The logic for replacing current driven rides seems reasonable. However, thereafter - especially in crowded non US cities - the possibility of replacing mass transit is going to be limited by road capacity. There is simply no room on the road to move thousands of passengers off mass transit and into self driving cars, even if they are smaller and faster than current human driven cars. Traffic will always grow to fill the roads available and that will happen long before they can out compete mass transit in cities.
Another advantage to autonomous vehicles will be their ability to communicate with each and form convoys on the interstate and work with signals and traffic controls in cities. This wil allow more cars to be on the road---especially considering traffic is often caused by the human element in combination with congestion.
Self driving cars would not be able to replace the Tokyo subways and trains, and would be hard to compete with Japanese taxis which are numerous, clean and reliable, although expensive. Some road space can be recovered by eliminating on street parking, self driving cars won’t park except at dedicated recharging garages.
Self driving buses can eliminate some cars, and deal with the current shortage of bus and taxi drivers in Japan (a reason taxis are expensive there). One company Tier4 is already providing autonomous buses service in a small city in Japan.
1. You don't even consider the Chinese self driving operators such as Baidu who is testing with ±400 vehicles. They seem behind Waymo but not entirely out of the calculus, same may be true for Yandex.
2. You take Musk's promises seriously. His track record, Tesla's cult culture, the fact they have not test driven autonomous miles at any scale and so many other 'small issues' should make one very careful believing that the structural problems in Tesla (no LIDAR, the plan to roll out FSD without careful training, culture of risk taking and poor safety record and much more) would somehow be wished away into a globally scaled operation.
3. You take calculations from ARK invest seriously... Hard to know where to begin here but they are the most unserious 'investment fund' I have ever seen and none of their models stands external reviews well.
4. For Waymo, there is no dependency on Uber at almost any point beyond the launch periods and there is a reason they do not need Uber in San Francisco where they now have large enough user base. The opposite is not true as most car rides are done in a single local and if Waymo is cheaper, has better car experience and is available - why would a user not prefer it? In November I rode both in PHX & LA and my preference for Waymo was very strong, not to mention it was half priced when compared with like for like cars (Waymos are a premium Uber flavour).
I find it hard to find any reason why any of these manufacturers has any particular hold or agenda over AV, now that Waymo proved its a setup that can be fitted to their cars. These companies have over capacity (not only in China) and would do a lot to satisfy a client like Waymo who looks like the future.
I am quite surprised by this analysis quality, I much appreciate most of your posts but this one is not well enough thought through (and AVs are the future, but it would be Waymo's model and others like it delivering it in the next years - I doubt Tesla would even be a player).
Musk’s track record in industrial products is one of delivering everything, albeit late, isn’t it?
Thanks for the point on ARK. I did share my back of the envelope a couple of times. I think they’re probably right in their orders of magnitude, if not the specifics.
Interesting re waymo!
I’m not sure what specifically you disagree with. I agree that Waymo is a contender. I’m just saying that the modular approach is inferior to the integrated one right now! But you’re right that if walk holds all the sway it can bend the arms of other players. Maybe that will be the case!
I'm looking forward to self driving taxis. I'm a 75 year old driving a F150. I plan on driving it into the ground, as they say. By then I'll be 90 and many will not want me driving then.
The technological changes in my lifetime have been astounding. I grew up with partyline telephones and black and white tv. I even have a computer in my hand for typing this message.
The Telsa sales graph is misleading (by now anyway). The numbers for 2024 definitely no longer rise exponentially. 2025 is looking to be way worse. Unless either Elon or Tesla's shareholders wake up and notice that the rose they once held now is brown and stinky, their prospect for winning the robocar game aren't that great.
Have you actually owned a Tesla? I have. Have you watched a channel like AI Drivr? You should. The real world of driving with this software will not convince people of its safety for a while if at all.
And this is from someone who would like it to exist.
The stakes are too high for it to be better than human drivers ‘on paper’.
My Tesla would very occasionally go into a phantom braking problem when passing vehicles in the lane to the left of it that very nearly caused a rear ending at 70mph on several occasions.
Tesla point blank refused to acknowledge this problem but quietly fixed it for a while on update til a later update undid the fix. Not a trust building exercise.
And that’s just boring old lane and speed control on a straight highway.
Correct me if I’m wrong and you’re a genuine enthusiast but these sound like the usual accumulated Tesla talking points.
Kevin Kelly has made the correct (imo) observation that for mass adoption - to justify Tesla before Musk turned himself into an unavoidable liability - public roads will need sensors at junctions and corner cases which will take a lot of public money.
The real problem with all that? Nobody cares as much as the likes of us early adopting enthusiasts. Well not in big enough numbers. It’s not a big enough problem to solve for most of the population right now or in the medium term.
Certainly reliable highway cruise control - def. Robotaxis - very limited local grids only
Heard it all before for nearly a decade - it’s waaay harder than it appears to get the last few percent of corner cases solved satisfactorily.
And that matters because that’s where fatalities occur at speed.
If you’ve driven one for any length of time you’ll see why the average driving person will not trust it with anything complex.
Waymo has the partial answer: drive inside regularly updated maps at fairly low speeds. And don’t over promise.
Tesla do amazing stuff but they are praying for a miracle to justify their valuation when they know there’s probably no answer without public infrastructure like sensors etc
And the buying public just don’t care. What are the numbers on take up?
After the 2024 free trial of FSD, only 2% took up the package. Keep in mind this is meant to be the central argument for the Tesla valuation.
Tesla reduced the FSD purchase price from $12,000 to $8,000 and lowered the monthly subscription fee from $200 to $99. Despite these adjustments, the uptake rate remained modest.
I had that exact problem with our Teslas, even down to the fix and later regression again, but only when using the Autopilot feature. Never when using FSD. FSD has its own, different inconsistencies still. But I drive with it much more than without it! It's safer and less stresful and leaves me less drained upon arriving.
So, which version of FSD did you have? According to the chart in the article the current version 13 has many times fewer driver interventions than version 12. I don’t know their actual safety rate compared to human drivers is, but there data shows fewer crashes per miles driven on Autopilot compared to the federal NTHB rate for all drivers.
Tesla doesn’t seem to have a plan for how to handle situations where cars get stuck (like the Waymo that was circling a roundabout for 15 minutes and the rider couldn’t stop it to get out). Waymo has rooms full of back office support drivers that can take over the driving so passengers aren’t in danger in these situations. I don’t know how Tesla plans to handle that, ask passengers to post a help message on X? Does the CyberCab even have remote operation capability?
I’m also worried that Musks bias against LIDAR is a mistake. He claims humans can drive well enough with by judging distance using only our eyesight, but humans mostly have two eyes for stereo depth perception, and we can move our heads to better visually inspect the situation. Even if Tesla cars have multiple cameras, they are fixed in a particular location and as far as I know are not stereoscopic. And the one camera at the top of the windshield seems susceptible to eventually loosing sight to a well placed bug strike during freeway driving, resulting in possible disaster.
I’m not sure of the version - and I sold it last week. But it was the full version and updated to the latest automatically.
It’s a solution looking for a problem with so many edge case issues out in the wild as you point out.
For example - imagine I’m driving as a passenger on a highway in a Robotaxi with no steering wheel. The overnight cool air has created condensation on the inside of the camera sensors on both pillars as the morning sun heats it up (happened frequently on mine) - the car is not completely blind but it sure has blind spots. Does it just pull over and wait for an hour for it to clear?
Is it teleoperated til it clears? If so how many virtual drivers do you need to reliably run a fleet for this and thousands of other edge cases?
An advanced cruise control is a huge advantage and Tesla is arguably the best at that. I’ll believe it when I see it for anything more.
Speaking of tanking in Europe, there may be a need for autonomous tanks (and drones) depending on what happens in the Russia/Ukraine war. Can Volkswagen make those, and didn’t they also have a problematic founder?
You may get into some of these ideas in your additional work, but I would submit that full-sized cars are more of a legacy than the pinnacle of optimized design for the world you envision. Even two seats is twice as many as needed for the majority of car rides, so something more like an enclosed motorcycle/tricycle with enough storage space for a backpack or side pack would be the ideal tool for the job in most cases. Like ordering an Uber, you could specify when you need a vehicle that seats more than 1 person and request a ride for 2 or 4 or more people. But a benefit of having so many smaller/narrower vehicles, particularly when these vehicles are largely or completely able to communicate with the others around them, is that you can fit far more vehicles on the road, such as having two one-person vehicles side-by-side in a lane. Our current roads could handle more traffic with greater efficiency.
And that doesn't take into account the reclamation of parking lanes of city streets that are no longer needed for parking and are available for actual transportation in a world where our built environment doesn't center on parking. There's also the newly available space of current parking lots that can be better used for just about anything else, and 2+ car garages that can be converted into living space.
The potential for societal change from self-driving cars is poised to be remarkably transformative across many dimensions.
I think the CyberCab form factor is ideal for most uses, with two seats and luggage storage for most use cases: shopping, dining out, traveling to airports, going to outdoor activities, bringing kids to events. A one seater with only room for a backpack isn’t good for these, but may mostly be good for a single person commuting to work, but even then safety, comfort and ability to use a laptop are points against it.
Was a nice read! Longest article I've read so far on substack and enjoyed reading your analysis as well as reading people's arguments against in your comment section. I think I will be thinking about this for the rest of the week. Look forward to reading more.
In my city, Toronto, there are tens of thousands of Uber and Lyft drivers (and a dwindling number of licensed cab drivers). Almost all are recent immigrants. (Many are spectacularly overqualified.) They struggle to make a living, thanks to Uber's take and oversaturation of supply in an unregulated market. Okay, we all accept that it's a nasty world, driven by tech changes, and nobody wants to be called a Luddite. But I am shocked that your essay contains barely a mention of what self-driving cars will do to those men and women whose meagre living depends on driving those of us who have a bit of money around.
If they are overqualified, then it shouldn’t be very hard to find other jobs. AI will impact jobs, with many software engineers, market researchers, graphic designers and jingle writers already looking at lost work. Do you think reducing pollution, accidents, congestion and inconvenience should be delayed so that some Uber drivers keep making money? I wonder what happened to all the people who worked as travel agents 20 years ago, I bet they found other jobs (hopefully not as Uber drivers
). And did those hard working Uber drivers not displace the jobs of conventional taxi drivers in the first place?
I can see this may result in the end of private ownership of a car, something that has been ingrained in American culture for over a hundred years. This would be an extreme societal change, maybe good, or maybe "Red Barchetta," by Rush.
Agree, but that would be Canadian culture, since it isn’t a state yet. And if you read Neal Peart’s motorcycling books, you can appreciate his love for the open road.
The only argument I’m dubious of is that they will remain affordable. They’ll be cheap for sure (incredibly so as they scale), but like so many American industries dominated by oligopolies, they’ll raise their profit margins once competition is impossible.
There is other competition. TuSimple and others are working on self driving trucks, and many Chinese companies are also working on autonomous vehicles. Tier4 in Japan has self driving buses in operation and their software is open source.
How will competition be impossible? The only way for autonomous services to compete with human drivers is for them to be proven to be safer or cheaper. If Waymo or Tesla starts taking share from Uber and Lyft, they won’t be able to raise prices too much or customers will go back to the human drivers. And if Waymo and Uber are wildly successful, other companies will enter the market, because as AI and EVs get cheaper it will be easier to do.
Fun article, but it rehashes many points from 5 years ago. This incredible advantage that Tesla has with all of its data? So far, Elon has not turned this advantage into robotaxis, which he promised years ago. I hope that we get there.
Yep, heard it all before. Since then hasn't the autopilot team gone through several rearchitectures? Just threw away years worth of work to try again. Their latest is using LLMs but who knows what hard limits they run into next.
I feel the only way to make EV's safe is what Waymo is doing now, which is trickle rolling out long march of 9s over decades, or world simulations that Nvidia is trying to build. Tesla and their mountains of data feeding Autonomous driving models for free thesis has not borne fruit.
Also, main issue is that this is an operations business. That's a different thing from manufacturing vehicles and selling them. Noone ever talks about that, like it's all going to be hand wavey solved somehow.
Tesla’s AutoPilot has fewer crashes and fatalities per mile than the national average, and FSD v13 is claimed to drive over 200 miles without an intervention. Musk says CyberCabs will be bought by private citizens who would then put them into service for hire, and they just deal with the insurance and cleaning, but this ignores the handling of situations where the autonomous driving gets stuck by a weird situation, which Waymo handles by having human remote monitors take over in those situations.
Waymo driving is getting better also, and they now also mainly use LLM based software. The one thing Waymo doesn’t do yet is freeway speed driving, which Tesla says they can do. I don’t know how they work in snow storms, they work pretty well out here in Arizona.
Yeah, we need more automation for that too. I’ve long envisioned an automated machine which can roll along damaged streets grinding up the asphalt and immediately repaving with some new elastic polymers binging the ground material so that roads won’t have potholes or cracks, and without annoying lane closures.
Thomas, great job as usual, just couple of builds:
Self driving are massively more capital efficient, their utilization is high. Most cars today sit idle AND occupy expensive space. 2) Cars are less space efficient on the road than public transport. The transfer of person-miles to SD cars from public transport will jam the streets. A new equilibrium will be needed. There will be of course self driving buses at a given time. Thanks for your work.
First, Tesla looks pretty cooked at the moment. They wasted a lot of their first mover advantage by spending time and resources on making Cybertrucks. By now even companies like Volkswagen and Stellantis are going to eat their lunch in small cars segments (which are still quite popular outside the US). Their self driving outlook has remained stable for about a decade now: it is always "a few years away". I personally find it a bit disappointing that it is now 2025 and a self driving car still isn't something you can buy soon but here we are.
Second, I don't think self driving taxis will replace public transport. They will complement it.
Guess why Mass transit is called that? It is just not possible to put those numbers of people in cars. It just doesn't add up in terms of square feet. Mass transit will lose some customers to those taxis, but they will also win customers who previously couldn't easily reach a station, but can now take a self driving taxi to the station.
Bus networks will be subject to the same sort of wins as taxis. They too are currently heavily restricted by the need to employ a human driver for every vehicle. A lot more routes will become viable, and shorter headways. If you double the number of buses, a bus network becomes much more than twice as useful.
Robotaxis meanwhile share a fundamental limitation with cars: they take up a lot of space and cause a lot of nuisance relative to the number of people they carry. That first part will limit how many will fit on streets in cities, and the second part will force almost every city to implement congestion charging. Buses can amortize that fee over more passengers so they will become cheaper.
So I think private car ownership will indeed become much less common. But the way people will get around will still be mixed, and it is not obvious to me whether public transport and mass transit will become less or more popular.
The big losers are going to be car manufacturers. We go from almost everyone owning a private car, to having a much smaller number of robotaxis. Maybe that helps explain why they're not pushing for self driving as much.
Tesla is overpriced as a stock and under managed as a company, with Musk doing his MAGA schtick. He says the value is in autonomous driving and robots, but those are all long term. CyberTruck was a distraction, but it probably only cost work on the new roadster, which probably won’t be seen anytime soon, and the low cost car to be built in Mexico, which would have been hurt by Trump’s tariffs.
Mass transit will continue to exist, but it depends on how it is run. In Tokyo, or other large Japanese cities it is great and many prefer it to private cars or cabs; but in LA, the Metro is full of drugged-out dirty mumble screaming vagrants, which means most people prefer the gridlocked traffic. Self driving works great in stop and go traffic, so I don’t think people will switch to the metro, even when it finally reaches LAX next year.
Self driving and 5G internet effectively means a lot of people will tolerate sitting in congestion for hours. You could spend 2 hours travelling 5 miles but it doesn't matter because you were getting work done on your laptop. However, this will also force that insane level of congestion on everything else. Things like appointments, ambulances, fire brigade, going to school, etc. will just stop working. This is the part that will force every city to implement congestion pricing.
And with the mass transit, that is a choice. The US is prosperous enough to have public transit without angry vagrants on it. By a wide margin.
I planned to buy Tesla a few weeks ago, but chickened out. I am convinced now.
And you did not mention robots, which I believe will be even bigger for Tesla. The economics for robots will be fierce, but we are talking BILLIONS OF THEM. The numbers just keep adding trillions to the business.
Thank you for getting me off my rear end.
Great, great article! This was fantastic...thanks!
The logic for replacing current driven rides seems reasonable. However, thereafter - especially in crowded non US cities - the possibility of replacing mass transit is going to be limited by road capacity. There is simply no room on the road to move thousands of passengers off mass transit and into self driving cars, even if they are smaller and faster than current human driven cars. Traffic will always grow to fill the roads available and that will happen long before they can out compete mass transit in cities.
Another advantage to autonomous vehicles will be their ability to communicate with each and form convoys on the interstate and work with signals and traffic controls in cities. This wil allow more cars to be on the road---especially considering traffic is often caused by the human element in combination with congestion.
Self driving cars would not be able to replace the Tokyo subways and trains, and would be hard to compete with Japanese taxis which are numerous, clean and reliable, although expensive. Some road space can be recovered by eliminating on street parking, self driving cars won’t park except at dedicated recharging garages.
Self driving buses can eliminate some cars, and deal with the current shortage of bus and taxi drivers in Japan (a reason taxis are expensive there). One company Tier4 is already providing autonomous buses service in a small city in Japan.
https://solutions.tier4.jp/l4ride
So many issues here:
1. You don't even consider the Chinese self driving operators such as Baidu who is testing with ±400 vehicles. They seem behind Waymo but not entirely out of the calculus, same may be true for Yandex.
2. You take Musk's promises seriously. His track record, Tesla's cult culture, the fact they have not test driven autonomous miles at any scale and so many other 'small issues' should make one very careful believing that the structural problems in Tesla (no LIDAR, the plan to roll out FSD without careful training, culture of risk taking and poor safety record and much more) would somehow be wished away into a globally scaled operation.
3. You take calculations from ARK invest seriously... Hard to know where to begin here but they are the most unserious 'investment fund' I have ever seen and none of their models stands external reviews well.
4. For Waymo, there is no dependency on Uber at almost any point beyond the launch periods and there is a reason they do not need Uber in San Francisco where they now have large enough user base. The opposite is not true as most car rides are done in a single local and if Waymo is cheaper, has better car experience and is available - why would a user not prefer it? In November I rode both in PHX & LA and my preference for Waymo was very strong, not to mention it was half priced when compared with like for like cars (Waymos are a premium Uber flavour).
5. There is also no dependency on car manufacturers - in Europe in 2025 YTD 600! different EV cars have been sold: https://eu-evs.com/bestSellers/ALL/Brands/Year/2025
I find it hard to find any reason why any of these manufacturers has any particular hold or agenda over AV, now that Waymo proved its a setup that can be fitted to their cars. These companies have over capacity (not only in China) and would do a lot to satisfy a client like Waymo who looks like the future.
I am quite surprised by this analysis quality, I much appreciate most of your posts but this one is not well enough thought through (and AVs are the future, but it would be Waymo's model and others like it delivering it in the next years - I doubt Tesla would even be a player).
I do mention Baidu!
Musk’s track record in industrial products is one of delivering everything, albeit late, isn’t it?
Thanks for the point on ARK. I did share my back of the envelope a couple of times. I think they’re probably right in their orders of magnitude, if not the specifics.
Interesting re waymo!
I’m not sure what specifically you disagree with. I agree that Waymo is a contender. I’m just saying that the modular approach is inferior to the integrated one right now! But you’re right that if walk holds all the sway it can bend the arms of other players. Maybe that will be the case!
A lot of good analysis from Ed Niedermeyer on the topic btw on Reddit AMA yesterday: https://www.reddit.com/r/SelfDrivingCars/comments/1j8vjup/im_ed_niedermeyer_author_and_podcaster_ama/
I'm looking forward to self driving taxis. I'm a 75 year old driving a F150. I plan on driving it into the ground, as they say. By then I'll be 90 and many will not want me driving then.
The technological changes in my lifetime have been astounding. I grew up with partyline telephones and black and white tv. I even have a computer in my hand for typing this message.
The Telsa sales graph is misleading (by now anyway). The numbers for 2024 definitely no longer rise exponentially. 2025 is looking to be way worse. Unless either Elon or Tesla's shareholders wake up and notice that the rose they once held now is brown and stinky, their prospect for winning the robocar game aren't that great.
Hmm. I was an enthusiast.
Have you actually owned a Tesla? I have. Have you watched a channel like AI Drivr? You should. The real world of driving with this software will not convince people of its safety for a while if at all.
And this is from someone who would like it to exist.
The stakes are too high for it to be better than human drivers ‘on paper’.
My Tesla would very occasionally go into a phantom braking problem when passing vehicles in the lane to the left of it that very nearly caused a rear ending at 70mph on several occasions.
Tesla point blank refused to acknowledge this problem but quietly fixed it for a while on update til a later update undid the fix. Not a trust building exercise.
And that’s just boring old lane and speed control on a straight highway.
Correct me if I’m wrong and you’re a genuine enthusiast but these sound like the usual accumulated Tesla talking points.
Kevin Kelly has made the correct (imo) observation that for mass adoption - to justify Tesla before Musk turned himself into an unavoidable liability - public roads will need sensors at junctions and corner cases which will take a lot of public money.
The real problem with all that? Nobody cares as much as the likes of us early adopting enthusiasts. Well not in big enough numbers. It’s not a big enough problem to solve for most of the population right now or in the medium term.
Certainly reliable highway cruise control - def. Robotaxis - very limited local grids only
I don’t own a Tesla so am not partial to it. I’m just looking at the data on engagements and on the fact that waymo is already here.
Projecting the current progression into the future doesn’t make you think the 2026-2027 range for ramp ups is valid?
No.
Heard it all before for nearly a decade - it’s waaay harder than it appears to get the last few percent of corner cases solved satisfactorily.
And that matters because that’s where fatalities occur at speed.
If you’ve driven one for any length of time you’ll see why the average driving person will not trust it with anything complex.
Waymo has the partial answer: drive inside regularly updated maps at fairly low speeds. And don’t over promise.
Tesla do amazing stuff but they are praying for a miracle to justify their valuation when they know there’s probably no answer without public infrastructure like sensors etc
And the buying public just don’t care. What are the numbers on take up?
After the 2024 free trial of FSD, only 2% took up the package. Keep in mind this is meant to be the central argument for the Tesla valuation.
Tesla reduced the FSD purchase price from $12,000 to $8,000 and lowered the monthly subscription fee from $200 to $99. Despite these adjustments, the uptake rate remained modest.
Way too much kool aid going round.
Re corner cases- we can make billions of them in simulation https://machinelearning.apple.com/research/robust-autonomy-emerges
I had that exact problem with our Teslas, even down to the fix and later regression again, but only when using the Autopilot feature. Never when using FSD. FSD has its own, different inconsistencies still. But I drive with it much more than without it! It's safer and less stresful and leaves me less drained upon arriving.
Agree I did too. But that’s still not a robotaxi by a long shot and that’s what’s being sold
So, which version of FSD did you have? According to the chart in the article the current version 13 has many times fewer driver interventions than version 12. I don’t know their actual safety rate compared to human drivers is, but there data shows fewer crashes per miles driven on Autopilot compared to the federal NTHB rate for all drivers.
Tesla doesn’t seem to have a plan for how to handle situations where cars get stuck (like the Waymo that was circling a roundabout for 15 minutes and the rider couldn’t stop it to get out). Waymo has rooms full of back office support drivers that can take over the driving so passengers aren’t in danger in these situations. I don’t know how Tesla plans to handle that, ask passengers to post a help message on X? Does the CyberCab even have remote operation capability?
I’m also worried that Musks bias against LIDAR is a mistake. He claims humans can drive well enough with by judging distance using only our eyesight, but humans mostly have two eyes for stereo depth perception, and we can move our heads to better visually inspect the situation. Even if Tesla cars have multiple cameras, they are fixed in a particular location and as far as I know are not stereoscopic. And the one camera at the top of the windshield seems susceptible to eventually loosing sight to a well placed bug strike during freeway driving, resulting in possible disaster.
I’m not sure of the version - and I sold it last week. But it was the full version and updated to the latest automatically.
It’s a solution looking for a problem with so many edge case issues out in the wild as you point out.
For example - imagine I’m driving as a passenger on a highway in a Robotaxi with no steering wheel. The overnight cool air has created condensation on the inside of the camera sensors on both pillars as the morning sun heats it up (happened frequently on mine) - the car is not completely blind but it sure has blind spots. Does it just pull over and wait for an hour for it to clear?
Is it teleoperated til it clears? If so how many virtual drivers do you need to reliably run a fleet for this and thousands of other edge cases?
An advanced cruise control is a huge advantage and Tesla is arguably the best at that. I’ll believe it when I see it for anything more.
Do you consider in your forecast also Europe, where Tesla sales are tanking because of Elon Musk?
Also, what about this take for getting the "regulation" approval: the guy owns the US government now and it's dismantling it.
It’s not a dimension I looked into here!
Speaking of tanking in Europe, there may be a need for autonomous tanks (and drones) depending on what happens in the Russia/Ukraine war. Can Volkswagen make those, and didn’t they also have a problematic founder?
You may get into some of these ideas in your additional work, but I would submit that full-sized cars are more of a legacy than the pinnacle of optimized design for the world you envision. Even two seats is twice as many as needed for the majority of car rides, so something more like an enclosed motorcycle/tricycle with enough storage space for a backpack or side pack would be the ideal tool for the job in most cases. Like ordering an Uber, you could specify when you need a vehicle that seats more than 1 person and request a ride for 2 or 4 or more people. But a benefit of having so many smaller/narrower vehicles, particularly when these vehicles are largely or completely able to communicate with the others around them, is that you can fit far more vehicles on the road, such as having two one-person vehicles side-by-side in a lane. Our current roads could handle more traffic with greater efficiency.
And that doesn't take into account the reclamation of parking lanes of city streets that are no longer needed for parking and are available for actual transportation in a world where our built environment doesn't center on parking. There's also the newly available space of current parking lots that can be better used for just about anything else, and 2+ car garages that can be converted into living space.
The potential for societal change from self-driving cars is poised to be remarkably transformative across many dimensions.
I would assume we’ll get there! But what you suggest is not MAYA enough yet!
I think the CyberCab form factor is ideal for most uses, with two seats and luggage storage for most use cases: shopping, dining out, traveling to airports, going to outdoor activities, bringing kids to events. A one seater with only room for a backpack isn’t good for these, but may mostly be good for a single person commuting to work, but even then safety, comfort and ability to use a laptop are points against it.
Was a nice read! Longest article I've read so far on substack and enjoyed reading your analysis as well as reading people's arguments against in your comment section. I think I will be thinking about this for the rest of the week. Look forward to reading more.
In my city, Toronto, there are tens of thousands of Uber and Lyft drivers (and a dwindling number of licensed cab drivers). Almost all are recent immigrants. (Many are spectacularly overqualified.) They struggle to make a living, thanks to Uber's take and oversaturation of supply in an unregulated market. Okay, we all accept that it's a nasty world, driven by tech changes, and nobody wants to be called a Luddite. But I am shocked that your essay contains barely a mention of what self-driving cars will do to those men and women whose meagre living depends on driving those of us who have a bit of money around.
I see this as a part of automation, which I’ve written plenty about. It’s a serious problem! Just not specific to this industry?
If they are overqualified, then it shouldn’t be very hard to find other jobs. AI will impact jobs, with many software engineers, market researchers, graphic designers and jingle writers already looking at lost work. Do you think reducing pollution, accidents, congestion and inconvenience should be delayed so that some Uber drivers keep making money? I wonder what happened to all the people who worked as travel agents 20 years ago, I bet they found other jobs (hopefully not as Uber drivers
). And did those hard working Uber drivers not displace the jobs of conventional taxi drivers in the first place?
I can see this may result in the end of private ownership of a car, something that has been ingrained in American culture for over a hundred years. This would be an extreme societal change, maybe good, or maybe "Red Barchetta," by Rush.
Agree, but that would be Canadian culture, since it isn’t a state yet. And if you read Neal Peart’s motorcycling books, you can appreciate his love for the open road.
The only argument I’m dubious of is that they will remain affordable. They’ll be cheap for sure (incredibly so as they scale), but like so many American industries dominated by oligopolies, they’ll raise their profit margins once competition is impossible.
We should be happy here’s waymo and tesla then!
There is other competition. TuSimple and others are working on self driving trucks, and many Chinese companies are also working on autonomous vehicles. Tier4 in Japan has self driving buses in operation and their software is open source.
How will competition be impossible? The only way for autonomous services to compete with human drivers is for them to be proven to be safer or cheaper. If Waymo or Tesla starts taking share from Uber and Lyft, they won’t be able to raise prices too much or customers will go back to the human drivers. And if Waymo and Uber are wildly successful, other companies will enter the market, because as AI and EVs get cheaper it will be easier to do.
Fun article, but it rehashes many points from 5 years ago. This incredible advantage that Tesla has with all of its data? So far, Elon has not turned this advantage into robotaxis, which he promised years ago. I hope that we get there.
Yep, heard it all before. Since then hasn't the autopilot team gone through several rearchitectures? Just threw away years worth of work to try again. Their latest is using LLMs but who knows what hard limits they run into next.
I feel the only way to make EV's safe is what Waymo is doing now, which is trickle rolling out long march of 9s over decades, or world simulations that Nvidia is trying to build. Tesla and their mountains of data feeding Autonomous driving models for free thesis has not borne fruit.
Also, main issue is that this is an operations business. That's a different thing from manufacturing vehicles and selling them. Noone ever talks about that, like it's all going to be hand wavey solved somehow.
Tesla’s AutoPilot has fewer crashes and fatalities per mile than the national average, and FSD v13 is claimed to drive over 200 miles without an intervention. Musk says CyberCabs will be bought by private citizens who would then put them into service for hire, and they just deal with the insurance and cleaning, but this ignores the handling of situations where the autonomous driving gets stuck by a weird situation, which Waymo handles by having human remote monitors take over in those situations.
Waymo driving is getting better also, and they now also mainly use LLM based software. The one thing Waymo doesn’t do yet is freeway speed driving, which Tesla says they can do. I don’t know how they work in snow storms, they work pretty well out here in Arizona.
If all those new trips and extra miles do arrive, we'll need more roads, and more road maintenance...
Yeah, we need more automation for that too. I’ve long envisioned an automated machine which can roll along damaged streets grinding up the asphalt and immediately repaving with some new elastic polymers binging the ground material so that roads won’t have potholes or cracks, and without annoying lane closures.
Thomas, great job as usual, just couple of builds:
Self driving are massively more capital efficient, their utilization is high. Most cars today sit idle AND occupy expensive space. 2) Cars are less space efficient on the road than public transport. The transfer of person-miles to SD cars from public transport will jam the streets. A new equilibrium will be needed. There will be of course self driving buses at a given time. Thanks for your work.
First, Tesla looks pretty cooked at the moment. They wasted a lot of their first mover advantage by spending time and resources on making Cybertrucks. By now even companies like Volkswagen and Stellantis are going to eat their lunch in small cars segments (which are still quite popular outside the US). Their self driving outlook has remained stable for about a decade now: it is always "a few years away". I personally find it a bit disappointing that it is now 2025 and a self driving car still isn't something you can buy soon but here we are.
Second, I don't think self driving taxis will replace public transport. They will complement it.
Guess why Mass transit is called that? It is just not possible to put those numbers of people in cars. It just doesn't add up in terms of square feet. Mass transit will lose some customers to those taxis, but they will also win customers who previously couldn't easily reach a station, but can now take a self driving taxi to the station.
Bus networks will be subject to the same sort of wins as taxis. They too are currently heavily restricted by the need to employ a human driver for every vehicle. A lot more routes will become viable, and shorter headways. If you double the number of buses, a bus network becomes much more than twice as useful.
Robotaxis meanwhile share a fundamental limitation with cars: they take up a lot of space and cause a lot of nuisance relative to the number of people they carry. That first part will limit how many will fit on streets in cities, and the second part will force almost every city to implement congestion charging. Buses can amortize that fee over more passengers so they will become cheaper.
So I think private car ownership will indeed become much less common. But the way people will get around will still be mixed, and it is not obvious to me whether public transport and mass transit will become less or more popular.
The big losers are going to be car manufacturers. We go from almost everyone owning a private car, to having a much smaller number of robotaxis. Maybe that helps explain why they're not pushing for self driving as much.
Tesla is overpriced as a stock and under managed as a company, with Musk doing his MAGA schtick. He says the value is in autonomous driving and robots, but those are all long term. CyberTruck was a distraction, but it probably only cost work on the new roadster, which probably won’t be seen anytime soon, and the low cost car to be built in Mexico, which would have been hurt by Trump’s tariffs.
Mass transit will continue to exist, but it depends on how it is run. In Tokyo, or other large Japanese cities it is great and many prefer it to private cars or cabs; but in LA, the Metro is full of drugged-out dirty mumble screaming vagrants, which means most people prefer the gridlocked traffic. Self driving works great in stop and go traffic, so I don’t think people will switch to the metro, even when it finally reaches LAX next year.
Self driving and 5G internet effectively means a lot of people will tolerate sitting in congestion for hours. You could spend 2 hours travelling 5 miles but it doesn't matter because you were getting work done on your laptop. However, this will also force that insane level of congestion on everything else. Things like appointments, ambulances, fire brigade, going to school, etc. will just stop working. This is the part that will force every city to implement congestion pricing.
And with the mass transit, that is a choice. The US is prosperous enough to have public transit without angry vagrants on it. By a wide margin.
I planned to buy Tesla a few weeks ago, but chickened out. I am convinced now.
And you did not mention robots, which I believe will be even bigger for Tesla. The economics for robots will be fierce, but we are talking BILLIONS OF THEM. The numbers just keep adding trillions to the business.
Thank you for getting me off my rear end.
Great, great article! This was fantastic...thanks!
I am waiting a bit on the robots. I can’t quite yet figure out feasibility or competitive advantage…