When Elon Musk took over as CEO of Tesla Motors 11 years ago, autonomous driving was not on the radar. The mission of the company was to accelerate the world’s transition to sustainable transportation. This would be achieved through the production of electric vehicles that were more compelling than any gas car.
By October 2014, at Tesla’s “D” event, the idea of autonomous features in the Tesla Model S was first announced — well after advanced driver assistance features had already been released by other car makers such as Mercedes and Nissan. However, Tesla offered a novel release strategy: build the hardware into all cars, and enable software upgrades to improve the features as time went on. Tesla’s unique ability to update their software over the air enabled this in a way no other automaker had ever done before. The first version of Autopilot had a single camera, along with radar and sonar. It was not designed to completely drive the car, but rather to make driving safer and easier. It was first released with just adaptive cruise control, by October 2015, steering within a single lane was added.
Less than two years later in July of 2016, autonomy moved front and center for Tesla with the release of Master Plan, Part Deux. This laid down a core mission for Tesla to improve human safety by a factor of ten. By this time it had become obvious to Elon that cars that drive themselves would be much more desirable and valuable that cars that could not. Therefore,it became a strategic requirement for Tesla to lead in autonomous driving.
The new Master Plan also outlined a philosophical approach that it was better to release autonomous software features even if they didn’t work perfectly and let drivers override the system. This would in rare instances cause some accidents — but as long as the probability of these accidents on Autopilot was lower than humans driving unassisted this was desirable. Thus at Tesla, releasing autonomous functionality into the market was seen as a moral imperative. An imperative worth the negative repercussions of abetting a small number of drivers in their unsafe and sometimes fatal behavior. Such a strategy would result in a net reduction in the number of injuries, deaths and accidents, but over time these numbers would tip increasingly in their favor towards prevention of accidents.
Next week at the Autonomy Investor Day event, Elon Musk will outline how autonomous driving technology will soon become the feature that cements the success of Tesla’s electric car strategy, as the cars become increasingly capable of outperforming human driving. He will base his prediction on some simple math.
But viewing him as the boy who cried unicorn, one too many times, most of the automotive industry will largely disregard both the timelines and the fundamental thesis of his argument.
For Elon, however, understanding the advancement impact of Tesla’s autonomy is just a simple exercise in mathematics: growth curves and probabilities. There is no real argument to be had if you have the data and you understand the math. Just as it’s hard to argue with Newton on mechanics, it’s hard to argue with Napier on growth (he discovered exponents).
And since this is Autonomy Investor Day, the focus is likely to be placed on the financial implications of autonomy for both Tesla and its customers.
Similar to the Model Y launch event, there is not likely to be much new information at next week’s event. It will package what was outlined at the beginning of the 2018 Q3 earnings call and in recent interviews on the topic of autonomy, before a lot of in-car demos for the investors.
The event will cover the release of the new ASIC specialized processing chip that has been in development for three years, initially led by the legendary Jim Keller of AMD and Apple fame, along with a team from Apple and AMD including Peter Bannon, who took over leadership of the team. They will review the speed at which this chip can perform simple matrix multiplication: about 20 times faster than the existing NVIDIA Drive PX2 chips that have been shipping in the cars for the last couple years. And notably with no increase in cost, size or power consumption.
Next they will talk about the advancement of the multi-camera processing neural networks that Tesla is using for image recognition. This, development has been lead for the last two years by Andrej Karpathy, who previously worked with Elon at Open AI. They will describe how the hardware limitations previously constraining this software have now been addressed with significant headroom. The new FSD Autopliot computer not only runs computations faster, but allows for bigger networks with more neurons, which increases the accuracy of all predictions. On top of this they are layering a rapidly increasing the amount of data to train the networks, and encompassing their unique ability to use human interactions at scale as input for training. This approach is not possible in “always fully autonomous” systems like Waymo’s that do not normally allow drivers to interact with them.
Reading between the lines Elon will effectively state that the approach everyone else is pursuing towards full self driving is fundamentally flawed. In other words, efforts underway at companies including Google’s WAYMO, GM’s Cruise, Ford and Aptive with their reliance on LIDAR, lack of low-cost distribution, and lack of constant human input in the training model is a non-competitive approach compared with Tesla’s strategy of deploying millions of cars equipped with a full suite of low-cost sensors (notably excluding LIDAR) that can train the network in conjunction with human actions.
Elon will restate a somewhat semantically ambiguous assertion that Tesla vehicles are now an “appreciating asset”. And while the actual market value of the cars may not outpace the depreciation unless demand radically outpaces Tesla’s ability to manufacture, this does not negate the core point. The undeniable reality is that Teslas are becoming better cars to drive as the autopilot software gets upgraded (this fact also applies to other types of software and hardware upgrades including faster charging times, higher speeds, and more charging stations). The desirability of these cars will accelerate due to the new full self driving hardware. And thus the utility of the cars will certainly appreciate in a significant and rapid manner as they become safer and easier to drive. This will flatten the depreciation curve and it’s possible that if demand for full self driving is sufficient to allow Tesla to radically increase the price, this will give those who ordered it at previously low prices a small windfall in their car values.
They will then outline how easy it is to upgrade this chip for all Teslas built since 2016, which could be viewed as a major accomplishment in backwards compatibility. This means that the fleet size of Tesla vehicles that can be upgraded to full self driving for a few thousand dollars is about 400,000 today, and will double in the next year..
On top of their rapidly evolving autonomous software, Tesla will outline the essence of their plan to compete directly with Uber and Lyft in ridesharing, in roughly a 3 year timeframe. These numbers make fleets like those envisioned by Uber and Waymo at 20k to 100k seem paltry. When the number of hours the combined fleet of Tesla’s cars could work is added up, it will easily eclipse all the human driven hours of Lyft and Uber today.
With over 5 million autopilot ready cars in the market by end of 2022, if half of these were committed to the Tesla network, this would be 2.5 million vehicles, that could easily drive 12 hours/day without interfering with their owners needs. This would enable about 12m hours of rideshare enabling roughly 95m rides a day, which would likely eclipse Uber and Lyft’s maximum potential capacity combined.
However, the Tesla Network’s cost basis will be substantially better than any other service. Cost per miles of a model 3 electric car will be about half of most others over the lifetime of the vehicle, which by all indications will be somewhere between 500k and 1m miles for every vehicle. Even with 50% battery degradation, an autonomous car could make a lot of money during a day, with the need for only one autonomous recharge per day.
High reliability of the drivetrain, along with low maintenance further sweeten the economics and operations. If you believe Tesla’s predictions on Model 3 drivetrain longevity (supported by real word-data on the Model S & X), a $40k Model 3 will last over 600k miles with need for one battery module replacement costing about $6,000. With all other maintenance, this might come to $60,000 for 600k miles or 10 cents a mile with another 2.5 cents for fuel, and 1.5 cents for tires. All in just about 15 cents a mile.
Connectivity will also create unprecedented efficiencies for maintenance and care of cars in high utilization. Where the biggest maintenance issues: car washes, car vacuums and tire service can be ordered and scheduled millions of times a day with a single programmatic “click” that emerging service companies such as Cox Mobility and Amazon Home services will effortlessly fulfill through programatic communication.
Tesla will then detail their thoughts on the Tesla Model 3 lease, which will serve as a feeder to the future autonomous ridesharing service imaginatively named “The Tesla Network”.
At $3k down and $500/month, this means that the first owner of the car will pay about half the cost of the vehicle for about 30k miles of use, or about 70 cents a mile. In doing so, they will subsidize the cost of their eventual driverless rideshare fleet, by leasing the vehicles to buyers, who may not be able to purchase the vehicle once the lease ends, and who for about half the cost of the vehicle will get the first 5% of the cars potential miles. Tesla will now have an asset that will cost them only about 4 cents per mile for the remainder of the cars life up to 600k miles, and it’s very possible they will go much longer.
And with this math Elon will in his own way declare victory over the legacy automotive industry, and leave it as an exercise in mathematics for the other auto makers to digest this all.
By his logic, for anyone else to create an offering competitive to Tesla would entail a complex series of steps:
- building electric cars that consumers love, profitably at scale, with low operational costs per mile, and very low maintenance requirements
- creating a modern, upgradable, flexible and extensible software architecture that spans the vehicle, chargers and the cloud
- ensuring the security architecture can outmatch the best hackers by seeking them out and plugging all vulnerabilities as they arise
- integrating sensors into these vehicles that could be deployed at scale in the market
- developing specialized chips comparable to Tesla’s new FSD computer
- developing neural nets and path planning software that can be trained by car drivers, and that work in a generalized manner across the world
- creating a means by which vehicles can always charge reliably in an autonomous manner
- deploying these vehicles into a driverless Uber/Lyft like service after regulatory changes allow this
And they key factor here, is unless you have all of this, you are not in the running. Every component of the equation is critical to the network.
Autonomy will also solve the other major challenges for Teslas. It will radically increase demand for Tesla vehicles even before driverless regulatory approval due to the attraction of nearly effortless driving. Once driverless becomes legal, it will solve one of the biggest problems for electric ownership for people who live in apartments or other environments where overnight charging is not an easy option. The car can simply go charge when it needs to and come back, earning money on the way. And of course at the same time it will transform the equation on ownership vs. rental or rideshare into something highly flexible and much more cost-efficient for anyone who wants to share their car.
And with all this outlined, in what may be the 2nd longest Tesla presentation in history, the reaction will be the same as it has been for the last two years: extreme enthusiasm from the Tesla believers, accompanied by yawns, claims of hyperbole, and complaints about the morality of Tesla and the implications of their actions on the rest of the industry from those who are not yet ready to believe. This skepticism may be warranted due to previous missed deadlines, or because the skeptics haven’t seen the same data, haven’t done the modeling or just aren’t inclined to believe extrapolations until they are fully realized. Or it may be due to the psychological tendency to rationalize away fundamental threats to your livelihood that are not easily addressed.
Demonstrations at the event of demo-mode Tesla Model 3’s driving on city streets will not sway those who disbelieve. Waymo has been driving on city streets for years they will note and refer to Navigant’s declaration of Tesla being near the bottom of self-driving technology.
No auto makers will convene emergency board meetings next week to discuss how to combat Tesla’s full self driving strategy. Nor will they rush to equip vehicles with lidar-less sensor suites that can be updated over the air, ensuring the collection of data at scale. The industry will yawningly dismiss Elon’s Musk claims regarding the timing/capability curve of Tesla’s full self driving features, and continue on their current trajectories.
And Elon would admit, it is possible that he is wrong about some of this, but he just doesn’t see where the flaw in the logic could be. If he is wrong, there will be a lot of unhappy Tesla customers who purchased full self driving software options, creating a major messaging challenge for the company.
There is no question that even with the new computer chip, customers this year will still experience situations where the autopliot does not behave correctly, and requires the driver to override their commands. There may be a few accidents with autopilot on, and these will receive extreme attention in the press.
But if Elon’s math is right, as we get into next year it will become increasingly apparent every month, that the curve Tesla is advancing on in terms of delivering value through increased safety and a more effortless driving experience is not linear. It will become obvious that the strategy to cohesively architect a network of electric vehicles, massive charging infrastructure, upgradeable software architecture, low cost sensors, specialized silicon, neural networks, autonomous software and battery manufacturing at massive scale gives Tesla a huge competitive advantage as it scales, leaving little room for the emergence of a Tesla killer.
And meanwhile, Elon will shift focus to his more recent primary goal for Tesla — to become the best manufacturer on earth. A feat which he views as a much harder problem than making the best car network on earth.