Did Tesla FSD Just Have its "ChatGPT Moment"?
Autonomous Driving Breakthrough Provides Significant Upside for Tesla's Share Price
Summary:
Following years of disappointments, Tesla's release of FSD 12.3 marks a pivotal breakthrough, hinting at the eventual realization of full autonomy. Notably, the market has yet to account for this development's potential value.
I estimate that Tesla currently trades around the fair value of its electric vehicle business, assuming moderate success, but failure to ever achieve fully self-driving vehicles.
I believe Tesla is the single best large cap stock to own for an extended AI rally, as the market ultimately recognizes and prices Tesla as an AI leader.
Disclaimer: This writeup is not investment advice and should not be construed as such. I am sharing my research insights and positioning on individual companies for informational and entertainment purposes. In particular, I do not take into account any readers personal financial situation. Please do your own research.
My Love-Hate-Relationship with Tesla
Investors that follow me on X (formerly Twitter) for longer may have seen me comment overwhelmingly negative on Tesla in recent months and years. In fact, I do have a love-hate relationship with Tesla stock. I was invested in Tesla during much of its phenomenal run between 2019-2021. Interestingly, on a price-sales basis, Tesla is currently as “cheap” as it was just before the Covid-19 market crash.
What initially had drawn me towards Tesla was the Model 3. I considered the Model 3 an outstanding electric vehicle, especially for the originally targeted price of $35k. That’s also where I got disappointed by Tesla first, as the Model 3 never really sold for $35k, or at least not for long. The other major reason for me to invest in Tesla was its autonomous driving ambitions. At the “Autonomy Day” in April 2019, Elon Musk basically promised autonomous robotaxis as early as “next year.” Fast forward, five years later, and the situation couldn’t be more disappointing until recently.
In defense of Elon Musk, back in 2019 Tesla was a heavily shorted stock and on the brink of bankruptcy. In 2020 Elon Musk revealed that Tesla was only “about a month” from bankruptcy between mid 2017 to mid 2019. While only die-hard Tesla fans would have truly believed in robotaxis by 2020, I have to admit that I was certainly expecting more from Tesla. Year after year Musk kept telling the market in one way or another that a breakthrough is near. He kept shifting goalposts only to disappoint again and again.
Despite the lack of a truly affordable mass market electric vehicle, and despite any tangible progress with its autonomous driving approach, Tesla’s share price kept ripping higher. For me, Tesla’s valuation by the market vs. real work progress didn’t make much sense, so I passed on the stock for the last few years. As of today (March 24, 2024), Tesla’s share price is down 31% year-to-date and 58% from its all time high in November 2021.
Is Tesla finally an interesting opportunity again now that the valuation is back to early 2020 levels? While I believe that Tesla as an investment is likely to work well from here just based on its electric vehicle business, that alone is not sufficient for me. What I was desperately waiting for in the past years was a revived hope for its autonomous driving approach. IF Tesla could win here, it would not only truly transform every of its EVs to an “iPhone of Cars,” no such a technology breakthrough will allow Tesla to redefine transportation. A world of robotaxis is a world that does not need Uber, but would increase the addressable market for Uber-type services by multiples times.
In sum, for me the Tesla investment case stands and falls with whether or not Tesla truly achieves fully self driving vehicles.
Tesla's transition from a coding-based approach to a neural network approach for its Full Self-Driving (FSD)
In May 2023, Tesla announced the transition from a coding-based approach to a neural network approach for its Full Self-Driving (FSD) system. The introduction of FSD Beta version 12 (v12) marked a move towards an end-to-end AI system. This changed represented a significant shift in how the company approaches autonomous driving technology.
Previously, Tesla's FSD utilized neural networks primarily for perception tasks, such as detecting the environment around the vehicle. Decision-making and driving inputs (steering, acceleration, braking, etc.) relied on traditional programming. Elon Musk had initially stated that AI wouldn't be needed for these aspects of driving. However, the plan changed to include an "end-to-end AI" approach, where the FSD system would handle everything from visual input to executing driving maneuvers directly.
This transition involves moving away from a "rules-based" strategy, where specific programmed rules dictated the vehicle's responses to various situations. Instead, Tesla aimed to use a "neural network planner" that leverages billions of video frames from real-world driving scenarios. This data trains the neural network to mimic the decisions made by skilled human drivers in complex driving situations, akin to the way a chatbot like ChatGPT operates, but applied to driving. This shift was described as recent at the time, with the neural network having analyzed 10 million video clips by early 2023 to train on behaviors akin to those of a "five-star Uber driver".
The modular approach used before involved separate systems for perception and planning, with a mix of deep learning models and traditional algorithms for trajectory planning and decision-making. The new approach aims to integrate these components into a unified deep learning system, streamlining the entire process from perception to action without manual rule setting or algorithmic decision trees.
This ambitious pivot reflects Tesla's belief in the potential of AI and neural networks to achieve more nuanced and adaptable autonomous driving capabilities, by effectively learning from vast amounts of real driving data. The goal is to allow the vehicle to understand and navigate complex driving scenarios more like a human driver, using learned behaviors rather than rigidly programmed instructions. This change is expected to enhance the performance and reliability of Tesla's FSD system, pushing it closer to the goal of full autonomy. Now with the release of FSD 12.3, Tesla may have achieved a lot of credibility for this ambition.
Why FSD 12.3 Changed my Mind on Tesla
Frankly, after years of Elon Musk overpromising and underdelivering in terms of FSD I had lost my hope that Tesla would ever be able to achieve truly autonomous driving. Initially, the change to the neural network approach in May 2023 only confirmed my stance. In effect, I interpreted the news that Tesla admitted that its previous approach did not work. Ok, the new approach may be more promising eventually, but it seemed that at least for now, Tesla would need at least another few years until we would have tangible evidence that the new approach works better and has a shot at leading Tesla to achieving truly fully autonomous driving.
On March 16, Tesla began rolling out FSD 12.3 to thousands of customers and my X timeline was full of enthusiastic reviews that took me by surprise, given the lackluster reception of any prior updates since May 2023. The big picture is that FSD 12.3 is showing significant improvements in handling complex driving scenarios. In particular, there are countless tweets of Tesla drivers sharing how FSD 12.3 suddenly masters all the scenarios that previously required interventions by the driver:
What I found notable is that numerous tweets are coming not from diehard Tesla fans, but from former sceptics that had exposure to the prior technology and observed a step change in quality. Phil Duan, who hints at a paradigm change in his response to the tweet below, is a senior staff software engineer at Tesla AI, where he leads the Fleet Learning team at Autopilot.
Moreover, it is hard to miss how much more humanlike FSD 12.3 drives relative to its predecessors, as evidenced by the tweet below from Michael Dell (CEO of Dell Technologies):
As if a step-change in improvement in a v3 version isn’t enough, is is notable how much more confidence is being expressed by Elon Musk and other Tesla staff members. I noted how Elon Musk became rather reserved over FSD in the recent years after having overpromised before. Now, he is hinting at significant improvements every two weeks, a clear acceleration in the rate of progress:
It is actually this acceleration in the rate of improvement that has changed my mind about Tesla. Within less than 12 months, Tesla’s neural net approach has exceeded in quality the old approach of hand-coding autonomous driving. Below is a Tweet by Ashok Elluswamy, who explains that the Tesla team believes that v12 has now net surpassed the previous v11 stack in quality.
In essence, Tesla’s new approach exceeded a decade of experience manifested in 300k lines of CC+ code within less than a year. This achievement is largely attributed to Tesla's use of a massive amount of real-world driving data collected from its fleet of consumer-owned vehicles. This data advantage enables Tesla to continuously train and refine its neural network-based FSD system on a diverse array of scenarios and conditions. Google’s Waymo, on the other hand, uses a combination of high-definition maps and sensor data. While Waymo's technology is highly advanced, Tesla's approach, if successful, can scale more rapidly and improve as it gathers more data from its extensive and growing fleet, providing a wide range of driving situations for machine learning models to learn from.
I believe in retrospect, Tesla’s release of FSD 12.3 on March 18, 2024 will be seen as Tesla's "ChatGPT moment" in the autonomous vehicle space. In the same way that ChatGPT marked a leap towards advanced conversational AI, showing potential that many didn't fully anticipate, Tesla's FSD 12.3 has had a similar groundbreaking effect on perceptions of autonomous driving. It hasn't brought full autonomy just yet, but its significant advancements suggest that the goal might not be as distant as once thought. Much like how ChatGPT pulled forward the timeline for AGI, FSD 12.3 has rekindled faith that Tesla could indeed deliver full self-driving, and perhaps sooner than we imagined.
Valuation
Tesla currently has a market valuation of $550b. I believe that Tesla’s electric vehicle business alone could be worth twice this amount in 10 years (by the end of 2033). This would represent an annual return of 7.2% and thus imply that Tesla’s current market capitalization is about fair, if we assume no value for its autonomous driving technology, its energy business, or even Optimus.
In a nutshell, here are my assumptions that would value Tesla’s electric vehicle business at the required $1100b by 2033:
8 million cars sold per year at an average price of $46k would result in revenues of $368b.
In 2023, Tesla sold 1.8 million electric vehicles at an average price of $45k. I am thus assuming a 16% CAGR in vehicles sold, far below Tesla’s historical growth rate of about 50%. While 2024 will be a tough year for Tesla, and revenue may in fact only come in at 15% growth, I believe Tesla’s growth rate will recover in 2025 and 2026 from lower interest rates. Moreover, Tesla will reveal a new $25k EV (model 2?) in 2026. I believe such a truly mass market model would sell very well and will eventually make my assumption of 8 million cars sold per year by 2033 look very conservative. While this new model 2 would pull the average sales price down, it is important to consider that Tesla is currently ramping the cyber truck, will likely release the roadster 2, and there will be 10 years of 2-3% inflation, which would push today’s $46k to a nominal $58k by 2033.
15% net margins would lead to net income of $55b in 2033, valued at a P/E ratio of 20x, Tesla’s electric vehicle business would have a future value of $1.100.
Tesla had net margins exceeding 15% in 2022 and 2023. That was achieved despite its investments into FSD, Optimus, as well as its other services business. Tesla invests increasingly in automation and at 8 million cars sold per year in 2033, it would be much closer to operations “at scale” than today. I thus believe that Tesla can achieve 15% net margins on its EV business in the long term, even if margins in 2024 are temporarily depressed.
Not long ago, Tesla has stated the ambition to sell 20 million cars a year annually. I am currently less optimistic about the pace of EV adaption. Nevertheless, I think 10 million EVs is currently a realistic goal to achieve for Tesla over the long run.
In sum, I believe the assumptions outlined above are pretty achievable for Tesla. If correct, the market is currently not very demanding of Tesla. In contrast, the market would value Tesla only for its potential as an EV maker, attributing zero value to FSD, the energy business or Optimus.
What could the FSD Optionality be Worth?
If Tesla were to achieve true FSD, I believe we are speaking about a business worth trillions of USD eventually. At some point around 2035, there should be a total of 100m Teslas on the roads. If we assume 30% of Tesla drivers pay only $200 a month, that would lead to annual FSD revenues of $72b. Since the hardware and software for FSD is already in place of every Tesla build today, the margins of this business would be extremely high for Tesla. At a net margin of 60%, annual earnings would equate $43b and it would take a PE ratio of only 23 to value FSD at $1 trillion by then.
If Tesla really achieves FSD, I believe all these assumptions will turn out ridiculously conservative:
Subscription Pricing: With FSD, Tesla could likely charge a monthly subscription fee significantly higher than $200. Eliminating the need for a driver's license and reducing accidents would not only lower insurance rates but also add immense value for consumers. This enhanced value proposition could justify a higher price point for FSD subscriptions, reflecting the savings and convenience offered to the users.
Increased Sales: The appeal of FSD could be expected to boost Tesla's vehicle sales substantially. As consumers recognize the benefits of autonomous driving—ranging from safety to convenience—demand for Tesla vehicles equipped with FSD technology would likely surge. This increase in demand could lead to higher production volumes and sales, further escalating Tesla's market presence and revenue potential.
Robotaxi Network: Tesla has long discussed the potential of turning its vehicles into a fleet of robotaxis. This would allow car owners to earn money by lending their vehicle to a ride-sharing network while not in use. This could drastically increase utilization rates of each vehicle, contributing significantly to revenues beyond the direct subscription model for FSD.
Autonomous Trucking: The freight and logistics sector could be revolutionized by autonomous driving technology. Tesla's entry into this market with the Tesla Semi could be transformed by FSD, reducing the cost of logistics, improving safety, and increasing efficiency in supply chains.
Licensing FSD Software: Licensing the FSD software to other car manufacturers or mobility service providers could open up additional revenue streams. This could also accelerate the adoption of autonomous driving technology across the industry.
Energy and Charging Services: Autonomous vehicles would increase the demand for electric charging infrastructure. Tesla could leverage this by expanding its network of Superchargers and offering energy services tailored to autonomous fleets, including robotaxis and trucks.
Insurance Products: Tesla has already started offering insurance products to its customers, emphasizing the safety features of its vehicles. With FSD, Tesla could significantly lower the risk of accidents, further reducing insurance premiums and potentially capturing a larger share of the insurance market.
As outlined, my initial calculations might turn out highly conservative. The true value of Tesla's FSD could extend well beyond direct revenue from subscriptions, encompassing a broad ecosystem of services and technologies that redefine transportation, logistics, and urban mobility.
Timeline & A Potential AI Bubble
Now of course you may say, but this is all far in the future, if it ever happens. And frankly, even today I’d rate the chances of Tesla achieving FSD at little more than 50% and if so, it will likely take many, many years until enough regulatory hurdles are surpassed for widespread adoption. However, the market is famous for pricing the future, so Tesla shareholders may not have to wait that long until a long phase of bagholding ends.
Will AI be the Bubble of all Bubbles?
Throughout the history of financial markets, numerous bubbles have emerged, driven by speculative fervor around technologies or sectors believed to transform the future, from the Tulip Mania of the 17th century to the Dot-com Bubble of the late 1990s. Each of these bubbles was characterized by inflated valuations and widespread investor enthusiasm for emerging industries, often leading to dramatic busts when expectations failed to materialize as quickly or as grandly as anticipated.
However, I believe that the burgeoning field of artificial intelligence harbors the potential to eclipse all previous bubbles in both scale and impact. AI represents a foundational shift not merely in technology, but in every facet of human life and industry. It promises to redefine the way we work, communicate, and solve complex challenges, making it a sector with seemingly limitless potential. As AI technologies, including machine learning, robotics, and autonomous systems, continue to advance at a rapid pace, they will likely attract immense investment and speculation.
Given AI's capacity to drive a new industrial revolution, the speculative bubble that forms around this sector could dwarf those of the past, fueled by genuine transformations across global economies and societies. This makes the AI bubble unique—not only because of its potential size but because, at its core, it is underpinned by technologies that could indeed redefine human civilization.
It is of course not clear if we are in the early innings or potentially already at the end of an AI bubble. I have asked my followers on X about this. 50% are convinced that nothing will prevent a massive AI mania / bubble, and I agree with that majority, I think its still very early days and I am thus trying to increase my portfolio exposure where it appears to makes sense.
Would an AI Bubble Really Keep Leaving Tesla Aside?
So far we have seen the Nvidia, Meta, and Microsoft share prices advancing like there is no tomorrow. Tesla, however, has been left out. It has even dropped out of the “Magic 7” due to its recent share price weakness. Would any AI bubble really occur without Tesla? I believe the answer is clearly no, IF my perception around FSD 12.3 is correct.
In the context of an emerging AI bubble, where the market begins to price in the significant optionality and transformative potential of AI technologies including FSD, Tesla's share price could experience unprecedented growth. The anticipation of true Full Self-Driving capability would not only revolutionize Tesla's business model but could also serve as a catalyst for re-evaluating the company's market valuation. If the market increasingly recognizes the vast array of revenue streams that FSD technology can unlock—ranging from direct subscription services to licensing, autonomous trucking, and energy and charging services—Tesla's financial projections could soar well beyond current estimates.
Moreover, as the AI sector heats up, investors seeking to capitalize on the next big wave of technological innovation might view Tesla not just as an automotive or energy company, but as a leading AI technology firm. For example, in addition to the transformative potential of Full Self-Driving (FSD) technology, Tesla's development of the Optimus robot could further amplify the company's valuation through a halo effect.
The FSD optionality, in this scenario, represents a combination of Tesla's potential to lead in autonomous mobility, its role in shaping future transportation infrastructure, and its capacity to deliver groundbreaking AI advancements. These factors, combined with a speculative market environment that values growth and innovation above current earnings, could lead to Tesla's share price reaching heights that far exceed traditional valuation models.
To be clear, I am not intending to pump / push / or advocate for a bubble valuation of Tesla. Based on my analysis, I truly believe that the market is currently mispricing Tesla’s chances to achieve fully autonomous driving and the implications it would have. I believe Tesla may perhaps double from the current valuation to truly reflect the current stage of its AI potential. However, chances are we are in the early innings of an AI bubble and I don’t think that Tesla’s optionality would stop being priced at a perfectly rational value. Anticipating this development may be rational in an odd way. Missing the top performers in a stock market bubble can be costly in terms of underperformance to a benchmark. I have unfortunately missed much of the rise in Microsoft and Meta and I have totally missed Nvidia. Rather than jumping on these stocks now, I believe the Tesla is currently by far the best stock to position for an extended AI rally, especially if it keeps making FSD progress in the months and years to come.
Disclosure: I am long Tesla with a 5% allocation in my portfolio.
From my 2024 Q2 Portfolio Update:
I sold Tesla right after the Reuters news that Tesla canceled the $25k car (Model 2). Later Elon Musk accused Reuters of lying without specifying details what part exactly was the lie. The consensus among Tesla investors appears to be that the Model 2 is indeed at least postponed and that Tesla for now focuses on the Robotaxi. Elon Musk later announced on X that the Robotaxi will be revealed on August 8. In sum, the Model 2 seems to be at the very least postponed for now.
Unfortunately canceling the Model 2, combined with the surprisingly very bad 2024 Q1 deliveries proved some key assumptions about my Tesla investment case wrong. I currently don’t see any possibility that the car business alone may be worth 2x today’s Tesla market cap anymore as I have assumed in my writeup. Solving FSD is now a condition for Tesla to provide superior returns. While I remain optimistic that Tesla can achieve FSD, it is still quite uncertain. I would place the odds at around 50%. At a $500b market cap, the upside of solving FSD is also not unlimited (I would argue 10x at best). Overall, Tesla is now clearly less attractive. I put Tesla back on my watchlist, but for now other opportunities provide better risk return.