特斯拉通往775美元的路径依然可行；技术问题可通过硬件升级解决，但Cybercab的大规模普及取决于更快的边缘案例验证以及有利的融资条件。

Is the bullcase for Tesla screwed if the chain problem doesn't get solved? What about the Roadster launch? How about what Arc Invest just said?

so they say, Kevin, you know, if the bull case for Tesla is getting to $775 per share and beyond as soon as possible, then we've got to scale cyber caps. But how do we scale cyber caps if we can't even get through the chain problem?

This is an example where electrified's kind of describing the issue with the chain.

It's a problem for those with hardware 3 cameras. See, hardware 3 cameras, they basically take horizontal objects and can essentially process them away through convolutional neural network data compressing.

That is a lot of information there, but it's essentially when you have this artifact that appears on a camera for a model 3 hardware that frankly disappears. Here's an Astra output of roughly what that would look like.

This is a chain imaged on a one camera output with hardware 3. So it's a combination of both older cameras plus older compute.

And then what the computer actually then sees after processing is that these 1 to two pixels of chain data disappear. You end up with this image here which essentially suggests the car continues to drive into the chain and therefore you have the chain problem.

people go, "Oh my gosh, this is this means it's not solvable. This means we can't get to full self-driving." The solution, well, I mean, there are two ways to solve it. There's the right solution, and then there's just an example of how you can quickly solve it.

One way you could like instantaneously char or solve this horizontal chain problem for Tesla full self-driving is you could literally turn the camera sideways and now all of a sudden these horizontal issues go away and the chains become very clear vertical artifacts which are surprisingly really easy for the AI to see.

But then you don't have a unified stream anymore between a vertical camera and a horizontal camera and your computer's like what is this? This isn't going to work. So you solve this using occupancy networks as well as temporal context.

Okay, in English, basically what you're going to do is when you see little tiny objects, you're going to turn them into big Minecraft style blocks and they're going to send a signal that this area is occupied.

But that's not going to happen just once and then it's gone in the next picture. The temporal context that Tesla uses in imaging is if they see something in one frame, they remember that context of where it is and what was there in the future frames.

So if you see the chain in one frame, you expect it's still going to be there in the future. That is a very realistic solution to the chain problem. Plus what are called skip connections.

Skip connections. Oversimplifying again. It's basically a way that you literally take some highquality imagery data from the cameras. You skip the compression entirely and then that way uh the algorithms at the end essentially get a high resolution context, a low resolution context of what everything is.

You know, 99% of the work could be low res but then you're doing almost so to speak this double check with a high-res version and then you're more likely to see the chain.

The problem with those three solutions, the temporal context, the uh occupancy networks, right, the little Minecraft blocking, and then of course skip connecting is it takes a lot of hardware, and that's what hardware 3 does not have.

See, hardware 3, and people forget this. I think this should honestly we should just write it down because there are so many examples where people dump on Tesla and they're like, "Oh, Tesla's FSD sucks."

But then they're using hardware 3 vehicles as an example.

That's not I mean that's like referring to an iPhone from 7 years ago. You know what is that? That's like the iPhone 9 or 10 or whatever.

And I'm not saying that doesn't still work. They do. I mean my black S or sorry black X runs on hardware 3 and she rides great.

It's just not as great as the latest and greatest cyber beast with hardware 4. Certainly the cyber caps.

When you look at hardware 3, you have 2019 technology, 12 cores, 1.2 megapixels of data, and 2x8 GBs of RAM.

Okay, this is 2019 technology people complain about and they're like, "Oh, why can't it see chains?" Because it takes better hardware to be able to see it. That's just the way it is.

Now, Tesla already comes with redundancy, right? This is the multiplication. It's basically two side byside 8 gigabyte RAM systems and it's kind of like a plane, you know, two engines, two pilots, two fuel pumps, uh two brake systems.

I've had to use that, you know, my very own aircraft that has come up before. We're like, "Wow, one of our brake systems is dead. Let's hope the E brake works." I'm obviously still here, so you could guess uh it worked.

But anyway, uh you know, two gear down system. Oh, you know, the the uh hydraulically uh actuated system doesn't work. No problem. We got a backup.

So, redundancy is built in. It's built in at the various different layers as well. But think about 20 uh or hardware 4. What that is hardware 4 is 2023 and it really scaled in like 2024 technology.

Uh so hardware 4, you're looking at almost twice as many cores at 20 cores. you're looking at like 5 megapixel cameras and you're doing 16 GB* 2, right? That's the 2023 technology.

And know that the cyber cabs are probably doubling at least the RAM. Again, we don't actually have the answers yet on what the RAM is for the cyber cab, but we believe it's going to be two 32 GB sidebysers.

So you can actually have much more of a an essential on edge device, the car neural net, you would have in some of these older generations. You could basically run bigger models with guess what? More context, more headroom.

Hey, how come AI hallucinates so much less today than it used to? because we kind of broke through the memory wall with high bandwidth memory and putting a lot more head space or head room on chip stacks.

You know, if you're now using a GB 6000 with 96 GB of VRAM and you're running a model that previously ran on 32, I can still run that model on 32 on the 96, but I now have a bunch more room, another 64 GB for context, which would be like the chat context that we've been talking about or how big of a contract or code base I could put in.

It's all kind of the same for Tesla. So, I think it's a little unfair for people to look and go, "Oh, well, you know, the 2019 hardware can't see the chains. This is the end. It's existential for Tesla." N man, this is called technology.

Now, in fairness, one of the concerns uh and this is probably the more realistic issue for Tesla. The more realistic issue isn't that they're, you know, superersizing the hardware on the Cyber Cab, which they are compared to what's available for retail right now, but that's okay.

Um, we expect every single generation, every single year, hardware is just going to keep getting better and better and better. The bigger issue isn't so much the chain problem, which I could show you.

There are two ways like graphically we can solve the chain problem. One is the way I said hardware 4 getting much more detail and compute headroom.

And then what do you see? I see the pixels of the chain. Wow, I see a full chain here. Oh, guess what? So, we don't drive through it and we stop before that chain, right?

But the cyber cabab does not only that but the cyber cab also does something and this actually brings up one of the risks. The cyber cab does something as uh avoiding known problematic areas.

Now they might have a fancier word for this but here's basically what that would look like. So a the cyber cab would take a hardware 4 plus so that 32* 2GB hardware which again is massively more well it's four times more than what we had at the prior generation but for ondevice AI that's huge you could run a much bigger model and again much more context right context matters especially with driving chain doesn't just disappear on you

but anyway uh cybercab probably also uses oh there's the phrase known entrance restriction where you kind of historically know that this is a particular area we're not supposed to drive through which may have been validated by humans and so the car sees that area and it doesn't even care if there's a chain or not.

It's just not going to take that little roadway there.

So again, chain or not, the cyber cab is just going to stay away from that area. This is Astra, by the way, which I thought was kind of a cool little mockup. I mean, it threw that together in like four minutes.

It probably burned like $50 worth of tokens. I I I don't know if it actually did it in um but you know when we put this together we start going okay all right so what does that mean for the cyber cap it's going to have better headroom better data there but the known entrance restrictions goes back to the edging thesis and this is the greatest concern that I have in terms of the speed that Tesla can actually mass market these puppies

so you might already know this we've talked about this before but electrified actually they put together I think with perplexity a nice little uh Versel app dashboard. Uh I'll link it in the description.

Uh you could also get it on their channel. But anyway, I thought this was kind of cool. It's just basically a little map of where these services you're expanding to.

But the most important thing, and we've been talking about it, is right here. It's the timeline, the live counters, right? And this is just a basic counter that says, "Look, we've actively been operating cyber cabs now in Austin for 14 days, but the whole robo taxi network has been there for about 448, which actually feels like a 100 days longer than I remember, but I guess time just goes by faster sometimes."

But anyway, uh, San Francisco Bay Area is starting, you know, to see some support for those cyber cabs, but a lot of these other areas, uh, uh, you know, Tampa, Orlando, Miami, Dallas, we're still unsupported on those cyber caps.

And so, I think what's going to happen is we what we really need is a dashboard that shows us how many days, maybe I'll make one, how many days was it from when robo taxis got introduced in Austin to we have cyber caps?

And so it is really not a hard thing to demonstrate, but let's imagine this. Okay, so you've got Austin was 0 to 436 days to get to the cyber cab, right? We need the very next place to look like this.

And then the place thereafter look like this. And as soon as it gets down to this where this is like, hey, we could basically edge out an area in 90 days or 30 days, that's when you need to get excited about Tesla stock because that's when the actual scale is going to come.

Right now, we're still in this sort of unproven hype of dude, we just mapped out Austin. And I'm not saying Whimo style mapping. I want to make that clear. Uh I don't think that they are mapping because they did say they're not mapping but I think what they're doing is they're test validating the area so much that they've effectively mapped all of the edge cases. So that's why I call it edge mapping.

And again the compression and time to release in a new area is going to be your proof in the pudding so to speak of when this company actually gets cyber caps to mass scale. Now, I want to touch a little bit on what Arc Invest said.

Uh, and then of course, yes, we've got the catalyst of V15 coming up. That's great, but again, all of it is always going to be a demand for more and more hardware. That's understandable.

Uh, let's focus on this uh interview here. This was actually a conversation between Sam and Brett over at Arc Invest. And I'm just going to give you some bottom lines uh that I got from this.

So, some of the bottom lines that I got from this on the Cyber Cab were that uh and I didn't model this originally when I was looking at my numbers for Tesla bullcase like $775 per share.

It totally makes sense that in addition to selling the car and taking margin on the car, you're also going to take a platform fee for people to basically submit their cabs to the robo taxi app through the Tesla app.

Well, through the robo taxi app that Tesla makes.

The interesting thing there is you're going to have to really really and we have we have mentioned this that if you get below a certain star review because you run a bad fleet service, they're just going to kick you off the app.

That I mentioned, but what I didn't mention is that Tesla's probably going to take a percentage like a McDonald's franchise fee. You know, give me 6% off the top. That's another potential product or profit vertical where Tesla gets big margin up front selling the car, but then gets the residual from full self-driving and the platform fee.

It's like owning all the McDonald's real estate almost and and collecting that 6% fee on all the franchisers that are running the actual businesses. So, you know, they're dealing with all the crap basically of, you know, people messing up their cars. Uh, and Tesla's not.

Now, they also mentioned uh Brett specifically thinks that the cost for a cyber cab is going to be about 25 cents per mile. Uh I estimated the c the cost higher than this because I included financing in my estimate.

And so I do think there is still an element of this business model that is interest rate sensitive even when it does scale.

So a second tailwind after we start scaling this is absolutely going to be financing rates.

Now, who knows? Maybe Tesla will offer, you know, business financing at a lower rate, but then they're eating the cost and discount points. So, that's worth knowing because sort of where you put Tesla on the scale of your investment portfolio.

You should be thinking to yourself, we need to validate the scale out for compressing edge mapping and we need interest rates down. You get both of those things happening at once, this thing can rocket ship because that's when you can really blow this cyber cab business up.

And then as Brett himself says, you'll be able to turn the screws on Whimo. Now, the one thing you won't be able to turn the screws on is the price for the stuff over at me.com because we don't have a coupon code right now.

And I apologize for that. I just want to be transparent about it.

Now, with that said, uh they're uh you know, Brett argues that Whimo, if Tesla can get their costs down to under $1 per mile, which they expect that Tesla will be able to, and I think they'll be able to as well.

They think that Whimo will be too expensive to be able to actually compete under that price.

That Whimo's a great experiment uh and it does a great job, but it's not a scalable job. And this is the classic sort of Tesla argument is that the Tesla strategy is scalable whereas the Whimo one may not be.

Uh and so we'll see because right now to me it certainly seems like we are mapping just the way Whimo does uh in terms of validation. Not necessarily mapping the map but mapping the edge cases.

Ah there's a chain there. Again the cyber cab would honestly probably realize the chain anyway but there are still plenty of edge case scenarios that you want to have mapped out.

And the benefit Tesla has is it could crowdsource all of this. This is exactly why now anytime you override your FSD, it asks you, hey, did you override for navigation, for parking, for a critical issue or other?

And and then the validators at Tesla can prioritize uh their refinements of this software based on this feedback they're collecting at a massive scale. It's actually pretty impressive.

Now, uh Brett, good insights. I liked it. Uh he also thinks it's really important we get V15 out that that is somewhat gating uh their expansion capability here.

Although I I would guess that the cyber cabs are already operating all of that. And that's sort of the difference is in the public you have a ton of Model 3s, you have some Model 4s, hardware fours I should say.

So hardware 3es, a lot of those, some hardware fours. Basically the sales from 2024ish on some of 2023. Uh and and then the cyber caps, they have the benefit of all being the latest hardware 4 plus.

See, and that's why we've had to like quantize down the hardware for um V14 to turn into hardware 14 light for hardware 3 because you've got to squeeze all that context into a smaller memory bucket. Otherwise, hardware 3 is going to lag.

And that's even worse. Like you could have a lagging FSD that just drives literally into a a park bench or something stupid or another car because it lagged. You don't want that.

Uh just to solve an edge case, have your 99% reality fail. You don't want that.

So hardware 3 is just frankly ultimately going to end up getting left behind. And that's going to be disappointing for some existing holders of the car. But I I think a lot of people well there are going to be people who are very upset and you know lawsuits are going to go flying or whatever around this.

Uh I think the vast majority of people are like all right well you know the car is

And I'm not saying oh well you were sold a lie or whatever. Okay that debate could be had separately. I get it. Uh but uh you know Elon has been very optimistic. We will say that.

So I'm not here to take a side on that. I'm here more just to say is it existential for Tesla? No.

The Roadster launch uh on 101 this whole like go for launch thesis. This is really cool. Uh this is, you know, we've covered this with the information's reporting. The information reported that this would end up being a stunt, an elaborate stunt, that no people will be inside of the vehicle, that the vehicle would be able to fly because of its gold uh cold gas thrusters co-developed with SpaceX, but it's so loud that it could potentially damage your eardrums.

So, it'll be remotely operated uh probably in uh at the SPA uh SpaceX facility in Texas.

I don't think it's actually a big catalyst for Tesla. I think it's more likely that that could be by the rumor sell the news event. But if you believe in the scaling of Cyber Cab and you believe in uh this being a rate sensitive that'll benefit because rates will come down in long term, you know, there's some value here.

Uh especially since I love the Wall Street Journal reporting on this lately. Tesla reclaiming the US EV market as legacy automakers retreat. I find this quite interesting. The company hasn't escaped the wider EV downturn.

Tesla sales are down 16% year-over-year. However, the rest of the EV market is down 30%. So yes, they're slowing, but they're slowing more slowly than everybody else.

Uh, and so that means there are fewer alternatives since legacy automakers are kind of giving up on electric vehicles, which leaves you with really if you want electric vehicle, leaves you with a Tesla.

And Tesla still has the best full self-driving stack.

That's where I think this these kind of headlines. We've seen these headlines for years. Hyundai says it will overtake Tesla in self-driving within five years. I've heard people say that now for what feels like a decade.

Uh I mean I got my first Tesla in 2017. So ah man, it's already almost been a decade, dude. Come on, man. Like that was magical when when I first had autopilot in December of 2017 and then I bought the stock and then it made millions of dollars.

Um you know, some of that magic is gone now. Obviously, uh it's been a long time. You get used to it. It's still amazing. I feel like I don't drive anymore, but I've been hearing the same crap since then.

Oh, you know, this company or this company's going to come in with FSD or uh some kind of competitive software, and I just don't see it yet. I'm sure it'll come, but it's obvious to me who's who's in the lead right now. And I hope that Tesla can keep this going.

Uh oh, I already erased it. I was just going to say it's all going to come down to the edge compression. That's what you need. Compress the edging. The faster we can get from edging to completion, the better.

Okay, here's somebody else arguing that there's a JP Morgan bombshell note claiming that nearly all robo taxi billions will go to Tesla, not you.

Basically, this is the argument that, hey, you were sold a lie that you were going to be able to put your own car in the fleet and rent it out and have it make you money at night.

Uh, you know, I think most of us kind of looked at that as a little bit of a pipe dream anyway from day one.

We could be hopeful for it, but let's be real, they're going to need the latest and greatest hardware, which they're going to be cyber cabs at first. You'll be able to buy those in the future once we get through the edging problem and submit those to the network.

But are we just going to throw our regular car into the network? Highly unlikely.

Uh and it's also going to create a real struggle I feel like for uh maintenance in terms of monitoring the charging and the cleaning. You don't have the scale for regularly cleaning these.

You know, what are you going to run out to the car every time somebody reports that it was stinky or whatever and then you have to stop what you're doing and clean it. You know, this is this is something that would work well anyway for fleet operators.

And I suppose Tesla could have like cleaning stations or whatever, but I don't think that's where the margin is.

So there's there's sort of like a this utopian scenario where oh but I I can make money while I sleep and then there's like reality which is no Tesla's going to offload the crappiest part of this business to fleet operators who could do it with scale and they're happy taking a little bit of a margin uh and a big tax break you know 100% tax write off on capital expenditures thanks to Trump and then Tesla is going to take the massive margin from the cyber cab taking a licensing fee on the app and uh taking uh uh FSD subscriptions which are like you know 90 plus% margins or whatever uh incremental margins since you know you're training to sell the car in the first place anyway to even have it be full self-driving you could argue the incremental margins are like 99%. That's pretty remarkable.

At the Tesla CyberCap launch they said that they've been able to reduce the length of the unboxed manufacturing process compared to the traditional process by 50%. which is still fantastic, but it ain't no 90%.

Now, the only place you could maybe argue 90% is that there's this forecast that maybe a vehicle will be able to come off, you know, their total lines once every 5 seconds versus once every 47 seconds for a Model Y.

That's where you could argue 90%. But she just told you that they have been able to achieve a 90%, you know, that 90% greater efficiency. No, they haven't. That's the point. That's the aspiration. We're not there yet.

And so that little word or that phrase there makes a real big difference because you don't want to buy the stock assuming they've already gotten to this scale. Those of us who are really paying attention, we're like, we already know that like this is still coming. But like not everybody does.

All right. Uh then of course you have to remember like you're constantly going to have negative press around Tesla. Like here's a piece 14year-old killed in Pacific Beach ebike crash involving Tesla.

Like I don't even know why they have to say involving Tesla. I think it's just because they realize it's clickbay.

Do I actually think we're going to be able to order a Roadster anytime soon? Honestly, no. I think this is a show event. and the Roadster event. If you want to buy more shares long-term for Tesla, I think you'll probably have a shot at doing that, you know, #notpersonalized advice after the Roadster launch event because people will see it launch and they'll be like, "So, when can I buy it?"

And then people go, "Oh, you can't." That was just to demonstrate how cool our thrusters are. And um we're we're, you know, we're still in R&D and people are going to be pissed about it for now.

People hear, "Oh yeah, th's coming. This is great. be a little the point is with Teslas, you're going to have to be really patient. I think that's uh the bottom line. Let them complete their edging and uh then we'll be a lot further along.