Transcription
I'm Randy Kirk and I'm here with Brian Wong. And Brian Wong, of course, is our resident science guy. Uh, and also futurist. I'm like a futurist small F. He's futurist big F. And, uh, we are both actually celebrating this week. Did you know? Do Do you know what we're celebrating about, Brian?
Um, celebrating the book, celebrating early Christmas. I I don't I think what we're celebrating is the fact that you and I uh you know even even Larry Goldberg mentioned it in the show that he's either been a couple of hours ago or will be a couple of hours from now. You know, Larry says, "Well, sounds to me like you and Brian are right again with regard to where the stock is going to be." And uh so let's get into that right away. Let's talk to me about what happened to the stock market [laughter] today. I mean, oh my gosh. And what happened to Tesla and why are we now finally uh solidly in this 250 range and what happens next? A 450 range.
I'm sorry I do that.
450 range. Yeah. Yeah. Um so it's looking like um you know we've been holding steady and flat for a couple months. Um you know hitting this 460 470 480 levels and then but the lows have been getting higher on that. So, it looks like we're setting up for a breakout. Um, you know, these kind of technical structures looks like we're a breakout. And then that align that technical stuff aligns with a um um the fundamentals which is that we Elon says we'll remove the safety monitors in um Austin in like 18 days. That December 30th would be roughly when we would achieve that. 3 weeks from December 10th, which that would be I think a huge catalyst.
>> right?
>> Because 100%. So, all right. So, you're saying that so what do you now that we're actually did this little breakout today and we were completely going counter to the market all day long from the very beginning. We were we were going crazy. And I had this brainstorm though that maybe this could be the beginning of the street understanding that Tesla is a different kind of AI company.
All the other AI companies are digital and we are and Tesla is the only one that is uh products.
>> Right. So [clears throat] I think that with the breakout it's looking at 550 to 600. So that seems like the the first level above the all-time highs where where we'd be at. And if we have the [clears throat] safety monitor being the beginnings of the March to April of CyberCap where we have um unsupervised at scale.
>> right? At scale well beyond Weimo. So CyberCap production.
>> you know, you're going to start making at least um a thousand a week, right? 2,000 a week, you know, just at at a basic level, you know, based on the history like even uh Cyber Truck >> um had got to 50 60,000 uh units in the first year and that Cyber Truck is five times harder to make than a Cyber Cab, right? So, you know, a simpler, cheaper vehicle that doesn't get [clears throat] to 250,000, you know, would be so 5,000 per week, you know, very very quickly. 5,000 per week at 20,000 some adult. Let's say $30,000 per vehicle. Then you're looking at um um 5,000 times 30,000 150 million per per week, you know, half billion dollars um per month, right? At a very low level, right? So that is a huge commitment of of money >> that if you're not absolutely sure you have unsupervised fully solved, then how can you do that? And so we have the um January February 10 times the parameter model more reinforcement learning. So even before we get those other models we're saying it's good enough to remove the safety monitor and then get 10 times better and then we have the cyber cap. So that's the confidence that we can do unsupervised at scale. So that March for the next four months uh five months [clears throat] is what will be um the in your face all the analysts have to say arch invest is right this is happening and so then now what how they view that knowledge that now they know scaling has to happen we have to get to the $3,000 per share are they gonna say it's going to be 2027 or is it um 2028 you know that trajectory would be locked in.
>> All right. So, the other thing that's happening right now is everybody is talking about AI in space. Uh you AI data centers in space, you and I had a conversation about it on Wednesday night a little bit. I think you've taken [clears throat] what you said on Wednesday night and really fleshed it out a lot because this is the big story. This is this may be the biggest story of our life.
>> right? Biggest story of civilization. Yeah. I think I think based on listening to all in podcast uh based on listening to uh the Peter Diamandis which is an old buddy of yours. I guess you went to Singularity University and everything and so so
>> I actually was a lecture at Singularity University.
>> You were a lecturer at Singularity.
>> I gave annual updates on nanotechnology for um for four years and um
>> so I wasn't attending I was teaching at
>> Yeah. All right. I'm I'm now I'm more impressed. Okay, [laughter] so let's let's turn over the story to the futurist right now, Brian Wong, who has
>> We want to hear from you, Brian. How does this data center and space thing develop? And before you start, wait a minute, I did some research today >> because we've been talking about whether Tesla would be in the making the solar arrays that they would set up a gigafactory to make solar arrays. I did some research and at least Grock says that right now those solar arrays are being made by a Taiwanese company. I didn't write down the name. I've got it somewhere. Taiwanese company is making the solar panels, if you will, but then they're being sent over to SpaceX and SpaceX is doing the final assembly themselves.
>> right?
>> Okay. So, and they said that and and Grock at least believes that this Taiwan company can definitely expand their production to meet the demand uh in plenty of time. So, it may be that Tesla won't be involved in that part of the game, but that's just one one uh one analysis, right? Um I've seen that as well. and that um the um someone whoever's doing the the the solar um uh panels for for space um SpaceX already has the supply chain and the orders in to go up about um 20 times from current levels they so their current level is on the order of 50 to 100 megawws per year >> based upon the 3,000 um scar Starlink um version two minis that they're putting up, but they already next year are planning to put the version three salaries for communications. This was before the ideas plans. They're saying communications, >> we have the FCC approvals. We're going to put up, >> you know, 30,000 50,000 and they will be bigger and better versions of those things. So, they had already planned out that they were going to go to about a gigawatt of that. They're going to they're planning to go up 10 times in terms of the the power for those satellites.
>> But now Elon has tweeted out that um they're going to go from 20 kilowatts per version three satellite to 100 or 150 kilowatts because they want to um put the AIDS center chip. They need more power versus communications, right? So they had the gigawatt plan. They had the FA factory supplier, the Taiwanese supplier, the scale up of of that. But now they say, "Okay, we need to go at least, you know, seven times, 10 times more than that as we up the power levels, but not just doing 50,000 satellite, they're going to go to a million." So thus they have to do the IPO because they need to get 1020 billion investment to do the um the 50 gawatt centers which that scale of solar plant exists in China. They have several um manufacturers at the 50 gawatt level adding up to um 600 gawatt to a terowatt of solar um for earthbased uh installation. Right. But we're gonna make it for space. So you have to make it lighter. No mass you a bunch of other things that make it lighter. So it's not the same thing, right? So China can't >> convert their their solar to space solar as quickly. One, you have to have a customer >> and that would be SpaceX and and doing it. Two, you need to, >> you know, do a bunch of things and SpaceX has experience, >> right? the only one who has 100 megawatt or more experience with spacebased satellite. So they have 9,000 satellites up there. Everyone else ever >> correct >> has about 4,000 and and half of those were launched by SpaceX.
>> Correct.
>> Right. Half the other non SpaceX were launched by SpaceX. So they started before Starlink was up six years ago.
>> Yeah. there was 2,000 other satellites >> and now we have on the order of you know 13,000 9,000 of which are SpaceX. So in terms of the knowledge of making solar for space at scale >> SpaceX [snorts] is way in the lead beyond China, beyond any other competitor. Um, okay.
>> All right. I have another little minor question before we get into your slides and that is with regard to some things Elon has been saying and I'm not maybe you're hopefully you're fully on board on this.
>> Is Elon planning to have the the version three satellites have both a data chip and the same communications as it would have had anyway? Or is it reducing the amount of communications and putting the data chip in? Or is it anyway what's what's
>> reducing the communications?
>> Reducing the communications. Okay.
>> Reducing communications. So they will still have so the the bulk of the communications currently um is for um large um antennas and arrays and also the electronics to communicate from space um down to earth either to a high u bandwidth internet connection or to your phone. There's two different types. Um, [clears throat] and you have a lot of gear in there. So, you would remove on the order of 60 to 80%
>> Wow.
>> of what's in there now to put in more chips. There are some chips in there now.
>> Mhm.
>> Like AMD um radiation hardened chips. They're about 5% the compute power of an H100.
>> I see.
>> But you do not want to use those chips for AI data centers. Um, you want to take unmodified um, regular AI chips straight off the TSMC Nvidia lines and then you want to throw them into your box and then send them up there. But you'll still keep 20% of the box that has solar power connections, laser communications, a bunch of other overlap that you want to keep the same. and you want to follow the same designs um as you already going to massproduce. It's like I'm already making uh 50,000 Model S. I'm going to go to 1 million Model Y. I want to keep commonality of parts wherever I can.
>> Right? So, the same kind of scale up of going from Model S to Model 3 and Model Y is what we're looking to do. and we're going to swap out parts that um are needed to do the new use case.
>> Okay. All right. Okay. I'll turn over to you from here to do the slideshow.
>> Yeah. And then you know um interject at any point.
>> Oh yeah. Yeah. Yeah.
>> Um so let me view slideshow. Okay. Let me sorry went to the back of this thing. Sorry. Get back to beginning. Okay, so we had previously said space AI is bigger than Tesla Optimus, which all the people on, you know, in our community of Tesla YouTubers, CERN, Basher, Herbert, everyone's telling you that Tesla Optimus is the biggest thing ever. It's all labor, right? And I will tell you why that is not the case. Well, I'm already gonna interrupt you because in uh I think in not in the Elon Musk mission, but somewhere along the line on this show many times you said that you thought that Starlink in all of its versions and everything that was capable of doing. You believe Starlink and and SpaceX was going to be larger than Tesla.
>> right? Larger than Tesla, large opportunity. Um but Tesla's going to go along with this because they're making the chips, right? So Tesla is a firm partner. Also X and XAI are firm partners in this thing. It'll be split up. The exact ratio will change over time. It could be 50/50 for a while. It could be, you know, 40, 30, 10, you know. So it'll change as things evolve, but each of the three companies will be major parts doing huge lift. Um there are scenarios where XAI becomes the biggest one. Because if you get super intelligence, right?
>> Then their part of the thing becomes more important.
>> Right.
>> Okay. So for people don't know Elon reply I replied to Elon about exowatts for AI in space tried to break it down and then Elon replied to me about lunar factories and mass drivers get to 100 terowatts per year. That sounds like a nice little short statement. It is incredibly profound what that means because the entire earth it when you go count for 100% power which none other power plants do you have to do a little calculation conversion the entire earth right now uses 3.4 terowatts of power per year. So when he says 100 terowatts per year, that is 30 times the entire power of the earth right now and he was want to do that every year, right? And this is not him, you know, pumping or something like that. This these I believe are firm plans that they're going to be working towards. And I'll go over in the rest of these slides, you know, what that all means.
>> Okay.
>> Okay. So everything depends upon Starship. the the thing that that the Tesla Q and the haters will say or SpaceX Elon haters will say, "Oh, it's gone up. It's blown up a few times." Blah, blah, blah. 11 launches, three times a booster has landed. And how many times has any other company, any other country landed a um a booster, an orbital class booster? It's actually now one.
>> One Blue Origin did it once. Yeah.
>> Right. And Blue Origin started before SpaceX >> and it's a much smaller beast.
>> It's smaller beast, right? And they have not relaunched relaunched it yet.
>> SpaceX has relaunched 500 times. And even the Starship, this Starship failure has relaunched two of the boosters, >> right? So Starship, the the test thing, has already done twice as many uh um three times as many landings of their booster than Blue Origin has and all other um companies and countries combined. And they've relaunched twice, which is double the number of booster relaunches of any other company and countries in the world. Right.
>> Or infinite actually because nobody's ever done it.
>> If you know in Cloud Blue Origins, first thing that they did like a few weeks ago >> because they haven't relaunched it yet. Yeah. So
>> relaunched it. That's right. Um, so just a scale of things. So the International Space Station and then the size roughly of the panels roughly like there's four of these double wings uh on the um on the space station. Actually there's a bit more. There's some other things you need more updated photo of it. And
>> this is the solar these are the solar arrays.
>> Solar arrays on the on the on the on the space station. So the SpaceX if they go to 150 kilowatts will have roughly the same size >> as one of these double wings, right? And then a football field to give you some idea how big this is. So it's like the width over the width of a football field about 15 yards wide. Okay.
>> Well beyond Optimus because Optimus is just labor just all labor, right? Um see energy in Earth cannot match the energy in space. The sun produces 10 10 billion times more power than hits the earth and we are only using one 500th one 1,000th of that equivalent total power that is hitting the earth. Um and we need to do all these other things that we're doing you know nuclear oil natural gas in order to achieve that. So space a sensor will beat Optimus because the energy cannot be built on earth. We cannot have enough Optimus to compete against the number of AI chips we're talking about and I'll go into how many roughly on the order of say a few billion Optimus versus many many trillions of AI chips in space right so the order of magnitude is 1,000 times more right so example would be no matter how good your small island is you cannot compete against the United States the most successful ful um small city is Singapore.
>> Right.
>> Right. Quite rich, really nice place, 6 million people. The GDP of Singapore cannot compete with the GDP of the United States with 350 million people.
>> Right.
>> Not even close.
>> Not even close. Uh so AI and energy will grow from 13% of the economy today. Energy and and that combined toward 99%. Labor is about 55% but it's dropping already and it will drop faster when this happens. So and so then there'll be a decoupling when you go fully automated just like when you go robo taxi fully automated, you will decouple from the human in the loop the human limitations will go away when you decouple. So just a few slides about why AI um will succeed for sure people always say oh how what do you use it for and that kind of stuff. So Grock 4.2 2, not fully released yet, but they're testing it for the last few weeks, did stock trading, real money that they did four different accounts against eight other models, including the old Gro 4. The Gro 4 did terrible. It's it's down from $10,000 to $43 in about, you know, less than 30 days, while Gro 4.2 is a leader in pink at plus 35% in less than 30 days. So trading to get that much money and all the other models have lost money >> when you combine all four models you know all lost mod money. Um so the thing is what is it that they did one they tell you exactly why they're in the trade why was it holding Palunteer etc. it was only trading six different socks and the NASDAQ because they were going with a particular company that has made special um derivatives call option type things puts that have 10 to 20 times the leverage >> right 10 to 20 times more risk >> and they say okay I'm so positive that I will put in a 20 times bet on Palunteer Google some of that and they do it for on the order of 10 minutes to 24 hours >> so it's it's day trading type stuff that they're for people who who don't know that um they all the model explain what they're doing and then what does 10 to 20 times risk mean that if people play roulette then 36 numbers plus the zeros um you have a 35 times bet if you put one number just a little better odds than 20% 20 times is betting two numbers at a time right two numbers at a time on one spin I have a one in 17 chance of of winning um three numbers is 1 to 11. So if you're doing those kind of bets at the rullet wheel, people who don't do gambling, not degenerate gamblers like me or something like that, then then you would know that if I got a bank roll of 500 bucks and I'm betting 10 bucks per per spin, in about um 200 spins, you'll be wiped out, >> right? Just a random luck thing. You that's your expectation. About three hours of play, you you you're done, right? Um, if I do three bets at a time, it would take only an hour, right? Because I'm betting 30 bucks at a shot, right? I'm speeding up how fast I'm wiped out. So, they're playing five different bets at a time, right? Not all their money because they're in their bank roll and they're still there like 24, 48 hours later, right? So, anyone who knows numbers knows that if you're still there at, you know, six times 10 times longer than than than random, right?
>> This is not random anymore.
>> right?
>> And it it means that they can win day trading completely. How much win day trading completely? That would be the the bottom third third row of this thing. You know, it's the software tools for day trading about $10 billion, but the people who are using the software are probably doing 10 times more, right? So, they're doing it's about 10 to 25% of US equity volume is day trading, right? So,
>> you're looking at, you know, half a trillion to a trillion dollars or more of day trading stuff. This is [clears throat] a huge market.
>> Yeah.
>> Right. If you if it means that for this game just like winning chess for a for for AI if you win this game that is a huge market that you have way more money than open AI's $20 billion blah blah blah it's like it's conquering a new thing and then
>> it doesn't mean that they can't get to program trading high frequency trading other forms of trading right once you've mastered one it's just like I master a chess computer has mastered bliss chess it's mastered some form of go it will get the test, >> right?
>> Right. You know, it's just it's going to happen. Just a history of how these things work. Um, also there's a score based on GDP val tries to get at can you get the skills and knowledge of someone with 14 years of experience in a particular job, nursing, doctors, lawyers, and old Gro 4, the inferior one that was losing on the trading thing, um, did 60% of that. and then the latest 5 point opening of gra uh opening 4.2 5.2 GPT 71%. So all of these models are going to be able to replace many of your jobs. So the fear you have about that not unfounded >> but the main point here is that scaling up AI to 1,000 million times will have massive economic value. Okay, just the main thing is AI useful conquering big industries. The other big industry is um advertising which um Meta [clears throat] and Google have said AI has been helping them to get 20% more productive on advertising $1 trillion global advertising market a half a trillion dollars for the big tech companies on on advertising right so that's a big market and then uh IT software making that's about a $5 trillion market 1.5 for software another for another 1.5 for um consulting 1.5 for hardware so um this that also can be taken over as well if you get fully automated programming or something like that. So not going to go too much into it just you know we can do a deeper dive
>> but it's but it's interesting also you've got both sides of that trade as well. So, you've got Amazon that is trying to maximize the amount of money they're making from the advertising, but then you've got the advertisers, which would be, you know, I have an I have an Amazon business and I'm trying to figure out how to make sure that I'm making money using their advertising to sell product, including the new book. By the way, it's also the same thing. You advertise the book on Amazon and you're hoping that if you put a dollar in, you get a$120 out, not 90s. Yeah.
>> And so and and it's very complex and complicated to try to figure out how to beat, you know, the other competitors, you know, in that world. But once again, AI would definitely uh cause you to have new new tools. And then for all the content and all the video there, the AI is making videos. Grock um 4.1 beta that I've been using. I noticed this morning, I think maybe yesterday, that they were when I asked a question, oftentimes they would put in um pictures with the um answer. So, it have some text answer and it have several pictures, you know, like a diagram, a a a patent photo, it's mixing the pictures. So, all the the work that I was doing on my science blog of like here's the picture to explain it, do that stuff, it's now doing that, right?
>> And and then making videos and short videos, long videos. Yeah,
>> that is happening. So,
>> um it's all through 2026, expect that to just explode, right?
>> Override anything. So, all the content stuff. So, again, the value question of AI is not a debate anymore. It is happening. It will happen. It be trillions of dollars. Also, that means the spending a half trillion dollars going to a trillion dollars per year. That will all pay off. you must spend for AI in order to get future business.
>> So, um, so that's I want to go over is no AI bubble. The spending will pay off. It's happening right now. It's going to go all over 2026 and they will continue to scale it. Um, one thing I live in Silicon Valley. One thing Silicon Valley knows that other people in the rest of the world, other people, other cities may not know is that once the unit economic works, so I can do one e-commerce trade, I can sell one book and that I can scale that then you scale it until it stops working. And then that's where you get all the money that they I went to multi- trillion dollars because I took one thing that works that is digital and then I scale it. Open AAI they got the attention is all you need paper from BU in China or you know some other work that was done by Google and they realized holy crap I scale this up a thousand times and then I get huge value. So that was them making the bet of this works at some smaller scale. It shows that it gets better and better as I do more and then I make the bet and say okay load up I this is what it is to go exponential. Right? So that was what happened before other examples that we have. So what [clears throat] I just proved is the first part. There is something to scale. It will make money. All we have to do is do more.
>> Right?
>> And now I'm going to discuss the scaling part.
>> Okay.
>> So okay next slide. So now we go beyond AI. It's all going to work. Um sorry. Okay. here's a scaling electricity um in the first while I get you know Edison whoever else was working on this stuff m make sure that electricity works and then they make um the first grids 1915 after the first you know 2530 years of them doing this stuff they get to a reasonable level you're at World War I levels [snorts] 35 terowatt hours look to the bottom of this chart going every 10 years and then you have us at 32,000 roughly 1,000 times more than World War I, 110 years later, 1,000 times more. So that can give you some perspective. If people can imagine the world of World War I and how much electricity there was to the world we have today and how much electricity we have, that is 1,000 times scale up.
>> Another area point out would be in the middle of this thing around 1935 during the Great Depression is when they, you know, first make the Hoover Dam or something like that. 300ish ter went up 10 times. We're up a hundred times from there. So that's the scale of a 100 times and a thousand times over the growth of electricity, right? And AI will be the new electricity. So a 1,000 increase world one, 100 times World War II. AI right now is only 1% of our current energy. So we're only using about the level of electricity that we used for the world for powering other stuff in the Great Depression to power AI, >> right? So, but then you can only take up so much of the percentage of the world's electricity before you know you have to build a lot lot more and that is very difficult on earth >> as we've been talking about as we've been talking about in the book and on a dozen shows.
>> Right. Right. So, the first level goals is what Elon talked about 100 gigawatts per year maybe 300 gigawatts per year. If you sustain that for 10 years at the 100 gawatt level, that's 8,000 terowatt hours, that is double the energy of the United States today, right? Over 10 years. Elon doing that, which he thinks he can start doing in 2030, right? And if he just sustained that level through 2040, he would have double the energy in the United States. You cannot think of any other technology that could to generate power that could double. you end up with, you know, United States plus the power in space, triple the energy in the United States, right? And and 60 70% of it for AI, right? Right. Is you cannot scale that. You're looking at 40 50 years in the optimal case for an earthbased solution. And Elon thinks he can even get to 300 gawatt per year, sometime in 2030s in this first scale up.
>> And that's 24,000 terowatts. That's double the energy of China. that is um um six times the power of the United States right from that. So again no other ways to do these kind of numbers and this is very achievable I will discuss some of that in terms of launches but first this is so profound that it is we're going from GDP per person to GDP per AI chip. So various calculations can show that you can make $20,000 per year from an AI chip. The world average GDP per person is about $10,000 12,000 per person. So an AI chip can be twice as productive as the average person on Earth. Obviously the developed world in the United States is four times higher. So I need four chips to equal an average US person or I need a better chip in a few years that is four times more productive than than the previous chip I got. But basically this transition from GDP per person to GDP for AI chip is where this un unleashes everything and the people kept saying you know singularity we pass through it when AI happens and we can't predict what happens right is not easy because I have to do it the the we can say if I can get to a trillion chips then that is a 100 times or more GDP than 10 billion people.
>> Mhm.
>> Right? So how fast this goes I we can actually forecast which is very useful if there's ways you can use that information to figure out what's going to happen and how to invest etc. So 100 times more electricity than there is today. So lunar activation though so that level 300 ter gawatt for a decade is 100 times more electricity than the AI uses in the world now right. So already 100 times more for Elon doing it for a decade from 2030 to 2040. And then he talked about he reply to me if you recall from the beginning was lunar activation goes to 100 terowatts per year. So that means starting at 1 terowatt for for 10 years where 80,000 terowatts over a decade three times current world electricity. 10 terowatts again still only 10% of what he thinks it can do with the moon. 80 thou 800,000 terowatts over a decade. Three uh three times more than all world energy every year. 30 times more than over a decade. Go up another 10 times. It's 30 times every year. 300 times more than everything we have now. So it's just impossible for anything on Earth to compete with this just because there's a million times more resources um out out in the sources and we're just a tight fraction just starting to use the moon to do some of this stuff. SpaceX launches done a bunch of calculations about 30 megawws per launch. It will then increase to about 60 megawatts and there's some stuff you can do to get even beyond that. But just the easy clearly doable is 30 megawatts per launch going to 60. So your first 100 gawatts, say 3,000 launches, then you improve to 60 megawatts per launch and you're doing 5,000 approved launches per year to get to the 300. So is that a big number? Uh we have almost 200 launches of Falcon 9 this year. So 3,000 launches is about 20 times more 15 times more than what we're doing now with Falcon 9.
>> And then uh 50,000 will be about 25 times more launches. So difficult but it's actually doable. It would involve more launch towers and other infrastructure which I've worked out but too long to go to here. Just saying that this is doable. difficult but duel one of the difficulties is this will need 5 to 10% of the US natural gas in order to make the fuel for the rockets right >> so we have difficulty scaling the rockets beyond this level right because we need so much natural gas for fuel you can do it but it gets again hard and you probably don't want to um again that 300 gawatts is three terowatts per year three billion 1 kilowatt chips over a decade So that goes to Elon talking about he needs more fabs to make billions of chips.
>> Um we have about um 10 million to 20 million AI chips in the world today. We're making about 5 million maybe 10 million next year and TSMC wants to scale by 10 times to get to 50 million uh chips in 2030. So 50 million chips is not 3 billion, right? That's what he says. He can't get enough chips. That's what he's talking about. That he needs to get more. Um they can scale up their factories more. Um they have to dedicate the factories. They need to go all out on lines. Um there's again supply chain scaling because I'm going from a particular level up to 100 times more. So you have to scale the supply chain. It's just a factory thing. Um and this is again 100 times more electricity than we have. So supply chain on the on the wafers, supply chain on the um um the fab line, the fab equipment, a lot of scaling. So launching from the moon that um Starship launch problem, too much natural gas needed to make the fuel.
>> No satellite launches, sorry, no satellite starships will be launched after the factory and the fabs are made on the moon. Um, you'll still be doing stuff, shelling other things, but the limitation of taking using rockets to take the fabs that you make either on Earth or the moon where you need them to go, you won't need rockets to do that anymore. You will use a mass driver, which is a magnetic um track to do it. Again, I've gone to details. Think about if people know that there's magnetic levitation trains. the the train that goes from um uh Shanghai airport to Puong, another sub city within Shanghai, you know, over like several dozen miles. That managation train is comparable to the kind of um mass drivers you'd be using. You want to get up to about um 2.8 kilometers/s to escape lunar gravity versus 9.8. So you get to 25% of the speed of earth one6th of the energy that you would need and you can do that with these kind of mass drivers. So fair engineering to do it another video to discuss just take my word for it. It is doable. I've studied the papers study the work upon it. There have been studies and work done for this thing. NASA studies have been done that this is completely doable. We have technology of comparable level that we could adapt to achieve this. But anything no more fuel. Go ahead,
>> Ryan. The my recollection was that there has been uh that the the technology that you're talking about that's already been figured out was to do this on Earth >> to try to to try to use one of these mass drivers to get the satellites uh from Earth into orbit.
>> There's all kinds [clears throat] of reasons that people have thought about doing it. Um, there's um elect electromagnetic um fighter jet launchers which is what the aircraft carriers used. So there's many military applications for this um and just a question of like you know how much have we actually done upon it but there have been studies there have been equipment made to consider can we make this work and now it's not that much more than the mandic levitation trains that we have the track might be a bit longer we need to accelerate a bit faster technologically engineering wise a doable exercise Okay.
>> Um, so and then at that scaling in terms of chip, we talked about how much of earth power we talked about before, but it's how many chips 10 billion chips per decade at the um 1 terowatt level. 100 billion chips over the at the 10 um 10 terowatt level and that 100 terowatt level that Elon talked about the beginning one trillion chips. So you know that is um about 100,000 times more than we have today. Now, were you going to be able to get permitting on the moon to to extract all this material?
>> Uh, good thing is there is no permitting on the moon.
>> Oh. Oh. [laughter]
>> Yeah. So, it's good news. Good news. You can just get there and just do it because there's no one there to say you can't. [laughter] And anyone who does say you can't, they have to say, well, come up to the moon and stop me.
>> Yeah.
>> Right. [laughter]
>> So, then here's some calculation about the token revenue trillions you can make. highly speculative because we don't know how the pricing evolves but the numbers should work out. Um, but if they don't then this thing does grind to a halt. If suddenly this becomes uneconomic for some reason it it you know will grind to halt. The advantage of doing things on the moon is that in theory it should be 10 times cheaper. M
>> so then if I could only make $200 billion per year and I thought I needed $2 trillion per year to do it because the moon can do it so much cheaper then you can make this keep scaling up because you kept lowering the cost
>> right
>> um and then there's some comparisons about how much material that we use because one trillion chips 10 billion satellites is about a billion tons right? Or several billion tons. We do make about 2.6 billion tons of iron ore per year. About 4 billion tons of cement that's in the center here. So that's a scale of civilization stuff we're talking about. We're going to, you know, civilization scale, which again, if we're doing it on Earth, we'd have to double up, you know, iron production or something like that. And it's just harder to do it here versus out there where we move the dirty industry out to do it. Um, but again, achievable, not You know when you say you know how much we're talking about we have done it before if we really wanted to if you had the economic reason and then this is kind of like an one examples of those solar panels next to people and the thing is the weight of this thing because you'd have like four times as much as this solar panel for your version three satellite right and the whole satellite including the panels and the the chips and everything else which is the most weight only like 10% of your weight is the solar panels this big thing is actually the six guys could actually probably lift it.
>> Mhm.
>> Right. So, it's it's you need to make it so that it's really light because you're launching it into space. So, um that is um what we're talking about for that.
>> Okay. So, um plenty of reason for the excitement that's out there. Uh we talked about this on Wednesday night, but uh uh uh on the um Moonshot program with Peter Diamadis, he said he their group, their crew was making the point that nobody was talking about this four weeks ago and that four weeks later it's all everybody's talking about and their enthusiasm was off the charts. They were so excited. every single one of the people on the panel uh was talking about what this could mean and how fast it could be and the fact that the science is basically solved that there's no that there's no big engineering challenge at least with regard to the first stages of this getting these satellites into space uh getting them uh getting having them take advantage of the fact that there's that it's cold in space so you don't need the amount of cooling and that that apparently Apparently the technology is there for the radiating away of the heat. At least at least they think it is. I mean there's evidence.
>> No, no, it's not. They think it is because the the Starlink satellites, you know, with 10 kilowatts or with 20 kilowatts, >> right?
>> Whether you use it for communication or whether you use it for compute, >> you got electricity, it's generating heat, you got to get rid of it. Right.
>> Right. Doesn't matter what I ran it through. I ran through something. I generated heat.
>> Right. at the level of 10 kilowatts or 20 kilowatts, right? And the the panels on the space station at the 150 kilowatt level, those are generating heat. You have to dump it off, right? So, those are solved problems at those scales.
>> Got it. Okay. So, now here's the here's the question that I think I asked Rock earlier today or maybe I was just reading about it. So much stuff every day that I can't keep straight. Um, one of the questions was in terms of batteries, uh, that, um, at low Earth orbit, uh, you're only facing the sun half the time or 60% of the time. So, you have to have battery storage, uh, in order to take care of that.
>> >> you put them at sunsynchronous uh, orbit. So there's so it's another low orbit but it's it's going at a particular it's on the terminator you know say you know like you have the the earth night and day the dawn dusk line right >> if you go around there then you're always >> in in in sunshine right >> so if you want to prompt your AI >> suns synchronous orbits >> suns synchronous okay >> dusk dusk dawn uh thing and it can be different heights it can be at you know 500 it can be at 800 so there's several layers of and and you think, well, if I'm loading this all up, the orbits are bigger than the Earth, >> right?
>> Right. So, I'm how many can I put around the Earth and not run into each other, right? And then I have, >> you know, any anywhere between, you know, 300 miles. So if I put a five mile gap at a layer between the things, I could have one, two, I could have like >> a hundred different layers of this thing wrapping around the earth 100 times, you know, going like um a thousand wide around the earth. So easily I can put all these satellites up there. And so what about the business of uh which which level of orbit would this also be advantaged by being in low earth orbit where you the latency would be less or does that matter?
>> Um [clears throat] some advantages for the lower latency with the closer orbit but um other factors could override whe whether you want to do that or not. um a bit more technical as to what it is. I would say um probably start off pretty low, 500 milesish, higher than the communications because you it's more important to have the 24-hour sunlight >> than it is to have a fast communication because I'm sending the question up. I'm going to think about it and they're going to answer you. The time it takes me to think about it is longer than the communication time. So,
>> so then, so then I'm going to I'm going to just spitball here because I don't know what I'm talking about. But if you're putting those at a higher orbit, then you wouldn't want them to communicate the communications directly with Earth. So you would have the lasers between the satellites where the where any of this inference that was being produced on the communication side would go down to a lower lower level satellite and then be and then be uh
>> it's usually not that complicated because um if I'm sending a question over to um the cloud, right?
>> The cloud is racks and racks of these um these things. So I have like a rack with say a hundred >> chips in it. You know it's actually 72 but it's 100, right? And then I got a data center with you know soon a million uh chips. So then I have 10,000 racks, right? If I send a question from chat GPT into the cloud and goes to that that rack, >> it's going to go to one server. It's not going to you know what's your question that that the one server can't handle it, right? maybe some massive research question I guess but in general one server will figure it out and it'll send back the answer right so uh for inference it's just talk to one get back out we're not going to be doing training in space >> for a while right training means that I'll have the
million chips, 10 million chips, working together to train up a new AI model. >> Um, that requires more precision. The errors that we're getting, uh, won't make that work. But for AI inference, the ChatGPT, actually using it, the Grok, actually using it, this will work. And it's the communication levels is low. It's like, I think, pick up a question, I send it over to Grok, it thinks about it, and sends me back the answer, right? Most of the work is there, but it's not, it's only like, um, you know, a million tokens. It's, it's, it's basically like, um, a fraction of a second of one GPU to give me the answer.
>> Okay. All right. So, [laughter] so, um, what's this worth?
>> Well, >> we got, so we got, so we got, you know, we got Optimus, we said is worth 25 trillion.
>> Mhm.
>> We've got Robo taxi, we said is worth 7 trillion.
>> As a market cap?
>> So, what is AI in space worth for SpaceX?
>> So, let's just only go at the, um, 100 gigawatt level, right? 100 gigawatts per year. Let's say we do that for 10 years. So, we're talking about terawatts, right? So, what's a terawatt of compute worth? Building it on Earth would be about, we're currently spending about, um, $50 billion per gigawatt. So, um, one terawatt, 1,000 gigawatts, that's worth about $50 trillion, right? And if we make it cheaper, then that would be, you know, only half the cost. You know, look at $25 trillion to build it. And then how much you're you're building it? You're presumably going to make more than it cost to build it. So beyond that $25 trillion level, right? If I go to the, um, so that's something that we could get in the, um, um, 2030s where we're, um, generating that. So you'd be generating $5 trillion per year on a build of like, um, you know, $25 trillion over the years. So make $50 trillion per year, like 50% margin on it, right? Then, you know, you're you're looking at, um, you know, 2.5 trillion per year stacking up towards $25 trillion per year toward the end, right? So that's the kind of, um, growth as this thing stacks up. And then, um, if we do the 10 times more, the one terawatt level, where we start getting some of the, um, the, um, lunar stuff going to 300 gigawatt level, the level that he was talking about. So that would be $7.5 trillion scaling up to about $75 trillion, right? So it would be rapidly, um, bigger than even, um, the bots. And it would be $75 trillion is just shy of the hundred some trillion that we have for our world economy now. So you're going to get double the world economy, which doubling the world economy in 10 years, um, is basically, um, the rule of 72, about 7% growth per year. So then you're going to be going from 3%, 4% GDP growth, >> towards, um, 10, 11% GDP growth just from adding this up. >> And the worldwide as well. >> And worldwide as well. >> That'd be worldwide.
>> One of the most exciting parts of all of this, of course, whether it's Optimus or whether it's, uh, space-based, uh, uh, data centers, is how it's going to lift up the poorest people in the world, >> right?
>> Um, um, yeah. Okay. Whoops. I, I, I confused myself and I lost my, my other question. I had a great one.
>> Well, let me, let me just go over the, the lunar-based version of it. So, yes, it may take an extra, um, decade or two, although that can come on, um, fairly quickly. Is that that is 1,000 times more than the level that we just talked about. So it's, you know, you could be adding, you know, let's say it takes 30 years to do, >> 10x, 10x each decade. You know, you're getting an extra sustainable 7, 10% GDP growth for for, you know, three decades and maybe forever, right? Where you'll be growing at 10% for like 40, 50 years. So this is something that doesn't like, oh, I do it and then I stop. It's like I do it, it keeps going and going and it go like for a couple of centuries before we run out of power and and material in the solar system. So, and then that's not even including how much we improve the AI. If not including, you know, the constant improvement of the chips and stuff like that, which can add another, uh, layer to it. Um, so it's the the level of, um, wealth and capability it really unlocks and cooks everything, but it gives a a shape to what are we talking about for an AI energy, really unlimited future growth scenario about how fast we can build this stuff out. Um, so for people who just say, once a singularity happens, we don't know what happens. We actually, it's very clear that once you just say it's all about energy and chips and compute, it's actually quite predictable. You can run the spreadsheets, we can kind of predict how we can do the timing of of what happens.
>> Yeah. And and, uh, you know, you just brought that up again, like, okay, what are we going to do? What are humans going to do? What's our purpose going to be? Uh, what's the value of a human at this point? And, uh, that's why you need to read the book, is because in chapter 4, we talk about the jobs that we can think of right now. And I think you'll agree when you read it, uh, the jobs that we can think of right now that humans will continue to do in bigger numbers than they're doing them now. I thought, uh, again, I'm always quoting, uh, WTF or the Moonshot, whichever brand you want to use. They have two brands. You're only supposed to have one brand. Uh, but they were saying the other day that, uh, uh, radiologists, they thought was going to be the first group put out of it. There would be no more radiologists because the computers now, right now, AI was one of the first things they did was you use them to read, um, the X-rays and and MRIs, and they're better at it. They're like way better than a radiologist is at reading the charts. And, uh, so, uh, Peter said, no, what we've turned out, what has turned out to be true is there are more radiologists than ever. Because no radiologist ever went to school because he was hoping to read charts. What they wanted to do is save lives and make people well. And so having the boring part of it done for them makes them more successful, more efficient. And so there's actually an increase in the number of radiologists, uh, that are working today compared to three years ago. And because they're giving the educated human interface to the knowledge and guiding you through the personal choices, right?
>> You know, engaging with your emotions and helping you get the medical treatment and choices that you want to make so that you are healthier, deliver the result of being healthier, longer lived, and feeling good about it. So we'll be helping each other to engage and like you will be more specialized in your one thing, but you'll be guiding other people to result. Yes, you'll be using a lot of help from the AI, but you need to know the topic in order to understand what, why, and for what purpose the AI is making the recommendation, right? So if you don't understand one, you can learn fast, you can learn more and more about what to do, but you can, and then understand what it is. But it's like, if I don't understand what the question is, I can't make the AI give me the right result. If I can't say it needs to be some synchronous orbit, right? It, you know, then and then then it can give you a wrong answer and like you can't verify it, right? Right?
>> So in order to to do it, you need people to understand and interpret the thing. And our ability to do that, all of us will get better and better at, just like the beginning of the, uh, personal computer age, virtually no one could use the spreadsheets. You load VisiCalc, it was very difficult to use, you couldn't understand it. Now we're all spreadsheet jockeys. But you have some guys who are more expert at it and be able to interpret and get more of the results. But everyone's level of using a spreadsheet has gone up when before almost no one could, right? So, you'll still have.
>> I saw, I saw this in my consulting. It was it was fascinating to me because here's all these guys. They know how to use a spreadsheet. This is like, you know, five years ago, seven years ago. They all know how to use a spreadsheet. They all get their, uh, results back from their bookkeeper showing them their, uh, their, uh, profit and loss for the year. If they went to a little bit of trouble, they could have been spreading them out like some of us probably do with regard to the companies that were invested in. And then, you know, make some projections for what it's going to be next time. And, uh, none of them could do that. None. Zero. [laughter] Out of 400 clients in 17 years, I mean, okay, maybe one or two of them, a few of them. I'm hyperbolizing, but I mean literally none of them were able to to get to that level that that would have been the smartest level that they could get out of the paper.
>> And the other thing is to for each person to get to a level of trust and conviction to make the the better decisions.
>> Yeah.
>> That AI could tell you something, a person could tell you something, >> right? But if I don't trust the person, if I can't believe, they can't walk me through it, I can't develop the conviction to make the decision.
>> So my analysis of this stuff, you know, early on, it, you know, I knew it was right. I knew enough to know this is guaranteed. I have now know it's right going out many years. You know, there's some level of of variability on it, but I know it's right. So I can make decisions based upon it because I have certainty. I have certainty relative certainty on the timing. I have a feel for what I'm discussing, right? So even if you have all the knowledge, if I'm, it's something being told to you, but you don't know whether to believe it or how much to believe it. There's a difference between 80% sure and 99% sure and 100% sure that, and also on the timing of if I know that's going to fit within this thing, and I'm willing to make the bet, >> commit to it.
>> Right.
>> Right. There's value in that. Right. Like whether you double your money or whether you go only at 10% per year is because it's whether you have the conviction around it. It's like, um, people who invest in the stock market. I can only invest in the index. I cannot trust anything else. I can only go with 7% per year. I can't get myself to invest in Tesla. I, you know, riding the the volatility. I can't do it. You know, Nvidia, those kind of things. It's like everyone's heard of those companies. They know all about it. But then, do they can they get themselves to the point where I understand it and I have conviction?
>> Right. So that understanding, conviction, trust, those levels of things have real value. And you may need people to bring other people along where it's like, I want to do it. I want to do it with you. I want to do it together with you. I don't want to do it, you know, that kind of consulting thing where you know where you are working with your clients. Yeah.
>> Bringing them along that that they wouldn't be able to do it. You know, theoretically they could, but because they don't have the, you know, the trust and conviction that you can impart, that you have, that you can impart to them, they can't go along for the journey, right? So there's all the kind of value. It's a longer topic to discuss, but that I think is the, um, for people to make the right choices, other people need to be involved to help them get there.
>> Sure. So I have a weird question for. When do you think Elon Musk came to the conclusion that satellite-based data centers would be a thing?
>> Um, I suspect about, um, six months ago, he it came onto his radar and probably within a month, he was, uh, absolutely convinced about it. Um, there were certain moves and other things roughly that kind of time frame. The, the, there was another company, StarCloud, >> right? >> Previously called Luminina or something like that, that was had the idea. They bought a rideshare on, uh, SpaceX's Star, uh, Falcon 9. So when you buy a rideshare, you must submit engineering drawings so they know what is it that we're launching, will it blow up or not? And that got bubbled up to Elon in a hurry. Just like the direct to cell phone thing created by Lynx and AST SpaceMobile, because Elon launches everything. He will then know, well, oh yeah, that's a nice thing to launch. Fine. And then two things come up that are like, wow, that's important. >> I already thought this other stuff, but now this, I kind of thought about before, but I wasn't. Again, conviction, understanding.
>> Yeah.
>> It brought him to conviction on it. And also Jeff Bezos saw that too. So the, there's some videos from Jeff Bezos from a couple three months ago where he says, data centers in space will be the thing. And, uh, all the move there.
>> He also probably heard the pitch from StarCloud from the CEO and them. So they have videos going back a year. Um, so it's possible that Elon saw it a year ago, but the fact that he may have been aware of it, he didn't get to the point of absolute certainty until later. Probably he was canoodling about it when he started talking about, um, I want to think about how to get to one terawatt hour as fast as possible.
>> Right. So around there, if we find the exact data, that is roughly plus or minus a month where he figured that out. Well, that's all I can think of. I, I'm sure that I could probably think of way more, but I think we've almost got an hour in at this point, and you've just explained to the folks that this is probably going to be five more, uh, programs over the next few [laughter] weeks.
>> Many, many all the time.
>> Well, Brian, uh, say good night, Brian.
>> Good night, Brian.
>> It's been great talking to you. Bye-bye.