Transcription
So buried inside Tesla's earnings call, underneath all the noise about the profit miss, there was actually a very trackable road map that Tesla has given us. Some real commitments, uh, that was on the record, and some of this is close, not years out.
We're going to talk about the Terra Fab. It's going to get a location announcement soon. Elon refused to squeeze it into the call because he says it deserves its own day in the spotlight. Think about what that tells you though.
Optimus production lines are going in at Fremont right now. We know that on the same floor where Model S and X used to be built. But they shared some photos of the actual production line. Elon called the line stunning to see, and they shared that it's going to be in production soon.
Mega Pac 3 starts production soon. Also, the Semi starts driving itself around the end of this year. Apparently, maybe early next year. AI5 chips hit volume production around the middle of next year. And he said AI6 is on track as well and something to behold. And they go into robots before they go into the cars. CyberCap customer rides are close, his words, not too distant future, and robo taxi keeps compounding 10% a week through the rest of the year. Every one of these is a checkable milestone. We're going to track them closely.
Today, we got Brian Wang joining us. Brian is a futurist, thought leader, and a popular science blogger with 1 million readers per month. Check out his YouTube channel, Nextpick Future, and his blog, nextpicfuture.com, is ranked number one science news blog. Welcome, Brian.
>> Great to be here, Eric.
All right. So, what we're going to do is go through each of the topics that was brought up during the earnings call, the ones that matter for Tesla investors, but there's kind of some sort of timeline. Some of them are more concrete than others. Um, the big one I think, uh, he alluded to, he thought it was really big, is a Terra Fab. He talked about that going to get a location announcement soon. Then he said, "We really can't talk about it too much in this call because it deserves its own day in a spotlight." Uh, what was your take about the Terra Fab announcement?
>> Um, I think that, um, you know, for the ultimate future of, um, the companies, it's going to be determining, you know, how many hundreds or trillions or whatever, you know, that the we ultimately get to. So it is, you know, something that will determine, you know, being able to scale up 10x, 100x beyond what we do over the next couple years. But I think it will take, um, for the Terra Fab itself, it will take, um, at least three years to be making meaningful production. In the meantime, the announcements of being able to work closely with Micron, Samsung, and TSMC to deal with current near-term bottlenecks. That would be what will impact the financials of, um, Tesla and SpaceX.
Yeah. What did you take about that? Like he, he said a few times he repeated thanking Micron. Uh, and I'm looking for the actual quote, but he said something about how that they've committed to multi-year commitment at a very reasonable price point agreement that they did because Micron could basically charge whatever they want to charge anybody at this point. But they gave some allotment to Tesla, and uh, he was very thankful for that. What was your take on that?
So, so for things like, um, them getting chips from, um, Nvidia for Ruben or for Blackwell, each one of those chips, um, or at least the racks need like 288, um, gigabytes or whatever of of of, um, of memory. So, if I was, if they were to get like a 100,000, um, chips, Ruben chips from Nvidia, they wouldn't be able to make them into racks and actually use them without the memory, the high bandwidth memory from Micron or from, um, SDH Highix, right? That they need the memory, otherwise the chips themselves don't work. Similarly, the, um, hardware 4 getting a double memory version of hardware 4, they need more memory, right? So everything that they're doing currently is bottlenecked by, uh, how much memory they have, you know, in terms of like supply. It's more, it's like if I had trying to make cars, but I had no batteries, I need to have like having the engine, you know, the, the GPUs is great if I have no batteries, I can't make the car. So they must have the memory, and that's also ultra critical for them. Um, and then ultimately scaling up to 100 times. That's what they need Terra Fab for. So each one is critical supply chain. Critical supply chain now, critical supply chain in the future.
Did you, um, some people are saying that, you know, when they said that the Terra Fab is going to be this vertically integrated solution, right? You got memory included in that as well? Or is Tesla going to create their own memory, or are they buying it from Micron, or both?
>> Uh, I think, I think they'll do both.
It's just like, um, buying from CL batteries as well as having 4680s.
>> They need both in order to do it. And then the mix will be determined by,
you know, execution and and other things going forward.
>> Right? But, um, they're dealing with, if I don't have memory, if I don't have the chips, I cannot deploy things. So they'll use any and all memory, and they need to make their own as well. They need to have their own version of stuff. Uh, same for the, for the chips. They need all their own and all of everyone else's. So again, they'll buy from everyone. And the reason they can cut such great deals because if you want to scale in the future, if, if Micron looks past three years out to five years, then they say the power dynamic is going to flip, and then I need to be nice to you now because later you're the one who will have control.
Yeah, good point. You know, first of all, I think that's such a huge news that nobody's talking about yet, that Micron, you know, struck a deal with Tesla and they've locked them in because memory is the chips is the bottleneck right now. So, if you can't get it, you can't do anything.
>> Um,
that's right.
>> Tesla's announced that they've got it. So, what's your guess? Uh, Terra Fab is going to get a location announcement saying, not going to happen anytime, right? We think it's Grimes County in Texas, right? They're going to announce that.
But that it's going to own, have its own day in the spotlight. Is that it's going to have their own announce, uh, discussion about Terra Fab? My guess is that it won't be, they won't do this until they make the announcement that Tesla and SpaceX are going to merge, and they want to make a big deal because this is the golden example of the why they need to merge. They talked about the synergies. Terra Fab is number one that they talked about. So my guess would be if, if the most aggressive timeline is August 3rd is when SpaceX is going to do their announce, uh, earnings call.
>> August?
>> August 4th, that that's when they're going to announce that they're going to buy Tesla. And then immediately after that, they will do a Terra Fab presentation, or they will do it before August 4th to get everybody excited, to get the stock going up and all that. It'll be part of a kind of road show thing. So between, if they announce it then, and then it happens January, at some point in between then they'll do it. I don't think it'll necessarily be the before. I think they would do it two months into it, you know, as they, as things are kind of in motion, but, you know, anytime in between that you could do it right after, um, to get people, uh, excited, um, and also to, to understand why it has to happen before the vote, you know. So then, so the vote, if they're saying do it, the vote would probably be in a month or or six weeks after.
>> November.
>> Yeah. In November. November.
>> Yeah.
>> Yeah. So then in the late August, September, probably September would make sense when they do it.
Okay. All right. Let's talk D thought the Optimus. Uh, production lines are going in in Fremont. And they shared some photos and, you know, they're just trying to show you that the first production production line, first generation of production line is in Fremont and it's already here. You can take a look at this.
What was your take on all this? So, first of all, that they're showing you that this is not just, you know, they're not handmaking. A lot of this is automated already, right? But that it's getting, that it's getting its, uh, getting started.
>> Well, from your conversation with Lars, I thought he, he mentioned that there was they're vertically integrating to, um, make a lot of components, things like actuators and that kind of stuff. I thought he, he made the statement about like 40 different production lines.
I threw out a number. I don't know how accurate that is, but just that's a lot of line, sub-production lines to feed the main line.
>> Yeah.
>> Right. Right. So that looks like just the final assembly of the of the bot, and then they would do these other sub-assembly lines. So the reason they had to take out all of the S and X facilities was because it's all the vertical integration that they'll have to have. So this is just the beginning. They're just like, give us a taste of the overall big picture. You know, probably have some like Tesla Day that you get invited to that will, you know, you get the full tour of walking 40 lines or something like that.
So here's my theory. Tell me if if you think it's anything close to right. You know, the way Tesla works, uh, the way that they've launched FSD is they want to be generalized, right? When they create something, they want it to work any place on Earth. Any place you can put it in the desert, it'll work, right? Uh, and then they kind of shared that today or at the earnings call that they're going to launch it city by city right now, but eventually they won't launch it city by city. It'll just be like any, any city that wants to approve it, go for it, or they'll go all this entire state, go for it. They'll go launch it big, right?
>> I think they're doing the same thing with robots with Optimus. So my theory is that they are putting the first thousand human or robots, not as factory workers doing, you know, proving pick and place. They're putting it to the university to train. And then their goal, and I think they've repeated twice at the earnings call, is that you just show the Optimus a video of what you wanted to do once or you demo it to it once, and it's got it, and it'll do it from then on. Like there's one-shot learning, anything, any task.
That's what they're going to. So that's why I don't think that you're going to see that they're going to sell it to customers to do pick and place. That's not it. That's what other companies will do. These guys are training it. We now know that certain apps, I think the app showed that, um, they're there in there's code there that lets you, the user, to say, yes, I authorize my Optimus bot to share its video with Tesla. Why are they doing that? Because they're going to, they're going to train. Is that, uh, your thinking as well?
So in terms of the overall pace of improvement, I think the, um, critical aspect is what's happening with, um, the Grok and cursor systems, as well as certain aspects of what's happening with deployment of SpaceX Starlink. So Starlink is going into all CyberCaps, right? So we have communications with all CyberCaps. And then we're going to put up 10 times as many satellites, and they will have, um, 10 times the capacity or something like that. So then be like 100 times the overall capacity. So when you do that, things like, um, GPS, you can do GPS with Starlink satellites. The Starlink have that capability that you can back into a GPS that is actually more accurate and more robust than than the 30 or 40 GPS satellites that that we have for regular GPS, the the higher ones. So that means that you can get positional data, other data, just like you're in the factory. The communication, the information that's going, positional information that's happening of making it work in the warehouse, I can make it work around the world because of this extra capability that Starlink is making it like the world needed laboratory. And then this pace of how fast Optimus goes is when does Grok 5, Grok 6, Grok 7 come and get to full AGI, right? That it, you know, we're going to more and more levels of it. 4.5 caught up with, um, the Opus 4.8, 4.7 level, and then we're going to have a 4.6 is probably twice as fast in terms of tokens per second next month, and then I think Grok 5 will be there, and then Grok 5 gain to fable level, and then beyond fable level, everyone going to that in like six months with Ruben chips and stuff like that. So the, the scale of the AI going to insane capability will be tied together with the Optimus capabilities, like the quality of the hands. It's more the quality of the brain and the quality of the brain software that will matter. So it's not just thinking in terms of this in isolation, but this with the AI, uh, progress that we're seeing, um, and also the speed of it that they've done this hard coding so that, you know, they map out all points on the B300 chip in order to have faster training as well as faster inference performance where they said that they would go to, um, you know, double the tokens per second, maybe go from 80 to 160 tokens per second with this capability, and they'll map out every AI5 chip when they run it off that as well. So it's a broader look at what's happening with the AI is in order to see how that matters. Which also goes to why Tesla and SpaceX need to be joined because physical AI and, um, regular AI are completely overlapping.
Yeah. And then, and then what was your take on their comment? So there's there's digital Optimus, there's digital Grok that's going to power, like you were saying, the fleets, but then there's, um, there's, uh, what was, where was I going with this? Um, FSD is right, the they're using the same FSD, full self-driving. They called it full self-computer.
Um, but this, this will also be used for Optimus, right? So photos in using your vision,
and then controls out.
How real is that? Like, uh, I mean, Elon said, and I don't know if I can believe him when he said that all the other demos out there are just pre-coded demos. This is true, you know, intelligence, I guess.
Full autonomy, full actual movement.
Full autonomy. Do you really agree, believe that that's true? That just like a car had to first learn how to see the environment and then react to it? In the same way, a bot should be able to see what it's, there's an advantage there for Tesla.
So everything that they've been saying last, um, two earnings calls is they want to get digital Optimus out related to AI4. They're not talking about AI5. So Megapod is AI4. They're talking about doing AI4 in the, you know, million, two million cars that would be doing digital Optimus. So everything about digital Optimus is is basically roadmap.
Wait, no, but digital Optimus is macro hard, which is controlling a computer. I'm talking about, you know, how Ashok said, I'm using FSD that's built into cars, but we're using that also for controlling the robot.
Right. Right. Um, I think the FSD system is the best real-world visual navigation system going. So they're repurposing it for visually controlling a computer and for visually having a humanoid bot vehicle control itself moving around. So it's just the best vision, real world. So FSD is used for navigation of the bot, but then digital Optimus is used for, I guess, the brain and intelligence of what the context is.
Well, also the visual interpreting the the screen information they're getting, um, five minutes of of screen information. So that's like, um, 30 frames per second times, you know, like a hundred, um, time 300, you know, there's like 10,000 visual images that are being processed out. So they're processing huge amounts of visual information doing that and the interactions and reactions to it where I'm doing mouse clicks and typing and and all that kind of stuff. So it's, um, heavily visually dependent systems and and I think I would argue they, I would argue probably they would argue that they have the most advanced real-world vision system, you know, in the in the world with that.
Let's talk, uh, Tesla Semi. So here they said that volume production is not till next year, right? But that self-driving should come around the end of this year. But he said, we're prioritizing Model 3s, Model Ys, and CyberCap first. So this is Tesla Semi. It's going to be small volume. So let's say he said end of this year, maybe early next, he said. What do you say?
So, it looks like, you know, the, there's the, um, the need for mega chargers, right? In order to have your fleet out there, you need to have the charging systems to go with it. So, they are selling, um, some wall-mounted mega chargers. They are selling some a couple of ports. So, for the Pepsi like depots, you can make your charging and they can do local routes. But if I want to go from California to Florida, they need to build out about 60 mega chargers all across those states, right? So the infrastructure in order to have longhaul trucking is needed. And then because, um, the robo taxi robot trucking is mostly solved software-wise, then they would, the economic case for robo trucking is so much better, you know, five times, 10 times better than just saving on on gasoline with electricity. That's a $50,000 per year thing, but robo trucking is a $500,000 per year thing. So they can just, um, and also the economics become so good, they can just build them and buy them and use them themselves. So then if I can't convince you to use it, it won't matter because I just deploy it myself. So them going to scale with robo trucking, they have all kinds of optionality around that. So because of the timing of like, I got this, I will have the holy grail already, then, um, then I'll go scale. Then if I can't convince other people to buy it early, I'll just do it myself, you know, just like, um, Starlink, you know, there, they launched 89% of all satellites because they have their own economic model for that. So with, um, with trucking, you know, they will then for sure go to as many vehicles they want because if you don't buy it, I'll just do it myself. I got no worries. I can fund it easily because they'll be the profit just be insane. So, um, next year with robo trucking, then they will sell everyone they can make, whether themselves or whether someone else. It's just like they won't be worried about like, how do I sell it? How do I convince someone? It's just like the.
>> Yeah.
>> Yeah.
>> Yeah. My understanding is that the way he said it was, um, production of the Semi is on track, number one. And, uh, it, it's, and that it will be, I think they're shooting for 50,000 to be at scale sometime next year, the end of next year or later, right? And so it's going to be small production for the first for the rest of this year. And then once they do, they will start adding, uh, self-driving after they've solved self-driving for Cyber. Yeah. So the fact that they've already announced and got version 15 out for robo taxi, and then they've got the seven features that they need to solve finally for that. Let's say they end up by the end of this year, then they'll redirect. He didn't want to redirect the focus of the team, then they'll add it to Semi. Semi-driving itself is a given. And basically, what I took out of that is that Semi, as they start selling Semi next year, it will have self-driving as part of it or shortly after.
>> Yeah.
>> Right.
>> Okay. Uh, let's talk CyberCap. So what did you take about the milestones for CyberCap production? One thing that he did say was something like, didn't he say, um, we are timing our Cyber production to be kind of, uh, along the same lines as when, um, the roll out for robo taxi.
>> Right. So I'm seeing it as, um, because they're talking about doing statewide production. They have FSD15, which is having no errors, um, you know, no errors for no errors or accidents, you know, zero incidences for 38,000 miles. So 38,000 miles, no incidences. That's like a person driving for 20 years at 15, 16,000 miles with the vehicle with having noise. So it's, if they're beyond most human, at least at human level of safety, and so then they can go to scale. Which if they go to scale, say in Texas or Florida, then you would need 10,000, 20,000 robo taxis in order to to make that happen. So I view the timing as, you know, not just Austin, but all of Texas, all of Florida, version 15 with thousands of vehicles. That would be at some point, hopefully this year, if not Q1 of next year.
Do you have an estimate, or has your estimate changed for how many CyberCaps you think will be on robo taxi service by the end of this year?
>> Um, it's, um, if we're just doing normal scaling where we are now, it would, it would be a small number. We have to see a phase change where we do whole states for there to be a big number.
Um,
what's a big number? A thousand, 5,000, 10,000?
>> For me, a big number is, um, 20,000 to 40,000, you know, because that would be like all.
>> What's your guess? Yeah.
>> By the end of this year, I think, um, it's only, it's still not going fast enough. It's still only 20% chance of of any kind of big number. Otherwise, we're at sub 1,000.
Um, you know, they'd have to like take.
>> Sub 1,000 by the end of this year? My gosh.
>> Oh, sorry. Sub, sub, sub 5,000, 10,000. So, you know, a smaller number, you know. So, basically that, um, if we don't see the inflection like we need to, I think that the Gen chance for an inflection, um, for a phase one inflection where we get to dumping these thousands of vehicles out there. Um, but because they didn't, um, make a bold statement about it, um, I don't have huge confidence, you know, it's possible. That's fair.
I just, yeah. I don't know.
I've been saying 10 or 15,000. I've been saying 15,000 like several months ago, and I still think we're, we could be on track for that. Um, they made a thousand, probably a thousand CyberTrucks, CyberCabs. They, they said that we are testing at first in, uh, employees, CyberTruck, CyberCab in robo taxi service. They said that, um, you know, we need to get through these, these are milestones that gets us to, you know, CyberCap roll out into robo taxi, and that they will match the CyberCap production to, to, um, to what robo taxi needs are. So I think we're still going to get to the 10,000, 15,000 by the end of this year.
Um,
we need, need to see. I would think you need to see the 2,000, 3,000 by the by, uh, September, October, right? If you're going to get to that other number, you have to pass through that other thing a few months ahead of time.
No, except when we talked about it in our other show, right? The way it's probably going to go out is very, very slow. The next several months, like nine, you know, basically what we've been seeing. Then it's going to be like when they're ready, they're ready.
Okay.
Then they're going to put, they're probably going to put a, they probably have in a couple months, they would have two or three thousand made right now. They've already probably made over a thousand by today, right?
>> That's great.
>> Yeah.
>> In two or three months, they'll have two or three thousand. That's not hard to hit.
And then that's September. And then maybe by the end of this year, like that last month, they'll go 5,000. Boom.
Because that's right, that's that's what they're promising. It could be missed, but that's what they're saying.
Well, but it becomes like, um, a ship full of 5,000 vehicles where, you know, like when we do the the supply chain thing that we used to do before where it's like, hey, um, a ship of deliveries didn't make it for the quarter, so then it slipped into next quarter, right? So if you're depending upon that ship of 5,000 or three ships of 5,000 each landing in December, that ship doesn't come in, then you're looking at next month, right? So, so we want to see it earlier so we have confidence we're not going to miss.
>> The endpoint.
>> Yeah. If, if the end of this year is then the target, it could easily slip several months afterwards. Yep. I agreed with that. Good. What was the big milestones that you, you thought that, um, you were shocked and surprised that they talked about that, uh, they said this is going to happen by this date?
Um, I think that, um, they're talking about things that are not half measures. That, you know, it's like, it's going to be all robo trucking. It's not going to be, you know, Semi without robo trucking. It's going to be all robo taxi, all state. So it's that they're skipping the, you know, when they do go, it's going to be this shock and awe kind of thing, right? So it's on the other hand, it's tough for many people to believe it because they don't have a confidence-building ramp to know, oh, of course it can happen because they already have a thousand. Of course it can go 10,000, right? It's because we're starting off low bases that, um, it's, it's wonderful to believe. But everything they're saying is we're going to pull off, um, our caution and we're going to go all the way that we're guiding towards these, um, really bold things. Um, I think it's just because it's just like when we're talking about does it go to 20,000 by the end of the year or it goes to 20,000 Q1 because you have these break points that they're not, they also can't have the full confidence around it. It's just, it's still all going to happen. It's probably going to happen in 12 months, but they, you know, there's a bunch of variabilities around that that could cause the delays. So, it's, um, it's, it's like this, and it could be another month, another month, another month, but then when it does, it goes up, and then we don't know when that is. The one thing I will say is that I think Optimus is people might be expecting it sooner than it really is. I think everything I've heard so far is it's a lot slower than you think. So they have started production of Optimus in Fremont, but every one of those bots that they make, 10 or 20,000, whatever they make, this is Gen 3. They're there just to to learn. They're are going to go to the university. They're going to go to some sort of factory and they're going to all be doing a bunch of tasks inside their own factory. They might send a bunch home to employees and maybe some lucky people will get to take it at home and test it. But they'll be doing a lot of training. It could be a whole year, and then only when, uh, the the new version of version four, hopefully at that point, that that's the one that's going to go 10 million per year, but it's going to be several years from now.
Right. The other thing is that, um, people are many people fixated upon, you know, it must do certain house tasks, it must do certain other things. That I think the personal owned, um, Optimus, so long as it has AI5 chip, which would be, you know, like a pedlop of performance and that kind of thing, and has all this memory, I can have 20 times the performance for my local model of LLM. So, and it has, you know, be able to talk to it, all that kind of stuff. So the AI5 version of anything, robo taxi vehicle, Optimus, that by itself will be an in-demand thing because of the, you know, do you want a local edge model version of your thing? And then stacking those chips up into Megapods will be in demand. That, that the, the need to get high performance locally is something that will drive that, irregardless of how it uses hands. You know, it can walk around, it can do some stuff that will be secondary. And also the fact that I can do an over-the-air update and I'll upgrade and it'll do all kinds of wonderful stuff. So I'm not worried about the functionality needed to drive sales. The hardware envelope around it is efficient to combine with AI to make those things in high demand. So they will be able to make as many as they wanted on that, um, the 10 million. They'll be able to sell it just for the fact that you have.
And, and he, they could, they did confirm the AI5 chips will hit volume production around the middle of next year. And he, that was one of the things he was so confident about how great these chips are, that they're on track. He's so proud of the team. Uh, no one's going to have made AI6 better. But I think what you just said is the big thing. AI5 chips will be enough to do generalized learning. And what Tesla's trying to do with Optimus is train it to be able to do generalized learning. Not this task, that task, this task. It's like, no, I want you to have the brain to learn anything. Anything and everything. One shot. I show it to you once, you got it. You got it. You're going to learn this and you understand what it is we're trying to do, the concept, the current, and do it, you know.
And then the volume production of it is that, um, they won't be dependent upon any other supplier for chips. They will have their own memory. So then going to a billion bots within their power, all within their control, versus as good as people are envious about like the demos from Figure and for other companies, they don't own their own supply chain. Right. And that we see that from, you know, all the other competitors, you know, like, um, Lucid or something like that, you know, it's like, you don't have your supply chain, you don't have your batteries, you're at a disadvantage. You won't be able to compete.
Yeah, I think you're right. That is the most important thing. AI5 chips. Thank you so much, Brian. That was good. That was fun. Check him out. Go check out his website, Next Big Future. Nextbigfuture.com. Thanks, everybody.
>> Thank you.