StarShip Is Going to Make Tesla Stock Rocket

追加: 2026-05-24

[00:00:02]There was a time when I believed that the number one business opportunity by purchasing an Elon Musk product was 10 million or 20 million cars in 2030.

[00:00:14]That was I don't know three decades ago.

[00:00:17]I know it seems like it was a very very long time ago. But um then there was this thing called Roboaxi which was going to be the largest TAM in the history of the world. And then there was something called Optimus which was going to dwarf that by 7x and make the biggest TAM of all times. And then I got on the on the, you know, call with Brian Wong and he goes, "Well, I don't know. I think maybe Starlink, uh, even before we had the connectivity to the phones directly. That'll be even bigger than Tesla." And then we got the connectivity of the phones. It just keeps on keeping on. And now the numbers are absolutely crazy. And Phil Bicel kind of wrapped it up a couple of weeks ago with a theory and he's back to expand upon that theory today.

[00:01:06]So Phil, let me let me turn it over to you.

[00:01:09]>> Yeah. Cool. Um, yeah. By the way, I I love how you frame that because nothing's changed with the business TAMs that we that you discussed. It's not like they're shrinking over here. you know that road taxi still got its potential and and we'll probably meet that potential and more I think the same thing with Optimus.

[00:01:30]>> Yeah.

[00:01:31]>> And then we're just getting bigger and bigger all the time. So yeah. Well, okay. Let me uh let me do a little presentation.

[00:01:38]>> Okay.

[00:01:39]>> Uh this one was an easy one to build actually. I think you can see that.

[00:01:46]>> Yes, definitely. Definitely.

[00:01:49]>> Yeah. So, if you look at some people in the media, you'd find out that yesterday's Starship blew up and that's the end of that.

[00:02:03]>> Um, >> yeah, the Guardian comes to mind. I don't know. I I did I say their name out loud?

[00:02:09]>> There's There's several I saw.

[00:02:12]I I predicted it with a friend. I predicted it in a post and sure enough, they wrote it. They wrote it. They saw this big I don't know whether they're incredibly dumb or incredibly agenda driven or some combination thereof.

[00:02:27]>> Yeah. Yeah.

[00:02:29]>> But I don't think I care.

[00:02:31]>> So So this is a screenshot that I did while watching the insanely cool >> uh Starship mission yesterday, mission 12. Um, and uh, what this picture is is, uh, Starship, the top portion of the, you know, I guess we call the whole thing Starship, but booster did its job.

[00:02:54]It brought Starship to a somewhat orbital trajectory, went off and launched a bunch of fake satellites effectively to uh, show off their satellite injection PEZ dispenser.

[00:03:09]>> Mhm. And then it came to a oh not so soft landing in the Indian Ocean here.

[00:03:16]Uh actually the fact that we have this photo from small buoy boat that has a Starlink.

[00:03:23]It's like a little these little buoy um like little almost look like a kids pools or something. They're like little floaties, you know? I mean they're not little, they're pretty big. Um, and sitting inside it is and and they actually they're sort of autonomous.

[00:03:42]They run to a certain spot where they expect Starship to land in the Indian Ocean.

[00:03:49]>> Mhm.

[00:03:49]>> Uh, I mean, obviously the the launch vessel is, you know, a thousand feet away for God's sakes. It's not like these little things are traversing the ocean, but >> they get in position. Uh they have cameras and they have Starlink terminals and they take photos of what should be Starship landing in place and I believe they have a drone up as well. U so they get a top shot if people are wondering how did they get that ship coming down.

[00:04:16]It's because they think they flew a drone above it. But uh this actually is a picture perfect landing regardless of the insane explosion because that's just what it's going to do when it hits the water. But it landed right where it was supposed to land.

[00:04:34]And you know if we look at this mission in total I also got these launch photos.

[00:04:43]>> Fantastic.

[00:04:44]>> Uh it was look I I don't know. I mean I sat there I got chills. I felt like I was 10 again. Um, it was really, really awesome. And while not everything, not everything went perfectly, it was a roaring success, >> right?

[00:05:04]>> Uh, you know, almost an 11 out of 10. Uh booster did its job, but booster had an engine failure and then they just I think they didn't attempt to do the boost back and they just let it fall into the the Gulf of America as as it was going to anyway as >> it was never meant to be caught. uh by doing its job, what I meant is it brought Starship into its theoretical, you know, orbital insertion.

[00:05:36]>> Mhm.

[00:05:36]>> So it could do its job. And this was amazing because this is a this is the V3 series of of Starship where they've made over a hundred different changes and massive changes, architectural changes, the pad has changed, the both both booster and Starship have changed massively, the interiors, the way they had mounted the engines. It's Oh, and then on top of that, never mind, they're using the Raptor 3 engine, so right >> it's the third gener. I mean, this is like a, you know, I called it the kind of a hat-tick because it's kind of a everything was about V3. Everything was 3 I3 here.

[00:06:24]>> It was a great mission. it was I mean if you have a different characterization of it feel free to stop listening to me forever because I probably couldn't just couldn't agree with you on that. Um and I framed it this way. SpaceX IPO is up next. All systems are go for launch.

[00:06:46]I put it in I put it in put it in space terms T21 days in counting as of >> yesterday and uh I clearly indicated that yesterday's success will push us way into the two trillion mark you know I mean look it was a very high-risisk thing to do to try to do this a mere 21 days a rational man uh would have waited to 5 days after the IPO to do this perhaps. I don't know. Um but uh Elon has um Elon's in a hurry and uh he has a different brain than the rest of us and and so you know after after a week of standowns of uh Thursday scrubbed launch uh we got a launch on Friday and you know it's a test mission. It really is a test mission. like it's not there for show and tell. I mean, you know, you can't you can't avoid it. U you know, >> million plus viewers easily uh by the by oddly by the way at by by the as this as the SpaceX broadcast, you can see here this is 13 seconds to launch. There's 549,000 viewers. And I watched that number get bigger over time. And I'm thinking to myself, all these people coming online had missed the launch. I guess friends were telling them, "The thing's up, man.

[00:08:16]it on SpaceX.com. So, they'll have to go watch the uh the replay of it. The the launches, oh my god, it's something it's something to see.

[00:08:25]>> Yeah.

[00:08:26]>> Um >> so, uh what I wanted to do is talk about this photo.

[00:08:34]Uh this is a super interesting photo.

[00:08:36]This is a picture of Starship itself uh at 143 kilometers above Earth moving at 26,000 kilometers an hour. And how do you think it got the photo? H >> I have a theory.

[00:08:56]>> Well, I have a fact. Your theory might be my fact, but so what they did was using the PEZ dispenser, they launched uh I don't know about 20 22 satellites.

[00:09:11]And the last two are actually kind of effectively real. The other ones were just basically mass equivalent objects so that they could prove that they could, you know, mass and dimension equivalent objects for I guess uh Starlink uh V3 satellites.

[00:09:32]>> I think it's V3. Like I said, everything everything is threes here. So yeah, it probably is V3 um satellites. So the last two they called uh they had a name for them. Forgot the name. Uh, Dodger dog.

[00:09:46]>> Dodger. Yeah. Uh, but what they what what they had on them is they had cameras and uh and lighting. And so once you saw them, if you watch the replay as they are ejected, uh, you see them getting pushed out and you see the light come on, it kind of blinds the interior camera which had been inside the Starship at that time. And then later they show the pictures back at Starship, which is incredible because there's a lot going on here. It means that not only are those things deployed, not only are they powered up and active, but the comms to ground are operational, and they are relaying live views back looking at the ship that just launched them. And to me the this tells another story.

[00:10:41]Uh if you're looking at this you would realize that that the business of SpaceX is intact. You know the idea here is if you think of the mission it nothing more than you know you can look at the grand aspirations of of SpaceX and think going to Mars we're going to the moon.

[00:11:03]really the business is launching satellites for the near term, so to speak.

[00:11:07]>> Uh, and this tells you that they're going to be able to do that in mass. You know, this PEZ dispenser mechanism inside is a mass launcher of satellites.

[00:11:19]And it makes uh Falcon launches, which I think we're up to two or three a week, 20 satellites at a shot for Starlink, makes it look like small change. Mhm. And and the power of each one individually on the V3s, I forget the number, but it's multip with increases. Yeah. It's >> Anyway, story is you're you're looking at a a proof that the business is going to be pretty big.

[00:11:54]>> Yeah.

[00:11:55]>> So, let let's let's talk about that. Um, two weeks ago we did a show here and uh I had the headline was who controls the spice controls the universe and the idea behind that was that you know this comes from the Dune sci-fi series about you know where the spice is the most valuable thing in the universe and he who controls the spice controls the universe and I was doing a uh mapping to the business of of SpaceX and I said that that the spice is compute.

[00:12:37]So I wanted to talk about that. If we look at the um post I did on the left here, uh this was May 20th, uh the day that the SpaceX S1 filing dropped.

[00:12:57]And I said, you know, like there's a lot in there that's I don't know how many hundreds of pages and I'm sure you could put a link at Grock or Gemini and ask a thousand questions and get thousands of answers.

[00:13:11]But I wanted to simplify this greatly and I think I can.

[00:13:18]The business of SpaceX is compute.

[00:13:24]And let's even take Starlink and stick it in a little box. Now look, I think Starlink is a massive business. Massive.

[00:13:31]I think it'll be the largest subscriber product in the history of subscriber products.

[00:13:39]>> Okay.

[00:13:41]But I think it's small potatoes to the compute side of this business. And that's saying something because again >> emphasize Starlink is big, growing massively and will eventually grow into the mobile business and threaten the mobile carriers themselves. And that would be a nice business to have uh and they will have that business.

[00:14:07]But the point of this post is that what I what I'm trying to say is that we uh planet Earth, us Earthlings in every country are on a collision course with power for compute and chip supply.

[00:14:33]And we are growing AI at an tremendous rate.

[00:14:39]We discussed that in >> Mhm.

[00:14:41]>> the Spice episode.

[00:14:43]And it is growing so fast that there is no way today uh if you if you if you remove Elon Musk from planet earth, if you remove Tesla and SpaceX from planet earth, th this industry would crash into a wall very quickly.

[00:15:06]There is no way that Nvidia will be able to supply the chips it wants to supply to the growing AI market.

[00:15:15]The AI market is taking up so much compute demand and there's that famous chart. So, uh I have it in this presentation but I don't want to run it again because we did it in the spice episode. Um I I'll show you what it looks like so you could reference it. it is this chart that I built. Um but we are growing so quickly from a demand perspective that if you subtracted as I said Tesla and SpaceX from the planet and you were left with Taiwan Semi making chips Nvidia designing chips and selling these chips uh we would end up with are slamming our heads into a brick wall because the demand is so high. So, here's the So, here, let me let me let me interrupt here. Maybe you were going to get to this anyway, or maybe you maybe you at least comment on it. So, the number one uh backlash, if you will, or the number one thing that people talk about in terms of all this demand is they can't see it. And I will and I will admit, in fact, people that watch the show know that eight months ago, 10 months ago, Brian Wong and I would have lots of conversations. Maybe you and I had conversations to where I said, "Where's the demand? Are we sure we're going to have enough demand? Are you sure we're not going to get to a place where we don't, you know, end up with uh with way too many chips and way too way too much compute and we we can't and all of a sudden it it it all collapses around us. Now slowly but surely over that time I've come to add one thing after another to what I see as the demand um in terms of inference in terms of a lot of which will be at the edge but a lot of which will need clouds um but the thing that I I thought of this morning while just listening to you open this up Jensen Wong Elon Musk some of these other guys anthropic etc these guys are brainiacs And I'm still having trouble comprehending what it is because it's so exponential and it's so wild and out there. I'm just hoping I think these guys are, you know, really putting it on a spreadsheet and going this is how much compute we're going to need for this, this is how much compute we're gonna need for that. This is the likely improvement in terms of u the number of uh the amount of energy uh per um per token uh you know etc etc. And they've kind of got a spreadsheet on this and they're like you know it's it's at least for the next five years including Terraab and including um uh SpaceX we we we won't hit that number.

[00:18:08]We won't hit that end of the end of the need curve.

[00:18:14]Well, yes, we will.

[00:18:16]>> Oh, okay.

[00:18:17]>> I mean, we're Oh, in terms of like >> the need. Yeah, the need. Yeah.

[00:18:21]>> Yeah.

[00:18:22]>> Yeah. No, it it's there. The demand will grow as I predicted because it's what's it's what we've run up for the last three years in terms of growth. And if you look forward, you see no slowing down. I mean, look, you want to you want an existence proof of this. Enthropic steel with SpaceX is the perfect existence proof. Enthropic will be paying SpaceX uh what upwards of about 15 billion dollars a year to rent a data center.

[00:18:58]>> Yeah.

[00:18:59]>> Um and they're not doing that because they've got 15 billion dollars to spend.

[00:19:03]They're making money hand over fist, >> right?

[00:19:07]uh especially in their coding side of their business. Um so and by the way three years ago the coding side of this business was really minuscule >> right so yet it's sort of like you think about AI we've come through this this place where it started as an AI assistant from you know chat GDP got used to that we college kids were running around you know doing their college papers with this thing and and now we've matured and figured out massively into these different spaces where AI compute is being put to work to do you know it's completely changed the entire software development industry >> in the last two or three years and it will continue to change that very industry but it will change all industries massively and we are just on the cusp of full generative what I called pixel a AI you know you know it's one thing to generate words and code which are tokenized represent you know it's that there's a tokenized represent representation of words and it's small data by comparison when you start generating pixelbased anything dynamic commercials dynamic movies dynamic you know any type of video or or or still um you're talking about, you know, commute compute demands that absolutely sore.

[00:20:44]>> Mhm.

[00:20:44]>> And then you've got agentic kind of compute coming online where we where what we had before was user sat down and typed in a question and chat and yes, they waited for maybe 30 seconds to a minute. In other words, it was sort of an agent going off and doing something on your behalf, but it wasn't running overnight.

[00:21:07]you know, it wasn't monitoring bank accounts or monitoring markets or monitoring uh chemical plants or, you know, things that are starting to happen.

[00:21:23]>> So, all of that tells me that the demand for compute for AI is is is skyrocketing.

[00:21:32]That chart I showed explains it fairly well. Mhm. Mhm. And so let's go back to what SpaceX is about. SpaceX, if we get rid of all the other businesses of SpaceX. Let's forget about the launch business. Seems strange, right? They do rockets. Why would we forget about the launch business? For the sake of argument, we're going to forget about the launch business. We're not going to worry about whether they do la NASA launches or launches for commercial satellites or military satellites. We're just going to put that in a little teeny box in the corner and say that's small ball. Okay.

[00:22:13]>> We're even going to take Starlink and put it in a little box. Although it's a much bigger box as I as I said. I believe it's massive. But but we're going to talk about compute and uh because of this demand and because we have we have two things that constrict this demand massive uh from from you know in other words from realizing the supply side of this. We have two constrictions two two factors that that stop us in our tracks. We cannot build ships fast enough to meet that demand and we certainly can't power them. Mhm.

[00:22:48]>> And I mean we certainly can't power them. So if you go to my chart without you know let me go to the top line of that chart. It computes that we uh planet earth will run we are running today at about 30 gawatts of AI compute around planet earth. Okay. All things that we talked about every box of AI that you can think of is about 30 gigawatts in under five years we'll hit a terowatt okay and to put that in perspective terowatt is twice what the US consumes on a daily basis right we we run at about a half a terowatt so could you imagine that AI requires a terowatt of compute That's power on these chips to do all the stuff we just talked about.

[00:23:48]And somehow magically we were supposed to uh take our grid that's growing at about 3% a year and was always historically designed for that kind of growth if you will and then suddenly say well we really in five years we need to get it up to the point where it can spill off power alone just for AI never mind >> right >> the the average growth for I mean the growth for everything else that electricity consumes. So guess what? You can't do it. And there's only one place to do it and that's space. And that's why I'm wearing my lovely uh SpaceX always sunny in space. It's the um this came from you know the terra fab announcement in March. Uh and the reason is that uh you know what we can do in space with is we can inject hundreds of thousands to millions of satellites. they can collectively become a data center.

[00:24:49]Uh they have free power effectively beyond the >> the satellite cost itself but on you know on an incremental basis it's all free energy >> and it's available energy. That's the other part. It's available. It's not like you know if you could pay for it on planet Earth could you get it? No.

[00:25:06]That's the problem. You can't even get it. um China won't even be able to produce it if they if they're growing at the rate they are with solar coverage and nuclear and you know buying more coal and burning more coal they're they're just not going to be able to to do it themselves.

[00:25:25]I when I wrote this post uh I thought what I was trying to do is I was trying to say like if you want to understand SpaceX you could read my Terra Space article front and center here which I quote posted it it's a great framing of the big picture >> sort of like what I call the SpaceX master plan >> it has a lot of pie in the sky stuff it talks about the future blah blah blah. And then I after I wrote it, I thought, "This is great. I love it. It makes me understanding." And then I had a couple people read it and they said, "Yeah, but that's out there type of thing. What is it about? Is it about doing missions for NASA?" And I'm like, "No, my god, no. That's small ball stuff. It's it's about compute.

[00:26:17]It's about building this orbital data center."

[00:26:20]And the interesting thing about it is, you know, it can grow it to almost any size. It'll start in the terowatt range.

[00:26:29]It'll eventually move into the ped watt range. Um, and uh yesterday what we saw was the railroad to space. We saw the launch vehicle that will I mean that's why I specifically uh chose to put this photo up because this is the whole business of SpaceX.

[00:26:53]It's the fact that it's there to inject satellites into space, these compute satellites.

[00:27:01]>> Now, in this particular case, it's it's Starlink satellites, but same same thing.

[00:27:07]Um, and so I wrote this post and I thought, let's get down to where it is. Let's get down to like, let's throw everything else out and let's just talk about compute.

[00:27:22]And this was up for less than an hour and our good friend CERN liked it and a bunch of other people liked it and then Elon Musk liked it.

[00:27:31]you know, and I'm not here to break news about that. Kind of keeping it between us and our listeners, but yeah, the point is he gets it. This is what he's talking about. This is what it's about.

[00:27:44]Um, and so, uh, I think this framing tells you why this is a massive business. I I think that, you know, a massive opportunity. SpaceX has a massive opportunity. What's what's super interesting about it too is a lot of when we when we first I think and we talked about this I think a couple times Randy is that we said when Elon first did the Terrafab announcement um it made it feel like the chip production he was he was saying I need chips for my products >> uh what I call insular use cases you know inside the walls type of thing. And what were those primary chip consumers?

[00:28:33]Well, robo taxi uh in every vehicle, you know, two chips per for inference uh for to to to run a FSD, which is big and will be huge.

[00:28:49]And then AI5++ for Optimus, which will be bigger. I mean, that's a massive demand driver. And by the way, those are two pieces of AI that we didn't even frame in the original part of this discussion talking about use cases. The embodied AI >> that's coming.

[00:29:11]>> Uh, this is different than, you know, it's like, well, everything we're using right now seems to be in a data center.

[00:29:16]Well, this stuff is out about around us, the robotics.

[00:29:20]Um, I mean, it may not be clear. Every car is a robot, right? and it's running around doing this thing and and Optimus clearly is a robot. So, um, but it but it made it seem like when we heard Terapab that what Elon was saying is, "Yeah, I need to make chips for my stuff >> at Tesla and uh I talked to, you know, Taiwan Semi and I talked to Samsung and while I love the guys and they're growing fast, they're not going to be able to cut it." And that's a true statement. I mean that is >> but his eye was on a larger larger game here and it was the non-inssular use case which I mean what I'm trying to say is if if there is no Grock in the future if he shuts it down if he decides I'm not doing Grock um and Anthropic's frontier model and Google's frontier model win, you know, or maybe one of them wins or they both win or maybe Open AI wins.

[00:30:29]>> Guess what? Anthropic, Google, Open AAI, yes, even Open AI are going to be SpaceX customers.

[00:30:37]>> Uh the the when we saw Anthropic do this deal, that's another funny thing. You know, I heard it referred to by many as, you know, Elon Web Services, kind of a play on Amazon Web Services, >> right? Right. And I thought, no, that's the wrong way to frame it. It is not.

[00:30:56]The enthropic deal of renting colossus tells us that anthropic is running into its own brick wall and desperately needs compute. And the fact that it happens to be terrestrial at the moment, it happens to be in a data center in Memphis is irrelevant.

[00:31:18]If this orbital data center was operational, that's the data center that would be providing the inference for coding and everything else. Sure, >> that's the data center. And at that point, he can shut down Colossus or keep them running or whatever. But to me, it's what I called >> the prototype of the business. It's like building the terra fab research center to produce some chips or to build the optimus um million plus proto lab factory in Fremont. Colossus one represents the prototype of compute >> for this for for SpaceX.

[00:32:03]So that you know fundamentally is how I'd frame it. And I would say that what what what we have is given where I think AI grows over the next decade.

[00:32:20]Uh I mean look at my chart. It will blow your mind. I I I wake up at night thinking about that chart because I keep checking the math in the chart and going this is insane.

[00:32:30]>> Um you know Elon will control the spice.

[00:32:34]He will control the spice through this company. They're there. Everybody, Google, Anthropic, even his sworn enemy, OpenAI, Meta, uh, Apple, Microsoft, they'll all be knocking at his door renting compute in space. Well, Google and Google Google already has, you know, they're they have not I they haven't announced a deal yet, but they're in negotiations right now.

[00:33:06]>> I suspect that Google will announce a massive deal with SpaceX.

[00:33:11]>> And, you know, it is clear to me that these guys have nowhere to run. You can't build a data center. Look, if you could get permitting, if you could go to your reddest of red states that were aggressively asking for your business and you were to tell and they were not only aggressive, but they said, "Well, we could supply you the power." The answer is no. They can't supply you the power.

[00:33:38]>> If they and then even if they could do that, where would the chips come from?

[00:33:44]Well, guess what? They're going to come from Terraab.

[00:33:47]they're going to come from Terrafab because Terapab is the only chip production possibility out there that could satisfy this demand. If we look at this chart here, this is the the green area here.

[00:34:06]You know, I didn't want to do this, but I will. Let me just do this chart really quick because it, you know, let's repeat the chart slightly. The blue line is where demand is growing. Mhm. 3.4x a year. The line below it, which is the yellow line, represents compute efficiencies through hardware.

[00:34:27]So, in other words, take your take your your your blue line and drop it down. So, you know, effectively pushing demand down.

[00:34:38]>> Um, so take a look at the green dotted line at the bottom. Mhm.

[00:34:44]>> That's what's available if Terrafab never shows up.

[00:34:46]>> Right.

[00:34:48]>> So that includes what Taiwan Semi's growth plans are and they are, you know, 90% of compute delivery.

[00:34:57]Um, if Terrafab shows up and begins doing its job, uh, I don't really like where that label Terapab ramp begins because clearly the ramp begins at the beginning of the slope there. But terap will will almost catch the line and you know it's in other words the ideal is that that that green space swamps out the red shading.

[00:35:20]>> Yeah.

[00:35:21]>> Um terraab will almost do it which indicates something obvious to anybody looking at this. There won't be just one terapab there'll be many terapabs. You know I mean these are terapab is massive. The the idea behind Terapab is that Elon is suggesting that this factory will will build uh AI chips that will satisfy one terowatt of compute annually.

[00:35:51]>> Mhm.

[00:35:53]And should he do that in what would be the largest manufacturing facility in the world? he will match that green line there.

[00:36:08]And you can see it falls off.

[00:36:11]And the reason it falls off is because it's not like demand doesn't keep going.

[00:36:16]Uh, by the way, the axis on the left is logarithmic. It's not linear.

[00:36:21]>> Mhm.

[00:36:22]>> So, what that indicates clearly is that once he gets one terapab going, he'll want another and he'll want another and he'll want another. um and he will be in a position to make them at that point you know >> yeah and to the earlier conversation uh there are two places where investors are asking the question about Nvidia about Tesla about SpaceX in terms of demand and you know we talked about it a minute ago and there's one one question is whether the demand is there for continued growth at all and then the second question is whether there's enough demand for the exponential growth growth and I think what your chart is showing even with Terraab is that if there is enough demand for that kind of for the kind of exponential growth that we're getting right now probably even Elon is going to fall behind unless there are dramatic breakthroughs in terms of efficiencies and and there could be >> yeah there could be for sure >> very dramatic increases in efficiency but I did wanna I did want to kind of sit on this for a minute because I don't know how many people are thinking about it, but I've been thinking about it a lot because I'm a manufacturer, you know, I think in terms of demand. And what what Elon also said in the S1 is that the big driver the big the number one um TAM now which is multiples of Starlink and multiples of Wi-Fi connectedness to your phone multiples of of the the use uh by you know of just setting launching satellites into space is enterprise.

[00:38:06]Well, I know something about enterprise.

[00:38:09]And so I think about just my little enterprise.

[00:38:12]Every single workstation would have a camera on it.

[00:38:16]Maybe two, maybe three, maybe four.

[00:38:18]Maybe the operator, if they're still human operators, or if it's a robot, it'll have cameras on it. All of these motions of the various equipment, the various operators, the valves, the pressures, um the incoming electricity, the incoming uh air pressure, the incoming gas, whatever it is, all of those are going to be monitored 24 well anytime that machine is on, anytime that equipment is running.

[00:38:47]>> And that's just in manufacturing. Then you get into >> government, which is an enterprise. And then you get into the services industries which are enterprises.

[00:38:56]>> Yeah.

[00:38:57]>> And all of that data from all of those different organizations has to be processed. Um and this is where this is where Elon apparently can see by graphing it out as to what he thinks is going to be the future's going to look like that it's $25 trillion a year.

[00:39:14]Yeah.

[00:39:15]>> Business. Yeah.

[00:39:16]>> Yeah. that that box is nothing more than saying we have a lot of data to compute >> right >> that's what it's saying you know whether it's in a like I said earlier in a chemical plant in a manufacturing of widgets a manufacturing of pies or cakes >> it's in a financial industry >> right >> um any service industry analyzing you know if you think about it Today you have this opportunity with AI to do stuff that you could never >> right >> do before.

[00:39:54]And since your competitors will start doing it, you're guaranteed to do it.

[00:39:59]>> In other words, if you're >> or gone.

[00:40:01]>> What's that?

[00:40:02]>> Or be gone.

[00:40:03]>> Yeah. Or be gone. I mean, look, if you're if you're Starbucks, you'll be analyzing your data dimensionally in dimensionally different ways than you ever have before. Aentic AI will be running through every point of sale transaction, every thing that you could imagine. Now, I'm not suggesting by any measure that I look forward to that day because we know how these companies become, you know, marketing hounds and uh, you know, generally piss us off as consumers. Maybe maybe AI will allow them to uh have a new place and not piss us off. I don't know.

[00:40:48]>> I don't know. You can drop back a little bit. I happen to know a little bit about the inner inner workings of In-N-Out. My son worked there for five years. in it.

[00:40:55]For those of you on the east coast, In-N-Out is the hamburger place in the West Coast and now half the country. Um, and In-N-Out has been doing all this data collection for a very, very long time. They know exactly how many people are going to go past that window per hour. They know exactly how at the beginning of the day, based on historic information and what happened yesterday and what the news is saying and what the weather is and all this other data, they know exactly how many hamburgers they're going to use that day. And I mean they get it down to the inch. Um and uh they know how many buns etc etc. They don't know how many employees they need to put on the line every day. And the reason you get such consistent service at In-N-Out and the reason people want to go there for that consistent service is because of the data collection that they do and yet they're still an absolutely friendly company.

[00:41:42]>> Right.

[00:41:43]>> Yeah. I'm not just I'm not suggesting doom and gloom because of that. I just uh because I'm no fan of Starbucks. I guess I took a negative on it. But yeah, I mean my point is if if Starbucks is now analyzing data, agentic AI running massive com, you know, AI processes 24/7, 365, you damn well we'll bet that Dunkin Donuts is doing it and every other service business will take a page from that book. I mean, obviously to your point, companies like In-N-Out are doing it today. they're just doing it kind of in um you know that look it's we came through this error the era of big data we called it big data where we were saying okay for the first time ever you know it's essentially the the chapter I wrote in our AI rising book >> the first time ever we now have the capacity I mean let's go back you know kind of on a historical timeline 2015 16 17 This is when we use this term big data and we were we were saying this is the first time ever that data produced could be data stored and potentially data processed.

[00:42:58]>> Okay. And we started from a computer science or a kind of architectural point of view or it you know however you want to couch it. We started building systems that distributed systems that could store the data in a distributed fabric and could process the data on to that distributed you know from that distributed fabric and it was largely my job at Rivian was you know I was own the big data job move data from our vehicles our IoT devices known as vehicles which are very prolific at doing generating data to say the least understatement of the year and >> and bring it to the data center and store it and then process it. Now in that box of big data we couldn't our processing before AI was very limited and what I mean by limited was you know it's it's kind of using traditional analytics methods >> right >> that nothing wrong with that but certainly not the type of thing where you can look at a data set and ask it a question that was not ever kind of planned out before, right? So, for example, imagine that you're the CEO of Rivian and you sit in front of um uh something like, you know, like a chatbot similar to um OpenAI or Grock >> and you want to ask questions about your vehicles in terms of how your customers are using them. Well, if you tried to do that a couple years ago, you'd basically be asking effectively your IT organization, produce a report that does this, >> right? Because I'm curious. Or you would go to, you know, somebody that's doing this information side and say, "Hey, is there a way, do we have the data that could tell us like, you know, how many this that and the other things happen with this kind of customer base or whatever the question is. I can't come up with with >> like it could be it could be how many times does does our car do our cars uh end up in some kind of a detour? How many times how many times do we have uh uh accidents in an intersection that cause uh cities to bring out the cops to direct traffic around it? You know, the kinds of things that are create a lot of these edge cases.

[00:45:33]>> Yeah. And and crazy crazy crazy, you know, open-ended questions to things that you could only dream of like, you know, you get to the end of the you say, "Well, tell me about our accident rate on our vehicles." Yeah. Categorize the accidents. and you get to the finish line of that and somebody's produced this beautiful report and then you say, "Can you categorize it by the color of the car and the the system would be like, huh, >> got to go back."

[00:45:58]>> No, we got to go back and rearrange and reindex the data to be able to produce something like that. Now, with an AI, we can go to this fundamentally large box of data. And when I say large, I mean the amount of data that comes off of of even Rivian's small relatively small fleet per day, probably per hour is larger than all the data that chat GDP has encoded for the purposes of of its current AI.

[00:46:32]I guarantee that's true.

[00:46:34]So, so for the first time, we've been able to bring that data to a data center. Not all of it. Believe me, there's too much of it to really still we're still operationally incapable. We don't have networks to be able to transfer that. We don't have enough storage that would be cost effective to bring it. But for the first time, we're seeing a lens into doing that. And now we're seeing a way to look at that data in any which way we want to look at that data. And even further, we're able to ask the AI, do discovery.

[00:47:09]>> Like, I'm not going to ask the questions. You ask the questions. You tell me something I don't know. Tomorrow morning, every morning, I'm the CEO of Rivian. I'm the CEO of Tesla. every morning tell me something factual that happened uh or something interesting or some assertion of something uh based on fleet data in the last 24 hours or or through the last you know so now oh you know hey guess what uh blue cars are selling really well now because you know people blah blah blah whatever um >> so you bet your bottom dollar as they would say that agentic AI is going to fuel up and and get heavier because now we can let these systems u go at data in non-uniform you know report-based ways and we can let these systems go at data on their own without us getting in there and have it come up with its own ideas of what's like it should be able to you know at some point our AIS will be looking at a chemical plant And one morning it will say to somebody in the operations uh this plant is built wrong.

[00:48:30]Not not that this valve is over pressuring this way or something like that. No this plant is fundamentally built wrong. If we rearrange where we put we bring the inputs here or uh we would uh reduce the incident rate of you know XYZ I mean that type of analysis which is very different >> um insight analysis you might call it.

[00:48:55]So all this leads to one like conclusion. Um the demand will continue to grow the way I think that demand curve suggests in my chart. Um it will outstrip supply in both chip and compute.

[00:49:13]>> Mhm. And the satis the the the supplier the only supplier on this earth that will be able to fulfill that potential uh be possibly able to fulfill that potential right I don't know what the timeline looks like will be SpaceX and we can add Tesla because >> what I mean by that is we're not I'm not I'm kind of in a way focusing on SpaceX because SpaceX will bring us its business will compute in space, right?

[00:49:46]Is that the that that large box of that big TAM, so to speak. But >> but but Tesla has skin in this game big time, right, with the chip production side, uh for sure. And uh so I don't, you know, exclude them in some kind of like, oh, they're left behind. No, they're part of this they're part of this this whole thing. Uh they will they will uh they'll be pulled along by this.

[00:50:10]But I guess the point is >> Elon controls the spice.

[00:50:14]>> Yeah. Yeah.

[00:50:14]>> That's the outcome of this, right? And and and I think that the only thing that will get in the way of this conversation um like we're going to have this IPO, >> okay? Uh I don't think any one of us knows where it will close in the first day and I don't think anyone knows where it will close you know month after but I will make a bet and I will make a bet.

[00:50:45]>> Yeah, you will make a bet. Yeah.

[00:50:46]>> Yeah. That on a 10-year timeline, you know, I'm not talking about like I don't I'm not going to get into the Wall Street year. What's a year going to look like?

[00:50:57]you know um this is what you know I know look terab isn't going to show up tomorrow computing space isn't going to show up tomorrow these things are going to take time >> um I mean heck we just saw the first reality of that railroad to space yesterday >> so give it time don't think I'm being aggressively optimistic on this these things aren't going to come in the way we think >> I think I think I think realistically you've got to give us a year to get a reliable Starship.

[00:51:33]>> Yes.

[00:51:33]>> Going to do six or seven, eight more launches before they start putting up actual product. Then you've got probably a minimum of two years, maybe it'll be a year and a half before you start putting some Nvidia chips into space or maybe some Grock I'm sorry, sorry, some um some Dojo chips into space kind of on a test basis starting to do a little bit of what can happen. So that could be a year and a half or two years. Then you got two and a half to three years.

[00:52:02]>> Yeah.

[00:52:02]>> If Elon is right, which he never is on time. um you know where where uh the Terapab might start producing some actual chips. So you know there's these are long timelines.

[00:52:14]>> Yeah, I think I think um you know I I suspect that late 27 sometime early 28 there'll be some prototype compute satellites launched for sure. I and I don't I don't care whether they'll launch on Starship or Falcon. It really doesn't matter. Like you gota you can move some things around here, right? Um, like you said, they don't necessarily have to be Terrafab produced chips.

[00:52:38]>> No.

[00:52:38]>> Um, they could just simply be AI5 chips produced by Samsung in in Texas, >> for example. Right.

[00:52:47]>> Um, they have the knowhow to design and I'm absolutely certain, beyond certain that they are developing their compute-based satellites as we speak.

[00:52:57]>> Oh, absolutely.

[00:52:58]>> Yeah. I mean, that's like ridiculously uh with great certainty. So, so they will they will trial balloon in some prototype way what it means to put compute in space. Um, but you can see that all the pieces will eventually line up, right? To your point, Starship.

[00:53:18]Well, to make it economically viable, we need Starship reus we need booster re reusability, not necessarily Starship itself reusability, but they need to get to reusability of of both both pieces of of of of that rocket.

[00:53:33]>> Mhm. And I do agree that it's, you know, six to 10 launches away probably from uh getting to the point where that PEZ dispenser will take the risk and actually it probably be be before take the risk and probably push out, you know, like you saw that stacking where two of the satellites were real. Mhm.

[00:53:55]>> I I suspect maybe in the next goround they'll say, "What the heck? Let's try to launch um 10 actual >> uh star Starlink satellites. Why not?"

[00:54:08]In other words, we they it might crash and burn. We might lose those satellites, but we we we'll we'll go for it. Um but then yes, so what we'll do is we'll get reusability on the on the railroad, right? We'll get to we'll get to the point where the railroad operates. We have to get to the point where Terra Fab produces that's a very very hard enterprise. I suspect that that will fall behind most people's optimistic version of that. Um I'm less concerned that compute in space will operate. I don't think the risk there is what people think it is at all. At all. I think if any company on earth understands building satellites at scale, operating satellites at scale and all the complexity associated with those, it would be SpaceX, >> right?

[00:55:05]>> You're looking, you can look at Starlink as the prototype with 10,000 orbital satellites far outnumbers every other type of satellite combined.

[00:55:15]if you don't think they know how to do that. And by the way, it's not just operating them, which is a big deal. If you watch the little promo video they did during the Starship launch where they talked about operating this network, it is not a trivial thing.

[00:55:28]They're looking, you know, collisions and all this stuff, >> but they don't really even talk deeply about the manufacturing side, which is a big deal. Mhm.

[00:55:36]>> So, so this company called SpaceX has the deep experience to be able to build these compute satellites and they only have a couple little problems to solve and they're actually not as big as most people think. So, so I believe that compute-based satellites isn't the fundamental problem. I think what will hold them back potentially is producing the chips, the terrafab problem, and producing enough launch sites to deal with the launch cadence.

[00:56:09]You know, I think that's the thing that we we saw um SpaceX uh potentially building out a property in um Louisiana for launching Starship. Mhm.

[00:56:21]>> And uh in the last two weeks of of the three posts that Elon liked to mine, one of the other ones was where I talked about, you know, the the launch cadence required to build a million plus uh orbital data millions plus satellite orbital data center >> is, you know, somewhere in order of three to five launches per day of Starship. I mean, heck, we just had one yesterday. We need to do that three to five times a day every day. And we need to get to a place where, you know, one of the things you saw yesterday in that launch is like they were talking about booster falling into into the Gulf of America, you know, >> and the gentleman that was announcing was saying, you know, hey, don't worry.

[00:57:11]When we do these launches, we clear land, sea, and air.

[00:57:17]Well, guess what? You can't do that all the time for all launches forever, forever, forever. I mean, booster will have to come back to his launch pads.

[00:57:25]It will have to be caught because it needs that reusability and we'll get there, you know, >> right?

[00:57:32]>> Um, but we will see launchpads potentially blowing up when a catch goes bad. I mean, we have great reliability with Falcon. I expect SpaceX will I mean amazing reliability with with with Falcon. I think SpaceX SpaceX wants Elon wants to get there with Starship.

[00:57:54]>> Sure.

[00:57:55]>> They will get there without a doubt. We will not be worrying about >> when we get to three or five launches today. Nothing will seem dramatic. It will seem boring. I mean, did anybody, by the way, three days ago watch the um Falcon launch Starlink mission so and so out of Vandenberg? Well, actually I did, but I guarantee almost nobody did because it was as boring as, you know, watching a train run by. So >> that's still and for a lot of people that's still pretty cool. And I still still still see people going to the outside of airports and watching planes la.

[00:58:34]But but it's a very it's a very small diminishing number.

[00:58:39]>> It's funny. My uh my youngest daughter is home from school and uh from college this summer and she's here and and I made her watch the Starlink launch the other day from Vandenberg and she thought I was cool, you know, but she's like, "Yeah, Dad, that's yeah, you know, like you know, and and what is that launching?" And I told her, you know, you know that Starlink terminal I have, honey, well, those are the satellites that and she's like, "Yeah, and so what does that do for me?" And I go, "Well, it brings you your internet, but you know, this is all kind of like just part of it." But but yesterday, I put her butt in front of the TV to watch Starship and she did say, "Oh my god, that is cool." So that, you know, it's new, it's novel, it's different. Um but but in order to get to this business we're talking about, it needs to be anything but new, novel, and different.

[00:59:30]It needs to be >> boring and obvious.

[00:59:35]Um and so you know so so it will get there. It will get there. They're on they're they're definitely on a path to get there. Um and yesterday I felt like they are making the you know it's interesting. You see V3 and you go why don't they just settle on V2 not move to the whole V3 infrastructure.

[00:59:56]Well, it's easy, you know, we could we could have a whole show on why they're moving to a new platform.

[01:00:02]>> Mhm.

[01:00:03]>> Why they need to do this for the actual business of reusability?

[01:00:08]>> Uh, and some people say, well, Elon keeps talking about V4, you know, is he going to Well, I don't know where he gets to the point where he calls it a day, meaning like he says, put a stake in the ground. No, I mean like what I mean by that is Falcon level business, right? or Falcon.

[01:00:24]>> You know, Falcon's a platform at this point. Yes, they've made changes to Falcon, but we and because of the rather mundane nature of of their launch business now with it, we don't see those changes. Like, we don't hear, "Oh, guess what? We've got a V3." I mean, I can't even tell you the changes made on Falcon, but yeah, I know they're there.

[01:00:45]Um but you know the point is uh the this kind of massive changes will slow down.

[01:00:53]They'll focus on the systems reliable reusability. They have to focus on kind of launch facility and production facility stuff. That's why they brought that up about what they're doing at Canaveral. what they're doing, >> right?

[01:01:13]>> You know, because they have to get to where this is where this is a launch business where they can do uh multiple shots per day.

[01:01:21]>> Yeah. Yeah.

[01:01:22]>> Yeah.

[01:01:23]>> So, I mean, this all comes down to say that I would guess that the IPO will shoot out of the gate hot, heavy, and all the things that people thought. It may settle back. Who knows? Um, I guarantee that there'll be these um, beautiful opportunity days I call red days. You know what the red days will look like? There'll be the ones where they try a starship and it blows up on the pad >> and uh, and because people don't people don't understand that process.

[01:01:56]Yeah.

[01:01:57]>> You know, the stock will go on go on sale be off 20% that day and that's your chance people. You know, >> that'll be the day that you >> because because that >> that thing blowing up on the pad is only fuel for the fire of that team to make it work.

[01:02:14]>> Yeah.

[01:02:15]>> So much. So, so true. So, >> that's all I got.

[01:02:17]>> So, so that's the second spiciest program you've ever done for me.

[01:02:23]>> Oh, it's also the second. Anyway, um so I think uh spice uh I I don't know whether people realize I think most people do that, you know, spice was maybe the very first thing that created international trade.

[01:02:39]>> Yeah.

[01:02:40]>> Um and uh became very very very important and so he who owned the spice in those days >> Yeah. uh was actually you know running the running certainly creating the the most amount of of uh of dollar profits in the world. So uh we've got the spice of that time we've got the spice of uh future times and this will be compute as uh as you so correctly point out. Uh so Phil Bisel as I freeze up. Anyway, there we go. I think I'm back. Um, as always, great to have you on to explain this stuff. And I think that was very, very helpful. And >> I really want to get into I somewhere along the line, I want to try to get some kind of a mathematical look at this at this demand that Elon basically just announced. I mean, we were not talking about $22 trillion a year of enterprise compute um before whatever two days ago. So, this is this is a this is a new idea and I want to I want to really dig into that one, but we'll do that on yet another day. And so, thanks again, Phil Bicil. Thank you all and it's been great talking to you.