot-capable cars on the road today. If Tesla wants to test a new iteration of their software, they can push it overnight and test the algorithm in the background, something they’ve called “Shadow Mode”. Shadow Mode or not, Tesla has driven more real-world miles than Waymo has – by several orders of magnitude.
While Waymo has a better algorithm today, Tesla has the data necessary to build one that’s just as good in the future. Waymo just doesn’t have as many cars on the road generating data for them. They’ve planned to purchase 20,000 I-PACEs from Jaguar, but these are future plans, and future cars don’t generate today’s data. The Verge reports that Waymo currently has about 600 vehicles in their fleet today – not even one day of what’s output from Fremont. By the time Waymo has finished adding 20,000 vehicles to their fleet, Tesla will have produced hundreds of thousands more. So Tesla definitely has an edge in its ability to gather data from the real-world for testing its algorithms.
Cars
Tesla’s second advantage stems from their cars on the road. Waymo isn’t interested in building their own vehicles; they’re more than happy to focus on building better drivers. And good for them! They can focus on their strengths and not worry about the hell that is automotive manufacturing. But suppose that one day, John Krafcik decides that Waymo is finally ready to open up to anyone who wants to use it. How long will it take for Waymo to acquire the cars it needs to operate a national network? In my mind, this is the key question, because for Tesla, the answer is one night. Once Elon decides that Autopilot is safe enough for full autonomy to be enabled, one software update is all that’s stopping every Model S, X, and 3 (with the right sensors, of course) from becoming Level 5-capable autonomous vehicles. The answer to the key question for Waymo is the exact answer to another question: how much leeway does Tesla have in the race to be the first to market? If Waymo takes 2 years to expand its fleet so that it operates in every major city across the country, then Tesla has 2 years from whenever Waymo expands their Waymo One product to crack autonomy, if they want to be first.
Time to Market = Time to develop AV algorithm + Time to build the fleet
Tesla’s advantage (as I see it today) is that (1) they have hundreds of thousands of cars on the road, all of which are ready for autonomy once a sufficiently performant algorithm is developed, and (2) these cars are gathering and providing a larger dataset than any other company has access to.
What about Cruise?
You’ll notice that I haven’t yet talked about GM/Cruise. I think about them in the same way as any other company: (1) Do they have the data? (2) Once they’re ready to go, how fast can they ramp up?
Regarding the first point, I don’t believe that they have the infrastructure in place to gather data quite like Tesla. Cruise’s Wikipedia article mentions that as of mid-2017, the total car count was estimated to be around 180. If they’ve expanded since then, it’s likely that they’re more like Waymo than Tesla in terms of the rate at which they can gather data. As the Bloomberg chart above shows, 2017 Cruise is about as good as 2015 Waymo, but the rate of improvement is impressive.
However, once they’re ready, Cruise is in a position to overwhelm the market with vehicles. While Cruise won’t be able to flick a switch on existing cars (since production vehicles don’t have the sensors necessary for autonomy), GM sells about 10 million cars each year. While Waymo has to work with a supplier and Tesla’s production capacity is still ramping up, GM faces none of these problems. This is Cruise’s critical advantage.
And Zoox?
Pretty much all we can know about Zoox is the article that Bloomberg Businessweek put out in July of 2018. I’m not too worried about them for now – they’re behind Cruise and Waymo, and they also don’t have the in-house ability to build future vehicles. Zoox’s ambition was to build their own autonomous vehicles from the ground up, though it’s unclear that that’s still the case with the recent removal of the CEO (which was a bit messy).
Does being first even matter?
With social media platforms like Facebook, the more users that are on a platform, the more valuable that platform is – you have a MySpace account because no one uses it anymore. But I’m not sure that Facebook is a good model for network effects in autonomous driving – I think dockless scooters are a much better model. People only install your scooter app if there are scooters in your area. In Austin at least, this means that most of the people I see riding scooters are those near the UT campus area and young professionals who live or work in downtown. Put another way, I don’t think there’s an intrinsic advantage that people gain from using an autonomous mobility service that others also use, but people will need to be able to actually use such a service reliably for adoption to occur.
It’s possible that network effects are driven by psychology and data. What I mean by this is that people feel more comfortable using familiar technologies, i.e. Sunny may trust Waymo over a competitor more if Sunny’s friends have had positive experiences with Waymo in the past; Sunny may not trust a lesser-known company when the default option works perfectly fine. Furthermore, companies with more real-world data (e.g. Waymo, Tesla, Cruise) will likely be able to develop more robust systems in the first place, allowing them to create a better service.
What about the fact that Tesla doesn’t use LIDAR?
That’s a big assumption of mine – that Elon is actually right about the whole “we don’t need LIDAR for full autonomy” thing. I believe him for the most part – I’ve used Autopilot extensively on the highway (confirmation bias) and I buy into his logic that “humans only need vision to drive, therefore AVs should also only need vision to drive”.
Conclusion
Tesla is still very much a valid player in the autonomy race. Their primary benefit is that they have cars on the road today, but their competitors are also quite strong. Waymo benefits from being the furthest along in developing a good algorithm, while Cruise benefits from massive scalability once the software work is done. Tesla probably won’t be first the first to develop autonomy, but it’s possible that they can be the first to market.
[1] A great comment by Dieter Zetsche, the Chairman of Daimler, is "Anyone who focuses solely on the technology has not yet grasped how autonomous driving will change our society.” This is a quote in a press release from 2015, so it wasn’t just off-the-cuff. It appears that maybe they ought to focus a bit more on the technology.
Disclaimer: I’m currently long $TSLA and $GM. I’m not long $GOOG because I can’t get isolated exposure to Waymo.