What about the rest of the costs?
Nvidia chips, Dojo1, Dojo2, … Human drivers, all of the people watching 10 billion miles of driving videos
That’s gotta add up.
Not happening, at least at Tesla.
The issue of Tesla’s profit and potential profit from FSD is interesting. Unfortunately, Tesla doesn’t break out the percentage of R&D attributable to FSD, and I suspect it’s mostly combined with R&D for robotaxi, and given the shared technology, it’s going to be impossible to separate them.
One thing the subscription model does is accelerate the revenue recognition aspects. Tesla has been holding back some of bought FSD revenue as deferred since the product isn’t complete yet. With monthly subscriptions, that revenue is fully recognized at the end of the month. It also helps with potential lawsuits over what was promised versus delivered.
Tesla’s overall R&D budget is around $6.4B annually, but that includes everything Tesla is developing, not just FSD software. Depending on what you think is the future business model contributions, all of FSD could be thought of as R&D for the robotaxi business, so getting some of that money back is neat for Tesla.
Musk definitely created this problem in the first place, making “appreciating asset” claims with higher and higher prices for FSD at first, while FSD’s progress didn’t reach his predictions, and then declines in the bought prices. The switch to an subscription model only helps to satisfyingly one of Musk’s pay package goals, as well.
It is a perfectly valid concept. E.g. is the ultimate FSD required a different controller, that would go to COGS/ If it were necessary to staff remote operations centers to support the cars, that would be COGS. But, all the stuff already in the car is COGS for the car, not for FSD.
R&D, not COGS. It is getting spent if they never sell a car or FSD.
But if there isn’t any of those things, is there a “good” being sold?
To use a silly example, suppose my purchase of a Cybertruck contained a contractual clause that prohibited me from putting any advertising on the exterior of the truck through a wrap. However, I could pay Tesla a fee to waive that clause - an “advertising license.” That’s not the sort of thing that’s a “good” - no company would ever apply a “COGS” accounting term to that. It’s a source of revenue, but is it really something you would apply the concepts of “gross margin” or “COGS” to?
Again, just curious. Not an accountant, so I wanted to know.
Not all that interesting. I have absolutely no idea if Tesla has a cost advantage, and neither does anyone else on this board. Developing a new vehicle can cost literally billions of dollars. If Tesla can only pump out a couple thousand units a year for a few years, then they almost certainly don’t have a cost advantage. And let’s say vision only is cheaper upfront, but has more accidents which negates the savings. It is possible, but I have no idea. Or we’ve heard that something like 85% (or whatever it is) of taxi rides are less than two passengers. So a two passenger vehicle is the sweet spot. On the other hand, you can increase the size of your potential ridership by 15% by offering another row of seats. Higher upfront cost, but there is 15% of revenue sitting there on the table.
We don’t know what Waymo’s business plan is explicitly, but we can make some good guesses based on what they’ve said. The most likely business model is a hybrid approach that eventually tilts toward licensing. Initially, Waymo uses its own fleet to prove the technology. That’s where we are now. Once that is established, they license the tech to auto manufacturers and maybe Uber. This allows them to collect the high margin software subscription dollars and leave the low margin manufacturing business to others.
Tesla wants to control the whole stack (although they’ve said they are willing to license their tech to others), and keep all the dollars from manufacturing through ride hailing. That can work. That’s basically what Apple did.
So I’ll go on record as saying I don’t know who has the cost advantage, if anyone. Not an interesting take, as I said.
I was being kind.
Anyone who’s followed Tesla knows that they’re targeting an under $30K price, and it makes sense that a purpose-built small robotaxi will cost far less than taking a regular car and post-assembly modifying it, as Waymo does. We have a way better idea of CyberCab’s cost advantage than we do about any redundancy hardware dependencies, yet you keep bringing those up as some kind of gate Tesla can’t cross.
Really? You think that’s possible, just “a couple thousand units a year for a few years?” From a company that produces over a million Model Y’s a year and hundreds of thousands of Model 3s a year?
Musk has tried to license FSD but no one wants it:
2 Likes
Supposedly, Waymo lost around $1.5 billion in 2024. Google and Amazon each have about $100 billion (and growing every year) in cash on hand while Tesla has around $40 billion.
All three companies can afford to undercut each other in the robotaxi business for ages. Robotaxi margins are going to be negative or close to zero for a long time.
1 Like
Anyone who’s followed Tesla knows they were targeting an under $40K price for the Cybertruck.
Anyone who’s followed Tesla knows they were targeting robotaxis would be available to the half the population by the end of 2025.
Some might say it is wise to evaluate these claims for yourself rather than to accept these claims at face value.
I again assure you that regulation plays a role in AV ride hailing. Ride hailing in general is highly regulated. As much as you disagree, it takes time to work through the permit process.
It is possible Tesla produces millions of cybercabs this year, but without permits they will all be sitting in parking lots. Tesla is at the very beginning of the permitting process and only in a limited area.
But I’ll bite. Let’s say down a friendly marker. How many cybercabs do you think will be delivered this year?
True. Elon’s push to cram a bunch of new tech delayed the vehicle and raised the price, and, very unfortunately, reduced the specs. A complete failure, exacerbated by Elon’s politics.
But, I don’t extend that to CyberCab and is pricing. Targeting $25k-$30k, so maybe it comes in at $35k. Still way cheaper than anything Waymo can cobble together.
I don’t disagree. I think that while Tesla is at step 1 of 3 for driverless rides in the SF Bay Area, they’ll be lucky to get approval by the end of Q3 - possibly not until Q4. Austin may clear them sooner, though.
There are sightings of manually driven CyberCabs in multiple cities, including the Northeast. I don’t know if Tesla has started the permitting outside of Austin or the Bay Area, probably not yet. The first approval anywhere, if followed by a good safety record, paves the way for more.
I have no idea was “delivered” means for robotaxi deployments. What I think is important is that Tesla gets approval for paid driverless rides in both Austin and the Bay Area by EOY. With Waymo today running 1,000 cars in the Bay Area and only 200 in Austin, I’m expecting only a few hundred Tesla cars in operation in each (probably starting with even less). As they prove themselves, it’ll be interesting to see what the ramp up in both vehicles and area served is, but that’s for 2027.
Trying to make specific predictions in a new area for Tesla is fraught with speculation, and is not, in my opinion, a useful exercise. What I believe is important is that Tesla moves smoothly through the permitting process and that first deployments go smoothly.
But, while Waymo appears limited by its ability to make/prepare autonomous vehicles, I don’t expect production to be the limiting factor for Tesla.
1 Like
At least the company has experience in this field:
https://electrek.co/2025/06/17/tesla-tsla-inventory-overflow-parking-lots-all-over-us/
There are dozens, maybe hundreds of such articles from just the US. I haven’t looked internationally, but there might be more. Like I say, experience counts.
5 Likes
My bear case, if I have one, is fairly simple and straight forward. TSLA PE is about 300.
If they fail to (timely) deliver on just about any of their projects (like they failed to deliver on solar panels, the cybertruck, the roadster, cheap EV), then that PE should decrease substantially. Will it? Who knows. Tesla investors as by far the most irrational bunch of investors in existence - as evidenced by the fact that you claim bears got crushed last year yet TSLA is up less than 2% from this date last year.
By contrast, NVDA with a PE of about 45, is up 30%.
4 Likes
I wanted to update my bear case. Rivian is coming out with FSD at $50/month. I think this is where Tesla FSD is headed ($50/month). As more and more car makers come out with their own FSD, it will just get cheaper and cheaper. So basically, TSLA FSD will end up having almost no impact on TSLA revenue/income.
I think the same thing will happen with Robotaxi. Maybe TSLA does start off with a cheaper Robotaxi, but Google and Amazon can afford to cut taxi rates to compete with TSLA while they figure out how to make their own cheaper robotaxis.
So again, Robotaxi will have almost no impact on TSLA revenue/income other than maybe causing TSLA to lose money on their Robotaxi venture for several years.
As for Optimus, I think any value there will go to xAI as they will create the ‘brains’ of Optimus.
1 Like
Yes, software, a license, whatever still has a potential COGS … even is what is being sold isn’t very goodlike.
Well, the real bear case for self-owned vehicles is that the ADAS function goes to $0/month. That it just comes standard as part of the car. That’s what BYD is doing with their God’s Eye system in China.
That might happen due to competitive pressure - but it also might land there because of regulatory/liability pressure. Especially if all the equipment necessary for the system comes standard in the car anyway. That’s because ADAS and AV drivers serve a safety function that make the cars safer for people other than the driver and passengers. A car with ADAS ostensibly poses less of a risk to other drivers and pedestrians.
So….how can a manufacturer simply choose to have their car be less safe unless the driver pays them off? They can just flick a switch, at no extra cost, and the car will be X% safer for other people using the road - and they choose not to turn on the life-saving feature for no reason other than they’re not getting paid a premium for it?
With other types of safety features that involve a marginal cost to the manufacturer, there’s recognition of a cost/safety tradeoff: not every car has to have premium anti-lock brakes, because it costs more to put those in. That’s okay - as long as you meet a minimum safety standard, it’s not negligent to not spend more money to exceed it. But if there’s zero marginal cost to you to make the car much safer, and you simply choose not to unless someone pays you $100 per month? Boy, that’s a hard case for you in both the regulatory and legal liability arena….
1 Like
Anyone have an idea of the accident rate of ICE taxis compared to ICE privately owned cars?
The larger question is whether Rivian’s autonomy will be as good as Tesla’s FSD is, not just today, but when Rivian actually comes out with theirs.
Rivian’s in an interesting spot. Unlike Mercedes and Lucid (and probably other OEMs), their end state is not leveraging Nvidia’s Drive AGX OS. According to close Rivian-follower Travis Ketchum, Rivian’s Gen 2 and Gen 2.5 for the initial R2 production are using Nvidia (hardware, not the new AGX OS), but Gen 3 comes later this year which is “full custom software and hardware.” I wonder, though, from where Rivian is getting the drive data - they’e sold far fewer vehicles than Tesla and for less years.
Unlike Tesla, Rivian hasn’t said their existing vehicles, or even the upcoming Gen 2 and 2.5 vehicles, will be compatible with the Gen 3 software (and certainly not the Gen 3 hardware). So, it could be that a reduced feature/capability L2 system is $50 on Gen 2.5 and earlier, while the more fully featured/more capable L2++ system on Gen 3 will be priced higher. Or maybe Rivian will just increase vehicle prices to compensate.
Mercedes has “paused” their L3 offering (I wonder where that leaves current owners), and is pushing the Nvidia-based L2++ system for the new CLA. Car and Driver says:
Three years of MB.Drive Assist Pro will cost $3950, but the technology will also be offered on a monthly subscription basis.
$3950/36 months = $109.70/month, so more than Tesla, and individual months like Tesla could be even more.
Also, like Waymo, the Nvidia-based system requires pre-mapping. Again from C&D:
MB.Drive Assist Pro is also geofenced to where the system has map data, but it will roll out to all cities, eventually.
FSD works everywhere and without any pre-mapping requirement other than what has been done for decades for navigation.
1 Like
Although Waymo only has 130 million or so real, road miles, they have billions of simulated miles. I suspect Rivian can also test their system with billions of simulated miles.
Combine billions of simulated miles with a geofenced coverage area and Rivian can probably catch up to Tesla in Rivian’s geofenced areas pretty quickly.
Then why was Rivian board member Kracfik claiming that Waymo’s collection of real miles was so important?
Simulated miles are great, but not a full substitute for real miles and real world weird situations. Also, when your simulations require not just 3D renderings, but also fully synchronized LiDAR and radar data, they’re much more difficult to generate.