Email: email@example.com – reach out if you have Q’s or feedback. I’d *love* to hear it please.
The most important part of this letter is that I am going to give you both numbers and a vision. However, the vision and goals of the company is far far more important right now than the projected numbers because, let’s be honest, projected numbers are always way off. So let’s try to understand where I think Tesla’s goals are, how realistic they are to come true, and then work back from there to find out if Tesla is investable today. All projections are from 10 years from now in the year 2030 but I do give a roadmap of what I think will happen on a year to year basis (I expect the year to year stuff to be wrong but 2030 projection to be much more accurate).
Also, before reading this, I want you to wipe out all preconceived notions of what is possible and what’s not. Instead, let’s rely on pure hard cold facts to build up and see if what Tesla wants to do is possible. I implore you to think “based off the facts, could this really happen” instead of dismissing something that seems outlandish and saying “nah, that’s not possible because it’s never be done”. Just look at the facts and let’s build from there.
I divide this thesis into a few parts:
- FSD (full self-driving)
- Vehicle manufacturing
- Energy Generation & Storage
- Vehicle sales
- Financial Models
Remember, try to think of the world as what it could be based off facts; don’t dismiss something just because it’s never been done before.
I believe that Tesla is gearing up to become a:
- self-driving technology company (MaaS, mobility as a service)
- Battery and vehicle manufacturer where both categories ex. Software have 0% margin (all profits from software)
- electric utility company
I believe Tesla’s goal is to have the cheapest cars on the market to gain marketshare and then use their technology edge to then make profits from software. As an example, I think that Tesla will eventually sell cars at-cost but then upsell on software services such as FSD & in-car app store. I also think their goal is to become the leading manufacturer of batteries (with both the highest density & lowest cost) in order to use batteries as energy storage for the grid on both a home level and utility scale projects. In addition, creating both a “solar roof” in line with the cost of a normal roof + utility scale solar projects.
They already have almost all of the tech necessary and it’s just about scaling up production. Tesla already makes their own batteries through their Roadrunner project (yes, they still purchase batteries from Panasonic, LG, and CATL) and with their purchase of Maxwell batteries they now have tabless anodes (increases cell density by up to 20%), dry cell production where they don’t need to oven cure their batteries (making the space required to make a cell down to just 1/16th of the original space, reduces Capex, and 20% cost reduction due to no expensive heating and processing), and are targeting $65 / kwh batteries in the next year. However, when it comes to batteries we’ll see a lot more on Sept 22nd on their battery day. On their FSD they now are completing their fundamental rewrite of FSD and with the completion of their Dojo training computer I believe they’re 9 months from going from point A to point B w/o touching the wheel in most cases and 2 years to being truly Level 5 (no need of a driver in the car). With a million cars on the road detecting edge cases, they’re far and away the best in class. Their production Capex per unit of capacity has gone from ~15k per unit at Fremont factor all the day down to an expected 2k per unit of production per year at Berlin as they expand production (including redesigning their cars to make it easier for production)
All the stuff for the future is here; it’s just about scaling.
Let’s go through each section though, shall we?
FULL SELF DRIVING
This is going to be the hardest to both comprehend and understand but before we begin I need to say a word about how hard of a problem this is. Elon has done a huge disservice to Tesla by saying it’s right around the corner for far too long however I believe he’s finally right: it really is right around the corner.
Let’s start with the question of “what does it take to become level 5 autonomous”? Level 5 refers to no need for a driver in the car (i.e. a “true” self-driving car). It’s pretty simple: you need to understand the world around you in both depth, understanding (e.g. this sign says to yield), a high level of redundancy in both sensors & computing, and decision-making. Fundamentally Tesla has asked “if two eyes can do it and a brain, why can’t cameras & a computer”? I think they’re right. Here’s their approach:
- A ton of cars on the road helping to get data
- Have cameras create a 360 degree field of view
- One hell of a super computer
- Couple other sensors such as radar and 360 degree view with ultrasonic sensors
Here’s a high-level understanding of how they operate: they have a team of 500 (to grow to 1k in the next year) human labelers that take up to 30 second pieces of video from the car’s cameras and make 3d “labels” of everything in that video. These humans will label a stop sign, street lights, cars, all the different signs, etc etc etc. These labels get put into a huge computer along with what the driver did (speed, acceleration, turn signals, degree turn of the steering wheel, etc) and then the computer “trains” on this data. In layman terms, a computer takes the videos and finds patterns of what the car saw and compares it to what the driver did. So the computer training the neural net should detect patterns where when a stop sign is present the car then predicts the brake pedal to be pressed.
Now, let’s talk about “shadow mode”. This is where you drive your car on a day-to-day basis and your car’s video keeps inputting the data into the computer and it asks “is the driver driving the car like I, the computer, would drive it?”. If it deviates significantly (such as the driver slamming on the brakes when the computer predicts all is normal) then the car saves that portion of the video where it starts to deviate from the computer’s expected control and uploads it to Tesla so someone can label the video and input the labeled video into their training algorithms. The computer then re-trains the data and the next software release the computer would ideally do what the human driver did.
I cannot emphasize this enough: the amount of data required for FSD is the limiting factor for so many car companies. Waymo, Cruise, etc don’t have the cars required to gather enough data to then train their machine learning algorithms on. You need hard real-world data because driving is something that has a ton of edge cases.
Let’s take a real-world example: stop signs. Normal enough. However, what if there is a construction truck carrying a load of stop signs on the highway? The computer might detect that you’re driving (in our case on a highway), there’s a stop sign, so slam on the brakes and come to a stop. Right? Well as the driver keeps going normally the computer understands “there is a huge deviation from what I was expecting to do and what the driver did… I need to send this back to Tesla to label and train on”. The video goes to the repository of videos the FSD machine learning algorithm trains on and then next time the computer trains it notices the pattern of “if a stop sign is on a truck (both labeled in the video by humans), these videos suggest I ignore the sign”. That’s why data is so important. It’s why you need millions of cars detecting edge cases of unexpected results that can upload the data back to train on… and only Tesla has this many cars on the road doing so or even having the ability to upload data.
Let’s talk about other companies’ approaches to FSD real quick. Some companies, mainly Cruise, Zoox, and Waymo, all use what’s called Lidar to create a 3d map of the world. Basically a lidar is a radar but in a different spectrum of light so it can “map” a ton of dots in a 3d plane with each dot’s distance. However, lidar is no longer the only way to create a 3d map. Researchers have found that given significant enough computing power and understanding the distances of just one part of a video you can then detect the distances of the whole part of the image. Basically what you can do with lidar + a small computer you can do with cameras, radar, and a huge computer… which is exactly what Tesla has. With lidar prices up to 15k per sensor on a gas car (which has a much larger running expense than an EV), how can a robotaxi network compete on price given their huge upfront costs in building the taxi? Said another way: Waymo’s taxi’s cost 6 figures for a minivan so their cost per mile is significantly higher than a tesla that has lower running costs + their FSD hardware is simple cameras.
So in essence what you need for FSD is just cameras, ultrasonic, radar, one hell of a computer in the car, and one super computer to train your machine learning algorithms on. Tesla was right in this approach. I’m a software engineer by trade who has dealt a lot with this world. If this evidence doesn’t convince you then trust me individually: Tesla has everything it needs to have the cheapest in-class FSD.
The next part is training: when you have a lot of labeled videos and want to build a machine learning program, it takes a beast of a computer to go through all of it and detect the patterns of how the program should behave. Just as Tesla did with it’s FSD computer in the cars, they’re in-house building a computer under their Dojo project that’ll live in their datacenter and be able to train their program against the data (get it? Project Dojo. Dojo is where you train? Engineering humor…). Now, rumors have it that just as their computer in the cars is an absolute beast, so will this computer. The release of this computer to start training the data will be in ~1 year’s time and when that happens I think we’ll continue to see a huge exponential growth in how good Tesla’s FSD is.
One thing I’d be remise not to mention is that I think there is a probability that the cars on the road will not be allowed to become robotaxi’s (which Elon Musk has stated that they would). I’m not sure if this will be the case or not but essentially I think the cars on the road today will allow the driver to do whatever they’d like (such as read a book, watch a movie, etc) but that it possibly will not be allowed to drive without a human in there. The reason is because of how the sensor suite is set up. If there is a bird that has exceptional aim & takes a dump on one of the side cameras then there’s no way to clean it off. This can be fixed via software (if you have a 3d representation of the fixed world & you know how fast you’re going then you could estimate where to go) but I’m not sold. The main forward-acting cameras & radar are build into where the windshield wipers are so I think those would be fine. However, I’m not 100% sold that the current cars can become robotaxi’s. Again though, the main point of this section is to show that they can do full self-driving.
I hope this shows you that Tesla’s FSD hopes are alive, well, and will come to fruition easily by 2030 (I expect FSD to happen in late 2021).
Note: this isn’t exactly how this all works but it’s close enough from a layman’s perspective
VEHICLE MANUFACTURING & SALES
Currently Tesla has:
- Model S
- Model X
- Model 3
- Model Y
- Semi (in pilot production est 2021)
- Cyber Truck (announced & est production late 2021)
- Compact vehicle (~20k price, to be announced in a few years)
- Roadster (announced; est production 2022)
EM has stated many times “the factory is now the product” and “the machine that builds the machine is the most important part”. His view is that the factory needs to be as absolutely as efficient as possible and they are on a mission to make their cars as cheap & affordable as possible.
As evidence of Tesla’s pure commitment to cost efficiency, Fremont has cost around $15k per yearly unit of production i.e. they spent roughly 7.5 billion on Fremont factory to have 500k capacity. However, China went up in less than a year (which is absolute breathtaking speed for a car factory) and its cost so far 1.2 billion for by year’s end 300k units of production and they’re estimating 1.5 billion for 500k units of production. That’s an 80% decrease in the cost per unit sold. On top of that, they’re expecting that to fall further to just 2k per unit of production for Berlin. As we can see, Capex should continue to ramp up due to 3 car factories being built or expanded on, 2 battery factories being expanded on (fremont’s road runner project & Nevada), but the efficiency of that has been dramatically decreasing over time.
With the car itself, I’m not going to talk much about this and instead just reference what the experts say. The go-to tear down automotive expert Sandy Munro has a YouTube channel he created during Covid to keep his team working and they tore the model Y apart. What they found was some of the best engineering him and his firm had ever seen on the electronics side of things (i.e. the computers & wiring) and astonishing car engineering. However, what’s *more* impressive was how they tore down a model Y built in April and noted that they believed 13 things could be improved on what the heat pump (the piece that heats the battery & car). They tore down another model Y for a customer in August and found all 13 suggestions they had were already made & in place. He goes on to state, almost speechlessly, that Tesla moves at the speed of thought: the moment they realize they can make the product better they do so without hesitation. He also states that under F and GM he’d be surprised if they got one of those improvements done in a year.
As an example of this innovation in car manufacturing, there’s a story that goes that EM was at his desk and he had a die cast model S sitting on it. He then asked his engineers “if you can just cast this die cast 6 inch car in one go, why can’t you with a regular car”? Cue: the giga casting machine that is now in production. Tesla has the largest casting machine in the world where it takes the whole underside component of the model Y and casts them into two pieces (later this will be improved to just 1 per EM). Instead of assembling 36 components per piece (72 in total), it’s now one large frame. No-one in the auto industry is doing this (and Tesla now has all the patents on it).
Another key point is how the car’s design is actually changing based off their factory designs. Berlin’s cars (which will produce the model Y to start and then the 3) are going to be significantly changed compared to US’s & even china’s cars. As evidence, they will be using 8 different giga casting machines to
Another example: EM wants to redesign the F’ing tire. He thinks tires can be redesigned to be cheaper than current tires and last up to 100k miles. This pure obsession with costs & reinvesting into the business is almost like Amazon. But seriously… they want to redesign the tire to make their cars just that much more efficient to operate.
Back to Sandy Munro, he talks about how impressed he is about Tesla’s commitment to manufactoring. As he says, “F and GM are screwed. They’re just screwed”. We also keep hearing “the competition is coming”… but as we’ve seen building an electric car is actually not quite the same as building an ICE case in point there are quite literally not enough batteries in the world to supply them. On top of the hardware & having a very different powertrain, VW’s ide.3 has had setback after setback now related to the battery management system’s software.
Car companies are in a recession except for Tesla who is profitable because of their manufacturing capabilities. What car company right now is willing to dump billions into R&D for electric cars much less make it vertically integrated like tesla?
Covid has allowed the lead to grow for Tesla.
Arguably this is the most important part of the thesis and quite possibly one I don’t want to write because they’re having battery day on Sept 22nd where we’ll find out how they’re going to scale *their* production from the 55 gwh to 5+ terrawhatt hours of yearly production.
Here’s what we know:
Tesla’s acquisition of Maxwell & Hibar was quite possibly the most brilliant move made in how much it’s going to transform them. They’ve already had their “road runner” project going for some time and it’s in a building near their Fremont factory. This year they’ve been issued tons of permits to expand production there and have hired over 500 more people to work in that building.
Maxwell Technologies was acquired for their battery technology. They produce “dry” batteries where yo udon’t need huge ovens to cure the battery’s slurry product (reducing Capex and, astonishingly, reducing the battery line’s footprint to something like 1/16th the size).
Hibar was acquired because they’re a leader of battery manufactoring.
I’m keeping this very light and will be filled in after Sept 22nd but I expect:
- EM tweeted out we will see their battery manufacturing lines in their Road runner plant down the street from their Fremont factory
- Battery costs to now be below ICE’s for Tesla
- Expansion of their energy business (which EM has said should be equal to their vehicle revenue)
- Dry annodes
- Tabless anodes to increase density
- Million mile battery where the car can go for more than a million miles without needed a new battery. This is key for a robotaxi
- Path to terawatt hour production in 2 years
- Path to each factory having terawatt hour production in a few years
- Non battery related but updates to Semi, Roadster, and Plaid powertrain for S & X
Here’s what I think Tesla’s end-goal is: to fundamentally be the absolute best battery manufacturer in the world. Bar none.
ENERGY GENERATION & STORAGE
Tesla was recently awarded permits in UK, parts Texas (where energy is de-regulated), and Germany (which has some of the highest energy prices in the 1st world) to become an electricity provider aka a utility company to sell electricity to the grid.
Here’s what I think Tesla’s end-goal is: to allow every household to have their own power plant (solar) on their roof and energy storage (powerwall) in their garage. For larger projects, they’ll build & own themselves their own storage (using the utility-scale megapack) and use the recently-release “auto-bidder” software to help control the grid where all the powerwalls & megapacks are connected to create a “virtual powerplant”. So let’s talk about auto-bidder: it’s a machine learning arbitrage software that connects with all other power walls & megapacks on that grid to help stabilize it. As an example of how profitable this is, the Hornsdale battery is seeing ROI of 20%+ and have expanded their battery to capture more of the Australian energy. This plant has a wind farm attached to it where the auto-bidder detects and sells back electricity when demand is high & stores when low. In Australia, they’re in the final phase of connecting 4,000 different home’s powerwalls to create a virtual grid and do exactly what they’re doing at Hornsdale.
Here’s what I see in the short term: Tesla will slowly sell megapacks to regulated utilities that replace expensive peaker plants. Since the current price today per kWh is lower than peakers, this is now a multi billion dollar market ripe for the taking in just the US (case in point Tesla has recently won multi-gWh of storage projects to replace these peaker plants such as in CA).
When battery costs come down (and solar prices too), I believe we will see Tesla power plants where they’ll have a solar farm charge the batteries and the autobidder software sell electricity back to the grid. If the price for batteries continues to fall as it has been with Maxwell’s chemistries in the pipeline, Tesla could very well become the supplier of the cheapest electricity available in the world. Like I said, this is hard to imagine but I do think Tesla will move into this area in the long-run. Even today, Tesla is starting to replace certain power plants with batteries due to their hardware being cheaper to both build & operate.
Let’s combine this whole ecosystem of technology that Tesla is either on the verge of having or already has:
Due to Tesla’s vertically integrated company where it has both the cheapest & best batteries & cars, I believe Tesla will sell all of its cars very-near-cost in the long-run and only make money through its full self-driving software. This allows electric cars & semi’s to become as affordable as possible (and helping save the planet, one of Tesla’s goals) while upselling very expensive software packages
Tesla will also release & own their own robotaxi’s that have larger battery packs & more efficient motors. Since they also very *very* conveniently an electric utility, there is almost a certainty that these will be the absolute cheapest on the road. Just remember: this market is for both passengers & cargo (through their semi) in both US, Europe, and China… at the minimum.
All technology in the above two paragraphs are available or soon to be available to them.
Their challenge? Scaling. That’s it.
In terms of the car itself, I beg you to please drive one for a week. Once you use it for a week you’ll never go back to a regular car again. You’ll walk away from your car with the car still running on accident or you’ll miss having a “full tank” every morning because you’ve charged the car to full during the night, you’ll miss not having to reach for your keys to unlock the car, etc.
As Joe Rogan says “that god damn Tesla is the future… all other cars are dumb”. Or Steven Crowder “I have changed my mind on teslas… it is pretty damn cool. It is *so* fun to drive”. Just do yourself a favor and drive a Tesla for a week. You’ll realize just how dumb other cars are. People don’t want EV’s, they want Teslas.
I do want to take a moment and talk about GM & F. I alluded to this before but car companies are struggling so hard right now that it would be board room suicide to suggest spending billions on a car that would take profits away from their ICE lineup. VW tried to build their id.3 and it was delayed because of their battery management software (something Tesla does real damn well). As a result, their CEO was fired… that doesn’t inspire too much confidence. Also, many countries such as the UK are outright banning the sale of new ICE cars starting at 2030. How is anyone supposed to compete against Tesla?
I’m going to update this section later but I’m exhausted and don’t want to figure out how to do the excel up here.
Refer to ARK Invests models for now. Their probability-weighted target is $1,400 by end of 2024 (their bull target is 3k). I personally think, doing back of the napkin math, we could see 5 trillion market cap by 2030 (with dilution, that’s roughly 4,500 / share) where Tesla gets a few cents for every mile driven & ~.01 kwh generated.
Today’s AH as of updating this thesis before sending it out is $380 representing ~30% cagr to 2030.
I own a lot shares & deep itm calls but this is explaining why & not . It’s 100% of my net worth and no, you can’t convince me to diversify for at least another 5 years. As a result of having everything on the line, I read industry news, 10-q’s, etc (not just Tesla news) on a daily basis.
I’d be really remise to not say something here about interest rates and long-term investing. All investments are pegged against the risk-free rate of what government bonds are set at. Currently, you can get essentially risk-free 0.65% return from the treasury’s 10 year notes vs. Apple’s ~3% in earnings yield, walmart’s 4.1% earnings yield, etc. Now, you take the probability of success in the cashflows a company makes over the years vs. that risk free rate and you come out to realize… basically anything beats 0.65%. So, people see Tesla’s 0.1% earnings yield and cry out OVERVALUED! However, if we think about how large of an opportunity this is to control both energy & mobility (including just even a small probability of this unfolding which I think I’m showing is probably a larger probability than most people think) and when risk free rates are so low… is it really that overvalued? I’m not so sure.