Yes, (at least) two big things to solve for with an in-house robotaxi service:
Autonomous vehicle (hardware + software)
Fleet operations
Both Waymo and Tesla have to do both.
Waymo clearly has the lead on 1, today. And there is zero evidence Tesla can gain an advantage in the next year or two. It’s just as likely or more likely that Waymo can extend its advantage - why not? Waymo already showed they could take the lead one time, why not extend it further (arguably, they extend it daily)?
Fleet operations is more of a known quantity (not bleeding edge tech like robotics). Why would Tesla have a meaningful advantage? Waymo is more experienced here as well.
Waymo expanded its unsupervised fleet to 6 additional cities in 2026 so far (and many more cities with supervised fleets for testing).
Tesla has expanded to 2 additional cities in 2026 (and barely, with tiny “fleets”, and its level of unsupervised is not transparent, expert opinion is that they are remotely supervised).
The Tesla “data advantage” and now “manufacturing and autonomy scaling advantages” are myths. They are something to aspire for. They are not things that have supporting data or a historical record of support.
My only comment is that this is probably the biggest bait and switch in the history of investing. Tesla went from “we will just flip a switch and instantly every Tesla ever made will become a human-less Uber and Lyft, and everyone can make money in their sleep without doing anything.”
And now it’s “We are manufacturing single-purpose automobiles which cannot be used for anything but robotaxi service, we are going to ‘scale’ because we have set up a production line, and we are rolling out nearly a dozen cars in three mid-size cities in 12 months…”
And the retail investors (and some of the big boys) can’t seem to tell the difference.
No wonder Musk set apart such a large dollop of his shares in SpaceX for retail investors; all the better to whip up frenzy rather than looking at what’s behind the curtain.
I think we are talking past each other. I view the major grind for Waymo as the mapping of location and geo fencing and constant need to update. I view Tesla’s approach as effectively solving for autonomy everywhere. The regulatory issues on a city by city or state by state process may be a grind for all participants. I suspect though if Tesla proves it works everywhere, it the overall process moves faster for them.
Again, I don’t disagree that it is still an if that Tesla figures it out.
For the above reason I mentioned, Tesla is solving for autonomy that works almost anywhere, at least in the US.
My understanding is this is an initial precautionary measure as a method to validate the vision only vehicles and calibrate and compare that the cameras are seeing the correct distance confirmed by the lidar vehicles. It will not be the process city to city for Tesla over the long run. It will be a vision approach in conjunction with AI. It is not the same approach as Waymo and I think this is why you potentially are missing the long term advantage for Tesla as it scales.
Of course, we have completely ignored tesla and musks proven track record of superior manufacturing efficiency, even if the autonomous systems end up relatively equal.
I see that as a liability. In one sense it is better to have those facilities before you need them rather than after you need them. But Tesla is clearly a couple years away at least from needing thousands of vehicles. In the meantime, all that capex is tied up with no return. That was a tactical mistake.
And I’m increasingly wondering if Tesla can achieve the level of autonomy they need with HW4 (which is what Cybercab has). Tesla’s capabilities are clearly behind Waymo in this area, and I highly doubt it is because Tesla lacks experience or technical ability.
If my speculations are correct, then Cybercab was a double blunder because they can’t use the vehicles they are manufacturing.
Capex is always a potential liability if it doesn’t return cash flow. The better way to think of it is Tesla has the existing room to ramp up as needed. I forget the actual numbers but the giga facilities I think begin returning cash based on existing sales inside of a 5 year period. The capex spend on Robotaxi add on is not significant yet.
We’ll just have to see. I own both Google and Tesla but I don’t own Google because I expect them to financially benefit much from the autonomous driving.
That’s the pitch. But Tesla is also using geomapping for their robotaxis. Elon Musk has hinted as such on a couple different occasions. This month, Reuters came out with an article detailing Tesla’s geomapping efforts (which they call data labeling in the article).
Inside Tesla, as these events approached, staffers worked long hours mapping routes and training the software on specific hazards to make the company’s self-driving technology appear more capable than it really is, four of the former Tesla employees told Reuters. The staffers said these labor-intensive safeguards are impossible to deploy on a broad scale.
Seven of the former data labelers told Reuters they wouldn’t trust FSD to drive them. “We have all seen it fail,” one said. Another said he wouldn’t ride in a Tesla robotaxi “if you f–king paid me.” One veteran self-driving engineer, who reviewed Tesla crash data for years, called its safety claims “b–lshit.”
The other part of the Tesla’s generalized solution narrative that I disagree with is that as far as we know Waymo and others are also working towards a generalized solution. Why wouldn’t they be? And what evidence do we have that Tesla is ahead in the race for a generalized solution? None that I’m aware of.
Elon has sold millions of vehicles without the multi-billion dollar advertising budgets of GM and Ford. He also refuses to maintain a Tesla public relations Dept to respond to media requests for comments on newspaper and TV news stories. All he needs to do is crap-post stuff on Twitter, and wave his wallet at judges and politicians.
It’s a secret sauce few understand, but many envy.
Despite the intervention of the powerful Delaware Court of Chancery, Elon got his billion dollar pay package followed by a trillion dollar one.
When those cash flows arrive makes a major difference in the investing thesis. Unlike previous models, Tesla can’t sell or apparently even use Cybercabs being produced right now. And there are good reasons to believe Tesla will never be able to sell or use them. At minimum, they are a couple years out before needing any significant number of Cybercabs. That changes the investing thesis a lot. Instead of five years, optimistically it is more like eight or more years.
It’s not being used in the same fashion as Waymo. I think I have clearly explained the difference in my prior post.
They very well could be. My thesis has two parts to it. One is a superior method of autonomy is achieved and then the other is the superior approach to manufacturing.
We’d have to know the cost to manufacture the robotaxi and how many they plan to produce in the interim. Tesla doesn’t provide a breakout of that information. But, the driving miles with robotaxi may advance the performance of the autonomous experience for retail purchasers of the other models. It could have a follow on impact on FSD subscription revenue.
McLovin covered it above, I see now, but this is written, so call it redundancy.
Well, to be fair and clear-eyed, an “understanding” is just a guess and a hope.
The actual reality and history is that Tesla is
doing very detailed mapping/validation with lidar in every city with a meaningful robotaxi presence (Austin, Bay area, Dallas, Houston)
and they are very much geo-fenced.
Further, even with all of that detailed effort, they still haven’t delivered L4 autonomy, so I find it difficult to ignore that history (and Waymo’s experience and other firms’ experiences) and say “oh yeah, that’s just now, soon they won’t need to do all that and can just scale nationwide.”
You can choose to believe that at some point this detailed validation and geo-fencing all goes away, and they “solve autonomy and scale everywhere”, but there is zero data to support that.
A recent Reuters article discussed how much hyper-local and manual effort went into Tesla’s Austin rollout.
(which again resulted in a small demonstration/test project, at best)
Okay, but you are the one making the assertion that it was wise to spend the capex before the demand arrives. In you calculation, when do the cash flows arrive?
street-by-street, hyper-local mapping/validation with lidar
differs from Waymo’s approach.
All I got was that for Tesla all of this detailed effort will “at some point not be needed and they will scale nationwide.”
I missed the explanation of how that magic happens.
All I got was:
Which looks like just a guess, without any actual data to support it because Tesla hasn’t actually done it in any meaningful operational domain (vehicles, mileage). Not even in Austin.
No, it’s based on Musk’s comments and interviews with the team and reporting on the subject. LIDAR is required for Waymo but is being used to verify with mapping that the vision and AI system is functioning as intended.
True and again I feel like I am being repetitive. Teslas advantages at scaling only matter if they solve L4 autonomy. We can dispense I think with any assumption they have or with any certainty will do so.
Having driven multiple versions of FSD over the years, it continues to improve and work towards L4 capabilities. The fact that Tesla has the most advanced version of FSD available of any manufacturer, says something.
Again, the article does not refute that robo taxis are vision only vehicles already functioning without LIDAR. The article also does not refute the claim that it is for purposes of calibrating the vision system and making sure it sees what LiDAR does in terms of distance and other measures.
You’re kind of repeating yourself without recognizing that distinction. As the former employee stated, it’s laborious and not scalable. Again, it plays into my argument that Tesla’s approach, if it works with vision and AI, becomes much more easily scalable than Waymo’s approach.
I think what it boils down to is that I believe Tesla can solve the L4 problem and you think they cannot. Both are equally unprovable positions. If they do solve it, I think I have a logical thesis for why they win the long game.
“While competitors like Waymo use a multimodal approach relying on multiple LiDAR units mounted directly on commercial robotaxis for real-time navigation, Tesla treats LiDAR as a development and training tool rather than a required hardware component”
In other words, once the AI system and cameras sufficiently operate without error, mapping is no longer needed and lidar no long serves a purpose. If the AI sufficiently understands the rules of the game in one city, it can in theory operate successfully in most others. With Waymo, at least as it operates now, mapping and re-mapping as streets change, is a permanent requirement to function.
Tesla has yet to do it in a single city, that’s the actual evidence.
Maybe they will, eventually.
But Waymo, is not standing still, they are optimizing a product already in service and have been gathering 4 million miles of experience and true autonomous data every week (and are beyond 200 million miles total) while Tesla is in its march of 9s challenge to improve safety.
Correct me if I am wrong, it still requires mapping of each geo fenced area in which Google intends to operate.
Longer term the expectation is Waymo will cut 42% of all sensors and the estimated production cost of the vehicle will still be 40% to 100% higher than Tesla. It has two disadvantages. It may be heavily reliant on third party manufacturers of the vehicle and currently has no strategy for eliminating its reliance on the mapping in real time.
The actual evidence is Tesla has made meaningful progress in Supervised FSD over the last 5 years and the robotaxi is its first attempt at full blown autonomy. The fact they have not reached L4 does not mean they are not progressing towards it.
Again, I don’t need to convince you to own Tesla. My thesis only needs to be sufficient for my own capital allocation decisions.
The counter is that supervised and unsupervised are worlds apart in terms of difficulty.
In the march of 9s to safety, Tesla is maybe at 3 or 4 9s: 1 incident in 1000 or maybe 10000 miles (based on flawed crowd-sourced data, but we can’t know precisely, because Tesla doesn’t share detailed safety data).
They need to get at least 6 9s, 1 incident per million miles to get on par with humans (or even better, if they want to get safety near where Waymo is today).
But, as Tesla can attest (as can Waymo and everyone else), the journey to “better than human” driving is a long slog.
Here are 10 reasons to doubt Tesla autonomy in the near term.
I’m happy to be wrong or corrected with a genuine argument otherwise.
Again, you are talking past my argument. I think Tesla can solve autonomy but it has not at this point and totally agree it will continue to be a difficult journey in fully solving the issue. Assuming it does, Telsa’s approach, reliant on AI and cameras, along with its manufacturing prowess, gives it several advantages over WAYMO in then scaling operations. Admittedly, one has to accept that both companies solve full autonomy.
You can with full utility make the claim that WAYMO is farther along in terms of the March of 9s. It is even entirely possible that WAYMO solves the issue entirely ahead of Tesla and yet Tesla is the long term winner because of the cost differential in scaling.
Again, it makes little sense to argue over whether Tesla can or cannot achieve L4 vehicle autonomy. It’s why I never make that argument. I fully admit that it is rationale to be pessimistic that Tesla will solve it and or optimistic that it will, so makes a market.