Why Elon Musk's Latest Venture is Likely to Fall Short
12-minute read
This is a linkpost; original at my Substack
Tesla Motors is going to enter the robotaxi business, and thereby increase the firm’s stock market valuation from around $740 billion to several trillions of dollars—at least according to Elon Musk. As incredible as Musk’s claim may sound, investors have learned the hard way not to bet against him. Then again, Musk falls short just as often as he succeeds.
When it comes to robotaxis, though, the former is more likely to be what happens. Why am I so confident? Because we know what it takes to succeed in that market, and Tesla doesn't have it. Tesla lacks both the necessary time, and the necessary culture.
Tesla and the Robotaxi Market
Robotaxis are still a niche service, available only in a few cities worldwide, and those cities are clustered in two countries: the United States and China. In the USA, the major players are Waymo, Cruise, and Zoox. In China, they are Pony.AI, WeRide, AutoX, and Apollo. In both countries, robotaxis are novelties, only available in sharply demarcated service zones. But they are present and offering rides for pay. And the size of their zones and the number of cities they service is growing. Significant disruption of the ride-hail market is still years away, but the potential for change to the market is significant.
Tesla, of course, is a disruptor itself. It’s not an exaggeration to say that the firm is the responsible party for the US market share of electric vehicles as a percentage of total vehicles sold going from zero in 2010 to ~7% in 2023. It’s a remarkable achievement, which the firm earned through excellent design, an unrivalled network of charging stations, and a sophisticated automated driver-assist system.
Building that automated driver-assist system (ADAS) is what makes Tesla think they can conquer the robotaxi market as well. Many high-end cars have ADAS. Mercedes, for instance, refers to their ADAS as “Drive Pilot”. Tesla’s ADAS is among the best, and is so good that the firm refers to it as “Full Self Driving.” The name is misleading, as Tesla cars are not ‘fully’ self-driving. During ADAS engagement they require a human to be present and actively monitoring and overseeing the ADAS. In response to mounting criticism of this misleading advertising, earlier this year Tesla quietly changed the name to “Supervised Full Self Driving” (FSD), which is less deceptive and more self-contradictory.
Could Tesla leverage its driver-assist systems to dominate the robotaxi market as it has the EV market?
Nonetheless, the quality of Tesla’s FSD is an advantage, and the firm has suggested for years that it aspires to develop it into what the name actually suggests: a true automated driving system that can operate a vehicle without a human driver present. Could Tesla leverage its driver-assist systems to dominate the robotaxi market as it has the EV market?
In April 2019, Tesla’s CEO, Elon Musk, as part of the firm’s ‘autonomy day’, announced that “there will be autonomous robotaxis from Tesla next year”. That prediction did not come to pass, but the ambition behind it was never renounced. Five years later, in April 2024, Musk announced “Tesla Robotaxi unveil on 8/8”. Three months after that, he explained that the launch date was slipping to October in order to accommodate an “important design change”.
Let’s do Musk the courtesy of assuming he means what he says and Tesla does aspire to dominate the robotaxi space. There are two factors militating against Tesla's success in this arena. They both stem from the company's fundamental approach to business and technology.
Tesla Is Late to the Party
Firstly, Tesla lacks the time.
To date, the successful robotaxi players have all demonstrated a willingness to play the long game. The most successful, Waymo, has roots going back as far as 2009, when Google privately initiated its self-driving car project, announcing it publicly the following year. The project was spun off as Waymo in 2016, but its new status as a standalone company did not lead it to rush into things; it would be another two years before it began offering revenue service, and only in a single, small market in Arizona. As of earlier this year, Waymo is offering full revenue service in a reasonably-sized market, namely San Francisco, as well as more limited offerings in Los Angeles and other, smaller U.S. markets. It took Waymo roughly 15 years to go from inception to meaningful revenue service.
Cruise has a similar story. Founded in 2013, it was, as of 2023, also operating revenue service in San Francisco, although at a much smaller scale—before a combination of bad luck and bad choices dealt the firm a substantial setback. Zoox, the US robotaxi company with the least notoriety, was founded in 2014 and only began revenue service earlier this year, in June 2024.
A robotaxi firm is still a firm and can’t be built overnight.
Waymo, Cruise, and Zoox were iterating their robotaxi design and service for between ten and fifteen years before putting up anything that one might describe as a serious operation. Even after assuming that Tesla can take advantage of learning curves, and can go from launch to implementation in half that time, it’s still reasonable to expect at least a five-year delay before revenue service.
Tesla might get a leg up in that it is already in the business of building cars and ADAS software. However, it’s worth remembering that Cruise has been, for part of its life, an arm of General Motors. And being part of a large carmaker didn’t allow it to develop much more quickly.
A robotaxi firm is still a firm and can’t be built overnight. It must write and iterate dispatch software. It must buy and build garages and storage yards. It must hire and train staff: maintenance workers, customer service agents, and lobbyists (more on that in a moment). This all takes time.
It will also take time to perfect FSD. It must go from being called “full self driving” to actually being fully self-driving. The system is currently nowhere near ready. By the firm’s own reckoning, Tesla vehicles with an engaged Autopilot (the firm’s trade name for its basic ADAS package) go ~7 million miles before an incident. That’s impressive: roughly a single order of magnitude better than the average human driver. Unfortunately, the statistic tells us nothing about how good Tesla’s software is at driving alone: that 7-million-mile statistic is for a hybrid system, one where there’s a human being behind the wheel, ready to take over when the ADAS encounters a situation it can’t handle.
A system that doesn’t rely on a human fallback driver is a different problem entirely, and solving it will require Tesla to spend much more time to improve its software before it can launch robotaxis.
“Tesla robotaxi”, Midjourney
Or will it? Tesla’s case is that it is pursuing self-driving technology in a dramatically different fashion than its rivals. Waymo, Cruise, and Zoox started building their self-driving systems years ago, relying on the traditional method of programming a complex set of if-then statements; testing; patching the code; and repeating. Tesla aims to go a different route, taking advantage of the revolution in machine learning currently underway. Their approach is to take the extensive library of data provided by the sensors in Tesla vehicles and use their “Dojo” supercomputer to build a self-driving neural network, in a similar fashion to how OpenAI used Internet data to build ChatGPT. They also think that relying on cameras alone, rather than expensive lidar rigs, will make Tesla robotaxis just as capable as humans or other firms’ driving software, while making the hardware a fraction of the cost.
Can Tesla do these things successfully? Is their approach more likely to bear fruit than the ‘traditional’ method of its rivals? That will be the subject of a future post. For now, I’ll merely observe that innovating in the software and the hardware at once is possible, and even likely, given the resources Tesla has. What is not likely, I think, is the ability to pull off both innovations successfully and quickly. And doing it quickly is imperative.
Wolves at the Door
Waymo, Cruise, and Zoox have taken the time to build out their software and hardware. Tesla can't.
Why not? Because, as Noah Smith points out, Tesla deliveries are down on a year-over-year basis, and the firm has laid off 10% of its workforce. This is a company that needs to focus on its core business of making personal vehicles. That's not just me saying that; it's the stockholders. At least, that’s how I interpret the fact that Tesla has lost $400B USD in market capitalization since 2023.
We know what it takes for a company to take big bets that take a long time to pay off. They need one of two things. One is wide margins, and the expectation those wide margins will persist; so, functionally, a monopoly. That’s Waymo, an arm of the Google search-engine monopoly, which continues to receive bountiful support from its parent; and Zoox, which Amazon bought for more than $1 billion in 2020. The second is patient investors who are comfortable with very long-term returns on investment. That’s Cruise, which has continued to receive support from General Motors, even despite the org’s difficulties.
Tesla doesn’t have these advantages. Investors are skittish, and, as Smith points out, the company has no monopoly. Chinese competition is poised to take the bulk of the global EV market with their super-cheap vehicles. Tesla needs to compete against that; its decision to prioritize robotaxis over a cheap "Model 2" EV would seem to be an admission that they think they can't. And if indeed they cannot compete against Chinese opponents, then Tesla can't afford the gestation time that robotaxis require.
Move Slow and Break Nothing
That lack of time is one reason Tesla won't succeed at robotaxi. Its culture is another, and it’s also the reason I hope Tesla backs away from this path.
To date, Waymo and Zoox have moved slowly, offering only limited service areas, growing them incrementally, and slowing things down when issues arise. It's clear that this is the right strategy. The proof is that when Cruise abandoned this approach, it paid a steep price. By pushing back against regulators when one of its vehicles inflicted severe injuries on a pedestrian, the company invited massive backlash that led to a months-long pause in operations, the reintroduction of human safety operators, and the resignation of the CEO. Only now, with new leadership, and a very cautious return to operations, is Cruise making a comeback.
To the extent that robotaxi rollout is working, it's because the companies involved have moved slowly, cautiously, and with humility.
I won't belabour the point, beyond saying that "slow", "cautious", and "humble" are not adjectives I associate with Tesla or its leader.
Beyond this, robotaxis are, and will remain, a tightly-regulated sector. Across California, local governments have sought to ban self-driving cars from their streets, forcing the state government (through its agencies) to overrule them and reserve the regulation of self-driving technology to itself. When Cruise indulged in bad behaviour, it was those agencies who revoked its license to operate statewide. Individual regulators have immense power over the robotaxi business, which is why having strong public-policy talent on board is so important. The first thing those lobbyists would say, I’d imagine, is not to antagonize the regulator.
Tesla’s CEO, of course, has conspicuously antagonized regulators and elected officials in California and the US federal government (not to mention the UK, Ukraine, and Taiwan). None of this inspires confidence that Tesla will be able to sustain good relations with any regulator long enough for its robotaxi business to succeed.
Tesla and the WALTER Theory
So Tesla will not succeed in the robotaxi market. In fact, its aspirations to do so threaten the entirety of the industry. Hasty, clumsy moves here will redound not only to the discredit of Tesla, but also to the other actors in the field. Transport investor Reilly Brennan describes this situation as the WALTER theory of automated vehicle sentiment: “Automated vehicles are still emerging and the media and public often lump them together — a symbol of the fact that the market currently has no outright leader. Each company’s success is individual, but their failures are often clumped together as if all operators were one company. This is wrong, of course, but sentiment is subjective.
“These companies Win Alone, Lose Together (W.A.L.T.E.R.)”.
By inadvertently injuring a pedestrian in a fashion that was only partially its fault, and then frustrating regulators about it in a fashion that was absolutely its fault, Cruise brought the whole self-driving enterprise into disrepute, with particular backlash in San Francisco. That backlash wasn’t aimed at Cruise, which was no longer operating there: it was aimed at Waymo.
Robotaxis have immense potential to improve the lives of city-dwellers by reducing travel costs, clearing the market for trips, solving the first/last mile problem in transit, and even changing land-use patterns. (Those benefits are beyond my scope here, but you will be able to read all about them in the book I’m co-authoring, The End of Driving, second edition, coming next year from Elsevier). But this potential will be unrealized if residents, and governments, come to believe that robotaxis are fake, or reckless, or dangerous.
That's why I hope Tesla changes its mind, and backs off from the idea of entering the robotaxi market. Conspicuous failure here won't just harm the company. It will harm the introduction of automated driving as a whole. That will make all of us worse off.
Respect to Rob Tracinski, Rob L’Hereaux, and Kevin Kohler for feedback on earlier drafts.