Photo: @alex_avoigt via X
A lot of things are happening lately at Tesla, with the Full Self-Driving taking center spot thanks to an apparent breakthrough. The software has become so good that Elon Musk might be tempted to bet Tesla's future on autonomous driving and, specifically, the robotaxi he's so pumped about. However, with regulatory hurdles galore and autonomous driving still a distant dream, the robotaxi will not be able to support Tesla's next growth phase.
My father was a dreamer, with many of his fantasies so far-fetched that nobody believed they were even possible. But he had many dreams, and some of them did come true, making him not only a dreamer but an achiever, too. I guess the silver lining to this is to dream a lot and dream big because something will happen if you work toward fulfilling those dreams. If you don't dream, you only follow someone else's dreams.
Although I don't consider Elon Musk a fatherly figure, I can see he is also a man of many dreams. With far more resources than my father had, he turned many of his dreams into reality, despite some being science fiction for many people. Having achieved so much while proving skeptics wrong almost every time must have made him consider himself infallible. There's a big danger in this, as being good at something doesn't automatically make you good at everything.
There is a lot to lose when you base your decisions on the assumption that nobody knows better than you. We've seen from Walter Isaacson's book how Musk manages his companies and how everyone else had a hard time putting some sense into his thinking. Isaacson described the struggles of Tesla executives to convince Musk that the affordable compact EV was a necessity for Tesla, while he only wanted a robotaxi. They finally sold him the idea as the only way the robotaxi development would make sense economically. In October 2022, Musk announced the new plans for the next-generation EVs.
Well, fast forward to today, things are not as straightforward as we thought. A recent report claimed that Musk changed his mind again about the affordable EV model. Again, he will bet Tesla's future on the robotaxi dream. Although Musk denied the report, calling it a lie, we know there is no smoke without a fire. After he denied it, Musk announced that Tesla would unveil its robotaxi on August 8. There's a lot to sink in and a lot to talk about, so here's my take on what's going on and how things might develop from here.
FSD V12.3 marked a notable improvement, with many beta testers being able to drive for days without disengagements. This confirmed that going all in on neural networks was the right move, allowing Tesla to leapfrog the competition. It wasn't completely painless, as the carmaker had to scrap hundreds of thousands of lines of code and rewrite everything. This bold decision paid off, as everyone testing FSD V12.3 discovered.
If you're wondering what was so special about the V12 that allowed it to finally drive a car with almost no intervention, think of ChatGPT. Large language models like it try to predict the next word in a conversation and have become very good at doing this. It's like autocorrect on steroids, but much better than that. Unlike autocorrect, which relies on algorithms to suggest corrections, ChatGPT tries to understand the text and offers predictions based on previous training.
That's an important distinction that translates well to the FSD software. Whereas previous FSD versions up to and including V11 used algorithms to react to fluid conditions around the car, the V12 software tries to understand the world. It may not be perfect right now, but it can learn to understand it better, and there's no ceiling to what it can learn. By asking to predict what the world would do and penalizing mistakes, the FSD software will eventually become ChatGPT-good at driving a car. But what does this mean for us, for everybody?
That will bring important safety benefits if humans agree to let it be in charge of driving. We already know that this would not necessarily be the case. It's not only distrust in technology, it's also the vanity that no one can do things better than ourselves. This can change once people see the benefits, whether on their own or forced by regulations. After all, the most important benefit autonomous driving can bring is much safer driving, resulting in a significant drop in traffic casualties.
That's the rosy part, but that's not why companies push autonomous driving development. The real benefit would be eliminating human drivers altogether. This, however, could lead to social unrest and, ultimately, to the disappearance of the driving profession. Millions of jobs would evaporate as society would not need human drivers anymore. This will not happen overnight, but it will be sudden and brutal enough to cause massive societal disruptions.
So far, there's no other company to be as close to cracking vehicle autonomy as Tesla. There are many contenders, but most of them are still waiting for a miracle to make their algorithm approach work. It's a prime example of the "sunk cost fallacy," where a company invests a lot in an outdated technology and doesn't want to write it off and switch to a new technology. Companies that have spent billions on their algorithm-based self-driving software don't want to start over again with a new technology.
This means that Tesla is close to becoming a monopoly in autonomous driving. Once AI-based autonomous driving systems take off, there's no way others will catch up. If you want a preview of what would happen, it's enough to look at the space sector. SpaceX launches and recovers rockets from orbit while others face the tough decision of closing shop. Competing on costs per launch is unimaginable, and catching up with SpaceX is nearly impossible.
Similarly, competing with Tesla in the autonomous driving space would be a pipe dream. Even if they start from scratch following Tesla's end-to-end AI approach, competing companies would need enormous resources, insane processing power (an unimaginable number of GPUs), and real-world data generated by millions of cars. It would also require a low-power supercomputer in each vehicle, not to mention the best AI talent in the world. If you think this is easy or cheap, just ask Apple.
Musk reiterated earlier this month his desire to license FSD to other companies. More than this, he agreed to a proposal to offer FSD and Autopilot hardware at cost to other companies, with the condition that the car owner would pay for the FSD license instead of the manufacturer. That would be an interesting proposition, allowing Tesla to expand its user base without much effort. It would also enable other carmakers to offer autonomous driving features with a minimum investment.
Elon Musk has said that cracking vehicle autonomy is crucial for Tesla. Probably, this is what he meant by this. Whoever gets there first will have an enormous advantage over its competitors. In other words, self-driving could make or break Tesla, not because it's crucial to offer such features but because it can't afford not to offer them. Consider it like the nuclear weapon of the automotive industry, and you'll see why everyone would want to license the technology once it becomes available.
This is all the more true considering that developing a competing solution is nearly impossible. However, this is not a silver bullet, and having solved autonomy doesn't mean it's game over for every other car company. There are potential roadblocks to adopting autonomous driving technology, and that could also affect Tesla and its business plan.
If certifying FSD as autonomous driving (SAE Level 4+) nationwide in the US seems difficult, achieving the same feat in other markets is nearly impossible. For instance, even supervised FSD is a big no-no in Europe, where regulations mandate that a human driver validate every decision. This basically defeats the purpose of automatic driving systems. Other markets have their own limitations and restrictions, which puts Tesla in a difficult position trying to advance its autonomous driving projects.
China, most notably, would never agree to let Tesla operate driverless cars unless all the data generated in the process remains in China. This will fork the FSD project, forcing Tesla to divert resources to offer similar features in China and the US. China will also want to protect domestic carmakers, making Tesla's position even more difficult. This could go as far as Tesla having to open its software to other carmakers in China just to continue doing business in the country.
This could make FSD a lot less compelling. With many limitations imposed by fragmented regulations across the globe, Tesla FSD will see its value diminished. Other OEMs would not be as eager to pay if it's only of limited use. A nationwide deployment with regulatory approval may mark FSD's most important moment, but this is still years away. This makes the latest information even more bizarre.
Musk has long considered the robotaxi to be Tesla's most important product, potentially worth trillions. He only agreed to produce an affordable EV to cover the costs of the robotaxi thanks to the economy of scale. Another reason was to be able to sell the compact EVs in case vehicle autonomy would not be ready by the time the next-generation platform is ready for production. This makes me believe Musk changed his mind after seeing the FSD breakthrough.
Indeed, if Tesla cracks vehicle autonomy, this means Tesla can produce a robotaxi and be the first to market with such a product. However, this would not mean much beyond the prestige afforded by being the first. That is, if it's the first. Tesla is hardly famous for being the first to market with a product. The Cybertruck is the prime example, unveiled in 2019 as a revolutionary electric pickup only to arrive four years later in an already crowded segment.
Even if Tesla unveils a robotaxi on August 8, as announced, it doesn't mean it will start production immediately. Being able to sell the robotaxi is another thing, and selling it in enough volumes to cover the development costs is another thing altogether. The compact EV had the advantage of a huge market and the possibility to sell immediately across the globe in insanely high volumes. The robotaxi has none of this advantage.
Besides high development costs, the robotaxi has virtually only one client: Tesla itself. It's a paradox, really. If the robotaxi is everything Tesla hopes to be, it makes no sense to sell it to others, individuals or businesses. It will be much more profitable for a Transportation as a Service (TaaS) business case scenario. Why sell the vehicles when it can just collect the money in perpetuity?
However, the robotaxi will not function in the present regulatory environment, not even in the US. This significantly shrinks its totally addressable market and, as such, its production volumes. Nobody would want a car that is not allowed on the roads or only allowed in restricted areas. This will make it impossible for Tesla to recoup the development costs. It's also why I believe that delaying the plans of the compact EV is a huge mistake. The robotaxi will not bring a new growth phase for Tesla and will only generate insane costs.
Although I don't consider Elon Musk a fatherly figure, I can see he is also a man of many dreams. With far more resources than my father had, he turned many of his dreams into reality, despite some being science fiction for many people. Having achieved so much while proving skeptics wrong almost every time must have made him consider himself infallible. There's a big danger in this, as being good at something doesn't automatically make you good at everything.
There is a lot to lose when you base your decisions on the assumption that nobody knows better than you. We've seen from Walter Isaacson's book how Musk manages his companies and how everyone else had a hard time putting some sense into his thinking. Isaacson described the struggles of Tesla executives to convince Musk that the affordable compact EV was a necessity for Tesla, while he only wanted a robotaxi. They finally sold him the idea as the only way the robotaxi development would make sense economically. In October 2022, Musk announced the new plans for the next-generation EVs.
Well, fast forward to today, things are not as straightforward as we thought. A recent report claimed that Musk changed his mind again about the affordable EV model. Again, he will bet Tesla's future on the robotaxi dream. Although Musk denied the report, calling it a lie, we know there is no smoke without a fire. After he denied it, Musk announced that Tesla would unveil its robotaxi on August 8. There's a lot to sink in and a lot to talk about, so here's my take on what's going on and how things might develop from here.
ChatGPT moment: Tesla FSD may have seen a breakthrough
People around Musk tried to convince him to hedge his bet on the robotaxi with a mass-market EV model. This was necessary to avoid a monumental fiasco in case the Full Self-Driving software failed by the time the robotaxi was ready for production. However, important things happened on the FSD front in the past months. The launch of the first end-to-end AI variant of the software brought a marked improvement and the perspective that development would advance much faster from now on.FSD V12.3 marked a notable improvement, with many beta testers being able to drive for days without disengagements. This confirmed that going all in on neural networks was the right move, allowing Tesla to leapfrog the competition. It wasn't completely painless, as the carmaker had to scrap hundreds of thousands of lines of code and rewrite everything. This bold decision paid off, as everyone testing FSD V12.3 discovered.
If you're wondering what was so special about the V12 that allowed it to finally drive a car with almost no intervention, think of ChatGPT. Large language models like it try to predict the next word in a conversation and have become very good at doing this. It's like autocorrect on steroids, but much better than that. Unlike autocorrect, which relies on algorithms to suggest corrections, ChatGPT tries to understand the text and offers predictions based on previous training.
That's an important distinction that translates well to the FSD software. Whereas previous FSD versions up to and including V11 used algorithms to react to fluid conditions around the car, the V12 software tries to understand the world. It may not be perfect right now, but it can learn to understand it better, and there's no ceiling to what it can learn. By asking to predict what the world would do and penalizing mistakes, the FSD software will eventually become ChatGPT-good at driving a car. But what does this mean for us, for everybody?
Social and economic implications
Tesla may have created an autonomous driving system with unlimited capabilities, and this sounds both amazing and scary at the same time. Imagine a driver with eyes in all directions, who is never tired and never gets angry or distracted, let alone intoxicated. Above all, it drives better and much safer than any human could possibly do. Can you imagine that?That will bring important safety benefits if humans agree to let it be in charge of driving. We already know that this would not necessarily be the case. It's not only distrust in technology, it's also the vanity that no one can do things better than ourselves. This can change once people see the benefits, whether on their own or forced by regulations. After all, the most important benefit autonomous driving can bring is much safer driving, resulting in a significant drop in traffic casualties.
That's the rosy part, but that's not why companies push autonomous driving development. The real benefit would be eliminating human drivers altogether. This, however, could lead to social unrest and, ultimately, to the disappearance of the driving profession. Millions of jobs would evaporate as society would not need human drivers anymore. This will not happen overnight, but it will be sudden and brutal enough to cause massive societal disruptions.
So far, there's no other company to be as close to cracking vehicle autonomy as Tesla. There are many contenders, but most of them are still waiting for a miracle to make their algorithm approach work. It's a prime example of the "sunk cost fallacy," where a company invests a lot in an outdated technology and doesn't want to write it off and switch to a new technology. Companies that have spent billions on their algorithm-based self-driving software don't want to start over again with a new technology.
This means that Tesla is close to becoming a monopoly in autonomous driving. Once AI-based autonomous driving systems take off, there's no way others will catch up. If you want a preview of what would happen, it's enough to look at the space sector. SpaceX launches and recovers rockets from orbit while others face the tough decision of closing shop. Competing on costs per launch is unimaginable, and catching up with SpaceX is nearly impossible.
Similarly, competing with Tesla in the autonomous driving space would be a pipe dream. Even if they start from scratch following Tesla's end-to-end AI approach, competing companies would need enormous resources, insane processing power (an unimaginable number of GPUs), and real-world data generated by millions of cars. It would also require a low-power supercomputer in each vehicle, not to mention the best AI talent in the world. If you think this is easy or cheap, just ask Apple.
FSD licensing by other carmakers
This leaves us with the only outcome possible, one that Tesla and Elon Musk have teased for a long time: licensing FSD software to other companies. Musk announced more than once that Tesla is in talks with other companies for a possible partnership, but nothing materialized so far. That will change once FSD proves its worth. Every carmaker would want to offer autonomous driving features to its customers or risk becoming irrelevant.Musk reiterated earlier this month his desire to license FSD to other companies. More than this, he agreed to a proposal to offer FSD and Autopilot hardware at cost to other companies, with the condition that the car owner would pay for the FSD license instead of the manufacturer. That would be an interesting proposition, allowing Tesla to expand its user base without much effort. It would also enable other carmakers to offer autonomous driving features with a minimum investment.
Elon Musk has said that cracking vehicle autonomy is crucial for Tesla. Probably, this is what he meant by this. Whoever gets there first will have an enormous advantage over its competitors. In other words, self-driving could make or break Tesla, not because it's crucial to offer such features but because it can't afford not to offer them. Consider it like the nuclear weapon of the automotive industry, and you'll see why everyone would want to license the technology once it becomes available.
This is all the more true considering that developing a competing solution is nearly impossible. However, this is not a silver bullet, and having solved autonomy doesn't mean it's game over for every other car company. There are potential roadblocks to adopting autonomous driving technology, and that could also affect Tesla and its business plan.
Regulatory hurdles make FSD a lot less compelling
So far, regulations are Tesla's biggest enemy, not only for licensing its autonomous driving technology but also for using it itself. Driverless cars are only allowed for limited testing in certain areas or cities, and each US state has its own take on this matter. Putting everyone on the same page would take years, which is why Tesla currently proposes "supervised FSD." Removing the human driver completely is not yet possible.If certifying FSD as autonomous driving (SAE Level 4+) nationwide in the US seems difficult, achieving the same feat in other markets is nearly impossible. For instance, even supervised FSD is a big no-no in Europe, where regulations mandate that a human driver validate every decision. This basically defeats the purpose of automatic driving systems. Other markets have their own limitations and restrictions, which puts Tesla in a difficult position trying to advance its autonomous driving projects.
China, most notably, would never agree to let Tesla operate driverless cars unless all the data generated in the process remains in China. This will fork the FSD project, forcing Tesla to divert resources to offer similar features in China and the US. China will also want to protect domestic carmakers, making Tesla's position even more difficult. This could go as far as Tesla having to open its software to other carmakers in China just to continue doing business in the country.
This could make FSD a lot less compelling. With many limitations imposed by fragmented regulations across the globe, Tesla FSD will see its value diminished. Other OEMs would not be as eager to pay if it's only of limited use. A nationwide deployment with regulatory approval may mark FSD's most important moment, but this is still years away. This makes the latest information even more bizarre.
Is a robotaxi able to offer Tesla the next growth wave?
On Friday morning, the internet exploded after Reuters published a report claiming Tesla canceled plans for an affordable electric model. Instead, Tesla will focus on bringing a robotaxi to market. The report caused Tesla shares to tank, losing 5% of its value within minutes. Elon Musk intervened to deny the report and called it a lie. However, moments later, Musk basically confessed the report was actually true by announcing that the robotaxi would be unveiled on August 8.Musk has long considered the robotaxi to be Tesla's most important product, potentially worth trillions. He only agreed to produce an affordable EV to cover the costs of the robotaxi thanks to the economy of scale. Another reason was to be able to sell the compact EVs in case vehicle autonomy would not be ready by the time the next-generation platform is ready for production. This makes me believe Musk changed his mind after seeing the FSD breakthrough.
Indeed, if Tesla cracks vehicle autonomy, this means Tesla can produce a robotaxi and be the first to market with such a product. However, this would not mean much beyond the prestige afforded by being the first. That is, if it's the first. Tesla is hardly famous for being the first to market with a product. The Cybertruck is the prime example, unveiled in 2019 as a revolutionary electric pickup only to arrive four years later in an already crowded segment.
Even if Tesla unveils a robotaxi on August 8, as announced, it doesn't mean it will start production immediately. Being able to sell the robotaxi is another thing, and selling it in enough volumes to cover the development costs is another thing altogether. The compact EV had the advantage of a huge market and the possibility to sell immediately across the globe in insanely high volumes. The robotaxi has none of this advantage.
Besides high development costs, the robotaxi has virtually only one client: Tesla itself. It's a paradox, really. If the robotaxi is everything Tesla hopes to be, it makes no sense to sell it to others, individuals or businesses. It will be much more profitable for a Transportation as a Service (TaaS) business case scenario. Why sell the vehicles when it can just collect the money in perpetuity?
However, the robotaxi will not function in the present regulatory environment, not even in the US. This significantly shrinks its totally addressable market and, as such, its production volumes. Nobody would want a car that is not allowed on the roads or only allowed in restricted areas. This will make it impossible for Tesla to recoup the development costs. It's also why I believe that delaying the plans of the compact EV is a huge mistake. The robotaxi will not bring a new growth phase for Tesla and will only generate insane costs.
Google News