Tesla Is A Software Company

| About: Tesla Motors (TSLA)

Summary

Tesla is a software company. This isn’t just semantics. It has important implications.

Tesla is working on two major software efforts: self driving and factory automation.

With regard to self driving, the rate of improvement matters more than the current level of performance.

Factory automation is an opportunity for Tesla to use its software acumen to leapfrog competitors.

Introduction: software eats cars

In the seminal essay “Why Software Is Eating The World,” venture capitalist Marc Andreessen describes how software companies have achieved rapid growth over industry after industry, often capturing dominant market share: Amazon (AMZN) for books, Netflix (NFLX) for movies, Apple (AAPL) and Spotify (MUSIC) for music, Skype (MSFT) for telecommunications, and LinkedIn (MSFT) for recruiting. In the years since the essay was published, we can add Uber (UBER) for taxis and Airbnb (Private:AIRB) for hotels. Amazon has expanded from selling books into retail more generally. Software continues to eat industries.

During an interview in May, Andreessen discussed the thesis of his essay with respect to the automotive industry:

...in 10 years, the winning car company is going to be the car company that makes the best software. … Our thesis is that…the value will flow to the software layer. The entire experience of being in a car will be defined by the software. At the limit, whichever company makes cars or makes the intellectual property that goes into cars — whichever of those creates the best software will then be the winning company and will get the majority of the value and including ultimately the majority of the profit.

Andreessen then discussed Tesla (TSLA) as a potential candidate for the winning car company, saying that CEO Elon Musk is “incredibly focused on software”:

He’s going to try have Tesla be the best self-driving car, which is entirely a software exercise. He’s on it. He’s pushing very hard. The other car companies are all in more or less reactive mode to what he’s doing. He definitely has the lead right now. But that said, the other car companies are trying to figure this out and we now have an entire wave of new automotive software companies in Silicon Valley — dozens on its way to hundreds of highly capable founders and engineers building new kinds of autonomous software for cars.

Besides the assertion that the majority of profit will flow to any one company, I agree with Andreessen's picture of the automotive industry’s future. Tesla currently has the lead in self driving, but success is not assured because large incumbents like General Motors (GM) are gobbling up autonomous software startups.

What makes Tesla unique in the world is that it is both a software company and a car manufacturer. It’s headquartered in Silicon Valley, giving it access to the best software talent in the world. A recent survey found that Tesla is the second most attractive company in Silicon Valley for software engineers, product managers, and data scientists. The first is Google (GOOG, GOOGL), whose spinoff Waymo is competing with Tesla in developing self driving. Globally, Tesla is the fourth most attractive company. No other car company made the top 10, either globally or in any of the individual cities surveyed.

The world's most attractive company? SpaceX (SPACE), a company that shares a CEO with Tesla in Elon Musk. The two companies collaborate on "software, engineering, materials and expertise." Tesla indirectly benefits from SpaceX holding first place.

A 2015 report by Morgan Stanley found that 60% of Tesla employees worked in software, whereas 2% is typical for other automakers. This is perhaps why 83% of consumers report being "very satisfied" with Tesla's in-car software versus just 44% for other car brands.

Tesla’s CEO comes from the software industry. Before SpaceX or Tesla, Elon Musk co-founded Zip2 and X.com, which merged with PayPal (PYPL). Tesla has the location, leadership and labour force of a software company. It also produces software that consumers like.

Tesla is using its software acumen on two major efforts: self driving and factory automation. First, Tesla wants to take humans out of the driver’s seat. Then, it wants to take them off the assembly line.

Tesla’s Hardware 2 cars will drive 11 billion miles by 2020

This could be the most important Tesla graph:


It depicts the projected accumulation of miles driven by Tesla’s Hardware 2 cars, cars equipped with what Tesla claims is full self-driving hardware.

To produce this graph, I modeled that Tesla would produce 5,000 Model 3s per week starting at the end of Q1 2018 (as management has guided) and 10,000 per week starting at the end of Q1 2019. I have no strong view on how the Model 3 production ramp will actually play out. However, I view these as reasonable assumptions for the purpose of a model.

On these assumptions, Tesla’s Hardware 2 cars will reach 11 billion cumulative miles driven in Q4 2019. At that point, HW2 cars will be driving 1 billion miles per month. (I computed these results on a spreadsheet by multiplying the number of Hardware 2 cars by the average miles driven per week, and then adding up the miles driven in each passing week.)

11 billion miles is a significant threshold. A paper published by the RAND Corporation estimates that a fleet of autonomous cars may need to drive 11 billion miles in order to provide robust statistical evidence that the autonomous system has a lower rate of fatal crashes than human drivers.

Tesla plans to obtain this statistical evidence by running its full self-driving software in “shadow mode.” The software will not control the car, but will receive input from its sensors and simulate steering, braking, and accelerating. The software’s decisions will then be tested against the driver’s actual input to determine which is safer.

We don’t know how many miles of driving data are required to develop self driving, but we can estimate how many miles are required to validate its safety. Right now, Tesla is the only company on Earth with a viable path to collecting 11 billion miles of driving data by 2020. Without a change of plan from competitors, or a failure to execute on Tesla's part, in 2020 Tesla will be the only company in the world with the data to statistically validate that its self-driving software is safer than the average human driver.

Tesla self-driving demo using the same multi-camera system and GPS included in new production cars.

If more or newer data is needed, Tesla will collect another 20 billions miles of data in 2020. That’s before assuming any increase in production from the 10,000 cars per week achieved in Q1 2019. Tesla’s production target for 2020 is 1 million cars, or about 20,000 per week.

Tesla’s end-game is to launch the Tesla Network, an autonomous ride-hailing service. I very roughly calculate that Tesla could earn $12.7 billion per year with a fleet of 1.6 million autonomous vehicles on the Tesla Network — a mix of customer-owned and company-owned cars. That figure includes vehicle sales, but not Tesla’s energy business.

At even just 15 times earnings (the historical average for the S&P 500), $12.7 billion in profit would yield a market cap of $190 billion. That’s 3x growth from Tesla’s all-time high.

The rate of improvement matters more than the current level of performance

Negative perception exists around Tesla’s full self-driving software development effort largely because tens of thousands of customers are exposed to the jankiness of Tesla’s advanced driver assistance software. If Waymo offered advanced driver assistance software, we would probably hear similar complaints. Serious jankiness apparently exists in Waymo’s full self-driving software, according to Ars Technica:

Waymo's cars have had persistent problems navigating left turns, according to (reporter Amir) Efrati, especially when there's no left-hand turn arrow to control oncoming traffic. Waymo cars struggle to navigate cul-de-sacs. Large mall parking lots also pose a challenge, since these private properties might not be well represented in Waymo's 3-D maps.

Disengagements data from the California DMV has been cited as evidence that Tesla is behind in developing full self-driving software. There are problems with this data, most notably the conspicuously small sample size. However, let’s assume for the sake of argument that Tesla is behind.

How far behind Tesla might be is only half the story. The other half is how fast Tesla is moving ahead. By analogy, if you give me a three-mile head start in a six-mile race against Usain Bolt, I’m still going to lose. It doesn’t just matter how far ahead I am. It matters how fast he’s moving.

Waymo only has 100 self-driving minivans on the road. Tesla has delivered 100,000 Hardware 2 cars — 1,000x as many. On a daily basis, Tesla’s fleet is encountering new driving environments and driving events that Waymo’s fleet of 100 minivans — and perhaps a few hundred other cars — simply cannot.

Waymo self-driving minivan.

With just the Model S and Model X, Tesla’s fleet is growing by roughly 2,000 cars per week. That’s 20x more cars added per week than Waymo’s entire fleet of minivans. As Model 3 production gets underway, that growth rate will accelerate. Waymo’s fleet, on the other hand, has not been growing appreciably. It has not announced any plans to grow the fleet to beyond a few hundred vehicles.

Tesla’s Hardware 2 cars are currently driving around 3.7 million miles every day. By comparison, Waymo’s cars have only driven 3.7 million miles in Waymo’s entire existence dating back to 2009. Alphabet’s cars drive 3.7 million miles and we call that Waymo. Tesla’s cars drive 3.7 million miles and we call that Tuesday. Waymo is currently accumulating miles at a rate of 1.7 million per year.

Waymo’s accumulated driving data is increasing slowly, whereas Tesla’s is growing rapidly. Waymo therefore has a slower path to fixing its system’s jankiness, to encountering new problems and driving environments, and to training its deep neural networks’ to accurately perceive the car's surroundings. The rate of improvement matters more than the current level of performance because it determines the future level of performance.

Let me say that again: the rate of improvement matters more than the current level of performance. This is fundamental. Given Tesla's unparalleled access to driving data, it stands to reason that the rate of improvement of Tesla's self-driving software also is unparalleled.

The analyst firm ARK Invest agrees with this assessment. In its research report on self-driving cars, it writes:

...ARK believes that the MaaS (mobility-as-a-service) market will evolve into regional monopolies because of the deluge of driving data necessary to train autonomous systems. Tesla seems to have a first mover advantage and appears to be the leader in autonomous data collection around the world, as it is alone in selling vehicles equipped with the necessary hardware sensors to collect autonomous driving data.

In the report, ARK Invest names Tesla and Waymo as two leaders in self-driving, and points out Tesla's data advantage:

ARK views players like Tesla and Google as leaders in the autonomous space. Tesla is the only automaker currently selling consumer vehicles equipped with the hardware necessary to collect autonomous data, giving it an unparalleled data library and the ability to advance and teach its autonomous systems quickly. Google may have the best performing autonomous car from a technological perspective and is taking steps toward commercialization with its recently formed entity Waymo.

Above, I described the importance of the 11 billion miles threshold. At the current rate, it will take Waymo's cars over 6,000 years to drive 11 billion miles. That means there may be no way to statistically validate that Waymo's self-driving cars are actually safer than human drivers, unless Waymo dramatically enlarges its fleet. This could prove to be a serious problem for Waymo. Without the data to show it, how will we know Waymo's cars are actually safer?

Tesla has ambitious plans to automate factories

Tesla is going meta with its plans for full factory automation. Tesla is using deep learning and computer vision to make fully autonomous cars. It wants to use that same technology to make fully automated car factories.

With Model S and Model X production, Tesla has a very low level of automation relative to the rest of the car industry. Even with Model 3 production, Tesla is hoping to at best match the automation level of manufacturers like Toyota (TM). Yet in some to-be-announced future factory Tesla’s aspiration is to reach an unprecedented level of automation: no humans directly interacting with cars, just machines. The humans’ job would just be to the keep the machines running.

CEO Elon Musk describes Tesla's vision for a fully automated factory.

Factory automation is an area where Tesla can leverage its software acumen. Deep learning is opening up new possibilities for industrial robots. Software is therefore becoming increasingly important in manufacturing. This is a potential leg up for Tesla because other car manufacturers have no particular competence in software.

Already, CEO Elon Musk has “refocused most of Tesla engineering... into designing the factory.” This is an area where a vertically integrated software and manufacturing company could foreseeably leapfrog old school manufacturing companies that are committed to incremental change and dependent on suppliers for software.

On the latest earnings call, we got an update on Tesla's factory automation R&D. The parameter that Musk focused on most was the speed at which robots move:

Like really I think speed is the ultimate weapon when it comes to innovation or production. And we are pushing robots to the limit in terms of the speed that they can operate at, and asking our suppliers to make robots go way faster, and they are shocked because nobody has ever asked them that question. It's like if you can see the robot move, it's too slow. We should be caring about air friction like things moving so fast. You should need a strobe light to see it. And that's incredibly critical to CapEx efficiency. And obviously we're going to be designing a lot of the robotic elements and what makes the robots internally. So yes, because current suppliers are just too slow to respond in some cases.

Clearly, if robots are building a car, the robots will build the car faster if they can move faster. This is what makes robot speed an important parameter. Production speed is similar to production scale. If cars are produced faster, more are produced per week. Fixed costs are then spread across a higher number of cars produced.

In a future factory, CEO Elon Musk wants these robots to move so fast they can't be seen with the human eye.

From the sounds of it, Tesla is not just developing its own software for robots, it is also designing its own robot hardware and its own production equipment to build the robots. Vertically integrating software and hardware has made Tesla a leader in self-driving car development. It is interesting that Tesla is pursuing vertical integration of software and hardware for factory automation. Perhaps it is as the computer scientist Alan Kay says: "People who are really serious about software should make their own hardware."

How things could go wrong

It's always important to consider risks and uncertainties. Here's how things could go wrong for Tesla:

  • The Model 3 production ramp could run into more serious trouble than it already has. If Tesla only achieves a run rate of 3,000 cars per week in Q1 2018 and 8,000 cars per week in Q1 2019, that doesn't change the result: it will still accumulate 11 billion miles of driving data by 2020. Likewise, if it achieves a run rate of 10,000 cars per week in Q2 2019 rather than Q1 2019, it will still reach 11 billion miles in Q4 2019. But if shortfalls much worse than these occur, the 11 billion miles threshold will not be crossed by 2020.
  • As I've written about previously, a large incumbent like General Motors could make up for what it lacks in software ability with sheer scale of data collection from millions of production cars.
  • Tesla could have insufficient sensor technology in Hardware 2 cars. It could turn out that full self-driving at a level of safety 2x higher than the human average requires lidar, thermal imaging, microphones, or other sensors that Tesla isn't using. (I explored this question in depth in a recent article.)
  • Tesla could continue to struggle with manufacturing. Full factory automation may prove too difficult. If the Model 3 suffers from reliability problems, that may hurt consumer demand, which would in turn hurt Tesla's data collection effort.

Prospective investors should be patient and have a high risk tolerance in order to consider investing in Tesla. Fully autonomous cars are a long-term ambition, and fully automated car factories are even longer term. With new technology, there is irreducible risk and uncertainty.

Conclusion

Software is eating the world, and cars are next. It stands to reason that the only car company that is also a software company is in a prime position to capture market share and achieve rapid growth during this transition. Tesla is attacking the traditional automotive industry with software on two fronts: vehicle autonomy and factory automation. Leading the industry in either one of these efforts would be a groundbreaking success for Tesla. Leading in both would give Tesla both a better product and a better manufacturing process than competitors. That is a winning combination.

Disclosure: I am/we are long TSLA.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

About this article:

Expand
Author payment: $35 + $0.01/page view. Authors of PRO articles receive a minimum guaranteed payment of $150-500.
Tagged: , , , Auto Manufacturers - Major
Want to share your opinion on this article? Add a comment.
Disagree with this article? .
To report a factual error in this article, click here