I recently got upgraded to Tesla (NASDAQ:TSLA) FSD 10.69.3.1 and my thesis is that it shows the way iterative improvements continue since I wrote about my experience with version 10.69.2.2 back in September. These software versions are also known as 2022.36.20 and 2022.20.19, respectively.
Up until now there were requirements for FSD such as a safety score of 80 or more and a minimum of 100 miles on AutoPilot. These requirements are no more such that everyone in North America now has access to FSD per a tweet from CEO Elon Musk:
Opening up FSD to everyone in North America is a prodigious development! We’ve already seen the way Tesla has used data from over 100,000 FSD vehicles to put improvements in place for version 10.69.3.1. Now they will be collecting even more data such that version 11 will be the best one yet.
The FSD 10.69.3.1 Experience
The 10.69.3.1 release notes say the following:
Converted the Non VRU Attributes network to a two-stage network, which reduced latency, reduced incorrect lane assignment of crossing vehicles by 45%, and reduced incorrect parked predictions by 15%.
My Tesla now seems smoother and more responsive during and after turns. I don’t know if this is because of the release note above but it is noteworthy that after making a turn, we’re then going the same way as some vehicles that were formerly crossing.
Headlines tend to focus on autonomy and LiDAR but there is much more to the story. Version 10.69.3.1 provides automation but not autonomy. Some feel that the benefits from this technology are all or nothing with respect to autonomy but I disagree. LiDAR isn’t essential at this time and we’ll continue to find more use cases for this technology as automation keeps coming along iteratively.
There is a big difference between being safe and feeling safe. The Verge reports that Tesla wrote the following to US Senators on March 4th:
Tesla’s Autopilot and FSD Capability features enhance the ability of our customers to drive safer than the average driver in the U.S.
I believe that FSD already keeps me safe in many respects but it will feel safer as they continue to improve smoothness and fluidity with future versions. There is still work to be done in this area but I am confident that progress will continue with version 11. Like cruise control, FSD helps with safety by preventing me from driving too fast. It also helps me feel relatively safe when having distracting conversations, whether they’re on the car’s speakerphone or with passengers.
Here are 10 areas where we should see iterative FSD improvements and they do not involve LiDAR:
If we’re turning left soon then get in the left lane.
Don’t go to the center of wide lanes – just stay near the left border of the lane.
Don’t pull right up next to erratic drivers – give them space.
Avoid excessive lane changes.
Avoid changing lanes in intersections.
After making a protected right turn, start following the arrows and merge left sooner.
Be a little quicker and responsive when it’s our turn at a 4-way stop.
The braking is usually good but there are times when it should be smoother.
U-turns tend to not be fully automated; the driver has to hit the accelerator.
Improve overall smoothness – especially in parking lots and on small streets.
FSD 10.69.2.4 kind of reminded me of an occasion when we scheduled an Uber to take us to the airport with little time to spare. The older driver we got that day was very cautious and it took him a long time to get us to the airport. It seemed like a fairly safe ride but it was stressful because of the time constraints and the fears we had about getting read-ended. 10.69.3.1 is more practical than 10.69.2.4 but it can still be slow in many situations. I would expect it to be slow for those who choose to be extra cautious with their settings but mine are not conservative. The choices for “Speed Based Lane Changes” are Disabled, Mild, Average and Mad Max; my vehicle is set to Mad Max. The choices for “Full Self-Driving (“Beta”) Profile” are Chill, Average and Assertive; my vehicle is set to Assertive. Responsiveness has already improved from 10.69.2.4 to 10.69.3.1 such that I don’t have to tap the accelerator as much. I expect this trend to continue with version 11.
FSD will continue improving such that use cases will expand broadly in the near future. Soon, I think some Hertz rental car customers will request Teslas with FSD such that they can plug in destination addresses and have the car make the navigation decisions for them. It is a long road ahead before we are in the era of robotaxis but many other use cases are within grasp such that FSD already adds substantial dollars to Tesla’s valuation.
Introduction To Margin Considerations For Valuation
Part of any valuation framework should include some understanding as to why Tesla has high margins on battery electric vehicles (“BEVs”) while other companies say their BEVs won’t be profitable until 2025 or later. Only after understanding current margins can we speak to their viability and the valuation implications. Simplicity is a key theme at Tesla and it helps explain many of the reasons as to why margins are high compared to others in the auto industry. I put together some margin considerations below in an order that makes them easy to explain but this is not the order of importance.
Regulatory credits are not sustainable in the long run; at some point they will end. In 2020, Tesla had unadjusted revenue of $31,536 million but 5% of this or $1,580 million was from regulatory credits such that adjusted revenue was $29,956 million per the 2021 10-K. In 2021, unadjusted revenue went up to $53,823 million and regulatory credits went down on a relative basis to 2.7% of this or $1,465 million such that adjusted revenue was $52,358 million.
If we assume that 100% of regulatory credit revenue reaches the gross profit and operating income lines then 2020 unadjusted gross profit of $6,630 million changes to $5,050 million on an adjusted level and 2020 unadjusted operating income of $1,994 million changes to $414 million on an adjusted level. In 2021 this means the unadjusted gross profit of $13,606 million changes to adjusted gross profit of $12,141 million while unadjusted operating income of $6,523 million becomes adjusted operating income of $5,058 million.
Unadjusted gross and operating margins for 2020 were 21.0% and 6.3%, respectively. On an adjusted basis, these are 16.9% [$5,050 million/$29,956 million] and 1.4% [$414 million/$29,956 million], respectively. In 2021, unadjusted gross and operating margins improved to 25.3% and 12.1%, respectively. These were 23.2% [$12,141 million/$52,358 million] and 9.7% [$5,058 million/$52,358 million], respectively, on an adjusted basis.
Looking at the first 9 months of this year in the 3Q22 10-Q, regulatory credit revenue was $1,309 million which is 2.3% of the overall revenue of $57,144 million.
In summary, regulatory credits dropped from 5% of revenue in 2020 down to 2.7% of revenue in 2021 and down again to 2.3% of revenue for the first nine months of 2022. Unadjusted operating margins were 19.2%, 14.6% and 17.2% for 1Q22, 2Q22 and 3Q22, respectively. The adjusted operating margin should be a few hundred basis points lower which is still very impressive! At some point in the future, regulatory components should be 0% of revenue so it’s important to think about this component when talking about future margins.
Teslas are not as prestigious as Ferraris (RACE), but they do have some of the same type of pricing power we see with luxury vehicles. A November 2022 Quartz article says Tesla’s pricing power is a major factor that enables them to make 8 times more profit per car than BYD (OTCPK:BYDDY):
Tesla’s intense focus on EVs and its strong brand, which allows it to raise [prices] and pass on higher material costs to customers, are two key factors behind the automaker’s high profits per vehicle delivered.
Ostensibly Tesla has 4 models which are the model 3, the model Y, the model S and the model X. However, they make the bulk of their money with just 2 models – the model 3 and the model Y. Wolfe Research Analyst Rod Lache noted that Tesla sold 940,000 units in the previous year with outstanding margins. At the 2022 GM (GM) Investor Day, he asked how GM would get up to high margins with their BEVs by 2025. GM CFO Paul Jacobson responded as follows:
So we’re looking at 1 million units of production across a much wider slot [than Tesla]. So some of those scale benefits that you could get from doing 940,000 units across just 4 models. We’re not going to have.
Again, Tesla has economies of scale with the model 3 and the model Y; combined, they had deliveries of over 325 thousand in the latest quarter per the 3Q22 press release.
Tesla has learned to simplify the body structure of their vehicles and this cuts down on manufacturing costs. CEO Musk explained this at the 2022 Annual Meeting:
And one of the things we’ve done is create the the largest castings that have ever been done and they’re very complex castings and so we’re able to take a 171 pieces of metal and go from 171 pieces to 2 and in the process make it lighter, stiffer with better ride handling, better noise vibration harshness [and] better sealing against water. So it’s really better in every way.
Here is a slide from the 2022 Annual Meeting that shows this simplification of the vehicle structure:
Due to the casting developments at the Austin and Berlin factories, fewer robots are needed to build the Model Y:
A November 18th video from The Tesla Space talks about the significance of Tesla’s casting approach:
Manufacturing the rear underbody of the model Y used to involve 300 individual robots and the reason for that is that a typical vehicle frame is made from a bunch of little parts all stuck together either by welding or fasteners or adhesives. So you need automated stamping machines to fabricate all of those bits and then a bunch of robot arms to put them all together. In order to fix that, Tesla enlisted the top die casting company in the world, an Italian firm called IDRA. And they had them build the world’s largest and most powerful casting machine – the Giga Press. And with that, 300 robots were replaced by one singular machine and casting process. And that’s only gotten better with the new Gigafactories at Texas and Berlin where they make for the rear and front of the model Y frame with these casting machines.
The video goes on to say that Tesla’s Cybertruck will use an even bigger casting machine. Further, it says Tesla will make a smaller and cheaper vehicle than the model 3
by using advanced manufacturing to push the production time and cost to an all-time new low that could mean casting the entire vehicle in just one big part like a Hot Wheels car.
In the FSD section above we were talking about automation with respect to driving. Here we’re talking about manufacturing automation which helps to lower costs and increase the speed for producing vehicles. The Electric Viking has a November 18th video in which he talks about the automation advantage Tesla has over Volkswagen (OTCPK:VWAGY) (OTCPK:VWAPY) (OTCPK:VLKAF):
Today [Tesla] uses 70% automation in its building process. Volkswagen said it uses less than 20%. Volkswagen has said it takes them more than 30 hours to build an electric car. [former VW CEO] Hubert Diess said it takes Tesla around 10.
One reason Tesla has nice margins is that they aren’t saddled with marketing expenses. Tesla spends close to $0 on marketing for each vehicle while legacy companies like Ford can spend $600 or more. Ford CEO Farley spoke about this at the Sanford Bernstein Conference on June 1st:
We spend $600 to $700 on a vehicle to promote it, and we spend nothing post warranty on the customer experience. And the problem is parts business, which historically has been very profitable, we only get maybe 10% or 20% of the customers who come back to us. It’d be much better if we try to develop an ecosystem where 100% came back. And we gave them experiences, and that’s our marketing. You buy a Ford Model e and after a year, we’re going to give you a complete detail of the vehicle, check all your software is up to date, you get a complete birthday for your vehicle. We should be doing stuff like that instead of doing Super Bowl ads.
An October 2021 Visual Capitalist article shows that Tesla spends on R&D as opposed to advertising:
So if you and I both run competing subscription businesses, and yours is older than mine, then even if our services are equally light with exactly the same churn curves by customer cohort, your average churn rate will be lower than my average churn rate, which means that more of your marketing dollars are going to support new subscriber growth and more of my marketing dollars are going to replace churn subs.
We’ve seen examples of the thoughts above. AWS has been offering cloud services longer than Microsoft Azure (MSFT) and Google Cloud (GOOG) (GOOGL) such that AWS has lower marketing costs and higher margins. Netflix has been offering streaming services longer than Disney + (DIS) and WBD (WBD). Again, Netflix has lower marketing costs and better margins. For the most part, Tesla is not a subscription business but some of the thoughts above apply to them anyhow. They have been offering BEVs longer than competitors such that they have lower marketing costs and better margins than competitors.
Saying Tesla is heavily vertically integrated compared to other auto companies, CEO Musk explains in the 3Q20 call that they make the machines that make the machines:
We have a massive amount of internal manufacturing technology that we built ourselves. We literally make the machine. In fact, we design it – so like, okay, what are the things we want to make, design a machine that will make that thing, then we make the machine. This is what – this makes it quite difficult to copy Tesla, which we’re not actually all that opposed to people copying us, but it’s quite difficult because you can’t do catalog engineering. You can’t just [say] I’ll pick up the supplier catalog, I’ll get one of those machines – one of [those] machines, bingo, I’m now Tesla. You have to – there is no catalog. What catalog? So we made the machine that made the machine that made the machine.
Tesla makes their own seats and they even design their own chips! A June 2021 Tesmanian article explains how Tesla is special with respect to vertical integration:
In a revised application for the construction of Giga Berlin, Tesla indicated that it plans to manufacture many components in-house. For example, the company wants to add steps for manufacturing seats, plastic components such as bumpers and mirror caps, and for painting them. It should be understood that this is uncharacteristic for automotive factories in the industry. Most other manufacturers buy off-the-shelf parts from suppliers and then simply install them in their vehicles. In addition to the aforementioned components, Tesla will manufacture at Giga Berlin the most important part of any electric vehicle – batteries. While all other companies buy battery cells (most often from Asian suppliers) the Californian manufacturer has developed its own.
BYD makes batteries for their own cars as well as for other carmakers. Tesla makes their own 4680 batteries but they also rely on suppliers like Panasonic per a May 2022 InsideEVs article:
An August 2022 electrive.com article notes that BYD has started selling LFP cells to Tesla for their Giga Berlin location. The article says there aren’t plans for BYD to enter the picture in Tesla’s Shanghai location at this time: “Specifically, around 70 percent of the battery cells installed in the Giga Shanghai come from CATL and 30 percent from LGES.”
Dealer markups can be extraordinary for legacy vehicles but Teslas are sold via a direct-to-consumer (“DTC”) model.
Summary Of Margin Considerations For Valuation
In summary, I believe Tesla’s margins are mostly sustainable such that they are worth much more than other car companies on a unit basis.
Trailing-twelve-month (“TTM”) gross profit, net income and free cash flow (“FCF”) are $19,923 million [$15,076 million + $13,606 million – $8,759 million], $11,223 million [$8,880 million + $5,644 million – $3,301 million] and $8,921 million [$6,146 million + $5,015 million – $2,240 million], respectively. Some of this is from FSD and energy but much of it is from the core car business. Much of the FSD revenue has been deferred through 3Q22 such that it isn’t a big part of gross profit and net income. However, the deferral shouldn’t impact FCF so the FCF figure isn’t as clean as I would like it to be in terms of separating the core auto business from FSD.
As we saw above, I think much of the gross and operating margins are sustainable. The business is still growing rapidly as we appear to be reaching an inflection point such that BEVs are starting to take over. P/E ratios have been coming down as interest rates go up but I think this core business still deserves a generous multiple given the growth. I think the core auto business is worth about 40x TTM net income or $450 billion.
In January, Mark Cuban tweeted that when making investment decisions, he asks himself how good a company is at AI:
Seeing the progress with FSD, I believe Tesla has been very good at harnessing AI. In a November 4th interview with Ron Baron, CEO Musk talks about autonomy:
I think at a very high level, I’d say that autonomy is an insanely fundamental breakthrough. And no one is even close to Tesla for solving generalized autonomy or generalized self-driving vehicles; no one’s even close. And with self-driving, as I was talking about earlier, the car becomes – call it roughly five times more useful but it costs the same to build. Now can you imagine what would happen if a company was doing like a 25 to 30 gross margins but suddenly that same thing was five times more valuable? What would that do to the value of Tesla and the value of that car? It boggles the mind actually.
Through the 3Q22 financials, it is my understanding that Tesla only had about 160,000 FSD beta testers in the US and Canada. There is a twitter discussion that talks about the financial ramifications now that FSD is available to everyone in North America. Depending on pricing, I think it may not be long before Tesla sells close to that many FSD packages in a single year. If they are soon selling 150,000 FSD packages per year at an average price of $15,000 then it means annual revenue of $2,250 million. We’re talking about software here where margins are excellent such that half of it could make its way to the bottom line. There is also the option to license this out to other car companies. There are a wide range of possible outcomes here but I believe Tesla’s FSD business could be worth $100 to $200 billion.
We also have other considerations like the Optimus program which is their humanoid robot. The outcomes here are even wider than the FSD business but I can see a valuation range of $0 to $100 billion for other considerations.
Putting all this together, I think Tesla’s valuation range is about $550 to $750 billion.
The 3Q22 10-Q shows 3,157,752,449 shares outstanding as of October 18th. Multiplying this by the November 23rd share price of $183.20 gives us a market cap of nearly $580 billion and the enterprise value is fairly close to this figure. The market cap is within my valuation range and I think the stock is reasonably priced for long-term investors.