Elon Musk Doubled Down On This Massive Bet

本站添加: 2026-05-27

[00:00:00]Elon Musk made a couple of announcements that are quite significant. First, while he agreed that SpaceX AI does not yet have the best AI model and that his strategy is to control the best hardware and become a hyperscaler, Elon said he will never give up and he still intends to get the best AI out there. He noted that SpaceX AI is only 3 years old, which is half the age of anthropic and a quarter the age of open AI. He doubled down saying, "Let's see where things stand 3 years from now."

[00:00:27]Second, Elon pointed out that SpaceX had achieved nothing of note after three years and was written off as dead after six years with three consecutive launch failures. But he said, "You may have noticed things are different now." Elon also promised he'll present a more detailed AI satellite design in a few weeks. Today, we'll give an update on the plans for Grock and estimates of how much money SpaceX can make renting out their excess compute capacity. It's much more than you think. Got Joe Bacti here with us today. He's an economist, investor, and tech entrepreneur. He has a YouTube channel covering both Tesla and AGI. Check out his website and community at pinionerlands.org.

[00:01:02]Welcome Joe.

[00:01:03]>> Good to be back. Hey, we're talking your favorite topic, AI models on the race to ASI, AGI, and this new model that Tesla, SpaceX have announced, which is compute as a service or a I think uh Elon calls it AI comput as a service and I think all in podcast they call it um Elon where uh web services.

[00:01:26]>> Elon web services, EWS.

[00:01:30]So Peter Demandis said this. Elon doesn't need to run the best AI model.

[00:01:34]So I think it's pretty clear at this point that XAI is behind. They do not have the best AI model out there. He needs to control the best hardware.

[00:01:42]Nvidia, he said, is the most valuable company on Earth and their models aren't the most popular. The hyperscaler play works. Okay. And Elon replied to him saying, "What you say is true, but nonetheless, our AI will be great.

[00:01:56]Whether it is the best remains to be seen, but I will never give up. Never.

[00:02:00]And then he reminded us that SpaceX AI is only three years old. That's half the age of anthropic and quarter the age of OpenAI. Let's see where things stand 3 years from now. Then he added this after he posted this post. He added another post followup. He gave it some more thought. He said SpaceX had achieved nothing of note after three years and was written off as dead after six years with three consecutive launch failures.

[00:02:26]You may have noticed things are different now. Elon plays a long game and he's not going to give up ever comments.

[00:02:34]>> I think three takeaways here each one separate but each one very important separate dimensions and points. First one from a pioneer and reality engineering perspective which is like our topic in pioneer lens. How do we engineer realities? What can we learn from the best pioneers like Elon? It's easy to say, well, long-term we always win. Because the big question is how how can you be long-term? What is the precondition to be long-term? And what is the precondition to be long-term dedicated is first principle strategic decisions because once you make a decision like we need to go to Mars or we need to build SpaceX and this is not a little business endeavor this is a strategic endeavor then you gain your long-term conviction to make this happen right that's also took me a while as an entrepreneur to understand that it's always there's always a shiny new thing you can go after like I you know I was in the cancer detection business all kinds of But then at some point you have to think about what truly matters in the grand scheme of things. That's why I'm focusing now on you know pioneer capital investment infrastructure for the age of AGI and on media because I think that is the most impactful thing you can do from an Elon perspective and that means now we have conviction to do the long game.

[00:03:50]It doesn't matter. We don't need to make money. We we just keep doing it because it has to be done. And the same is true for Elon and SpaceX and of course for Tesla. And everything he does now is of strategic importance for civilization.

[00:04:03]Not because Elon does it, but he does it because it is that. And that gives him the long game. Number two, point number two, do they need the best model or not?

[00:04:15]Uh and uh Peter Dander says they don't even need the best model. Nvidia doesn't have the best model, but they're the largest company on Earth. Okay, number one, he's right. Peter Dmenus is right.

[00:04:25]They don't need the best model. They just need to own significant pieces of the stack and be the number one in the stack. They will be the number one in energy, which is kind of a big deal when you think about it. They will be the number one in the galaxy. Well, who knows about the galaxy. In the solar system and no idea, I don't make statements on aliens, but in the solar system, they will be the biggest provider of energy, which is a big market.

[00:04:50]um and likely in inference chips. Let's see about that. They have to do some catchup, but it's very possible. Uh then okay, so that would be already enough plus orbital lift and everything.

[00:05:06]But then number three, I do think you I never like com commented too much on Grock and all the fanboys on X that said how amazing Grock is because I always felt it's not that amazing. I'm very critical. It's like well it's great but if I go to Claude if I go to Chip it was a little better.

[00:05:27]Rock was like there had some advantages but then Claude totally started killing everything six months ago and it's just vastly superior. Okay. So I always said Grock is totally behind behind Anthropic. I didn't say that because I don't want to get killed by the fans, but I never said anything. It's like I want to do this. I don't know. But here's the interesting twist of the story which I find extremely interesting.

[00:05:52]When I started to look into anthropics power, like what makes them so great, I realized very quickly that what makes them so great? It's Elon's and X's specific weakness.

[00:06:07]It's actually not the core tech. It's not the LLM. It's the scaffolding around it. It is all the little prompts and the little tweaks and the little product things that make them very powerful. And we know X effectively doesn't have a product team. It's a total joke, right?

[00:06:23]There are no people in product. So, which is bad. But it's not just random product design. It's like what Enthropic does best. They are basically doing the piping and the scaffolding of the core models.

[00:06:35]Same is true for the agents now and the agent revolution, right? It's not the core models that make now the big difference. The core models are already the core models already have an IQ of 150 160 I would say at this point. So it doesn't matter. That's not where the money is. The money is now how do we use these things and how do we automate the usage? That is what anthropic absolutely killed and that's why they're way ahead of everyone. But now comes the twist.

[00:07:00]Cursor.

[00:07:01]What is cursor? And why was cursor the hottest thing in the world 12 months ago and then totally fell off a cliff and now suddenly he's back? Because cursor from the onset always said we are not LLM. We are just scaffolding. We just taking these LLMs and then doing stuff.

[00:07:22]They became very hot because they were the first company that actually got something done with AI encoding. So they grew like crazy. they're in all Fortune 500. They're all over the place. Then it became uncool to do that. And then people said, "No, you need to own the core. This will be replaceable." Now, it turns out it's not only not replaceable, it's more valuable than the LLMs.

[00:07:44]And here's the thing. I was critical of Elon without saying I'm always scared of the fans. I was like, I'm not I never criticize Elon, even if I secretly criticize him. But on X, I was like, it's fine. You proved you're the best in building giga scale things, but your core team is a little sucky. It's like fine, you have a little LM, but you can't get it done. You have no product team. You don't know how to build this thing. What is so amazing about Elon, and now I give him kudos, is where is He's just the best. He is a true pioneer because he understands what he's good at, what he's bad at. He understands where his teams are good at, and he understands where they're bad at. He understands he actually responded to one of my tweets about product where I make like his snarky remark. He's like, "Oh, we try." So, he knows he's bad at this stuff, but he doesn't pretend he's good.

[00:08:29]He looks at the situation. He saw, "Wait a moment. This is a scaffolding problem.

[00:08:33]Scaffolding is non-trivial because you need to do things no one even knows, right? There's some crazy stuff prompting yourself to death and then testing of this whole thing." Cursor is the most advanced scaffolding company on Earth. Better than Anthropic. They just didn't have an integrated system. They didn't have LLMs. They were missing key pieces that Enthropic has. That's why Enthropic beat them. Elon connected the two. It's like, wait a minute. I have a pretty decent LLM. Maybe even a very good one. I don't even know because LLM is purely they can't be used that much, right? That's why we don't know exactly how good Grock is. I think Grock is probably pretty decent. It's probably very close to the other guys on an LLM level. Scaffolding is missing. Cursor, the best scaffolding in the world. They actually came very close to beating or on par with CHGPT and the latest claude model in coding using Kimmy which is like a Chinese definitely second tier model >> and cursor is on par with CHBD and anthropic with Kimmy and Kimmy is clearly behind their LLMs which means the scaffolding of curser is vastly better >> than even anthropics and now guess what yeah combine it with Grock and after what one month three weeks I don't even They are already like beating everyone >> with partial with just only partial data from cursor. Yeah, Grock is use just starting to use cursor together and they're already >> they gave him the NLM and they gave him the compute and then cursor brings in the scaffolding and I want to just emphasize this to people. It's like uh scaffolding is like human civilization.

[00:10:05]It's like saying it's just the brains and the scaffolding is the civilization.

[00:10:09]Okay, what is more important? The brains or the civilization? I'm not saying the brains are unimportant, but the brains are the same as a bunch of Neanderthalss or like homo sapiens 100,000 years ago.

[00:10:20]If you put them into a data center, suddenly they have AI. So clearly the scaffolding matters, right? Just so people understand that this is not just a little bit piece of >> and now of course it's not just amazing that Elon understood that understood his weakness understood what makes anthropic strong that is the combination of scaffolding LLM. He also spotted the unique arbitrage in the market that there is this little not so little like the largest deployed coding system cursor that was completely destroyed valuation wise. I mean, not complete, but it was like out of favor because I was like, "Yeah, what is that?" Elon got it, went to the >> It's not It's not cheap. In order for SpaceX AI to own Cursor, they have to pay $60 billion.

[00:11:03]>> I know. And so to buy, >> but went to these people and say, "I want to do a little experiment, give you some access, and buy >> Yeah.

[00:11:12]>> and buy an option, a call option >> for $10 billion."

[00:11:18]Okay. Who does that? So uh OpenAI tried to buy cursor for three billion by the way. They try to buy cursor for three billion. Kurser said no. Elon comes along like I give you 10 billion for a call option to buy you for 60 billion.

[00:11:31]And K like oh and you're Elon and you're giving us best compute. So they said yes but even with the knowledge that I just dispersed here most of the other CEOs would not even now do the deal. So you need to be Elon. Of course Elon has the advantage of SpaceX being driven to two trillion. So that's just what 2% which is insane.

[00:11:52]>> Yeah. I don't know how he's I I see if he's if he buys them through stock instead of cash. Yeah.

[00:11:56]>> He has to buy them through stock because the option expires 60 days after IPO. So this is happening now.

[00:12:03]>> He has to execute the option 60 days after IPO which is June June 12th. So he has to execute by August 11th. He has to execute the option and of course with stock. So >> and if he does it sooner actually then it it rises up. Wait, no. Maybe he wants to do it afterwards when it's already the value is already at 60 billion worth as opposed to too soon. Okay, that's why he did. That's why he's saying 60 billion but later.

[00:12:29]>> So, so this is all and the cursor team by the way is super smart. But it's you have to connect. You have to understand these things that I just understood because it's like not easy. It's all very new. But the scaffolding versus LLM then the combination and of course compute it's another dimension but it's mostly ALM scaffolding plus compute.

[00:12:45]It's a perfect m it's like completely perfectly executed like because it's not just the perfect missing piece. It's also the best in the world that at the same time fell out of favor because everyone thought cursor is weaker than anthropic. They're losing it because they didn't connect the dots that they are by definition using Kimmy and being nearly as good have proven they're better at the core competency of anthropic which is scaffolding than anthropic itself. And all you need to do is add the LLM, the compute. So >> yeah, >> when Elon says check back in in three years, there's a high probability that you do not have to wait three years.

[00:13:24]There's a high probability this happens in the next 12 months because the cursor team is very good and they were held back with these bottlenecks that they could not solve LLM and compute. But Elon has >> and there's Grock build which is I think the the Grock version of compute cloud code and AI cortex or OpenAI's u what am I forgetting his their name of that but uh yeah AI open's version of their coding system um >> codeex >> I forget codeex yes I forget thank you so you know grock build is just started and they're behind but thank you for that that it's absolutely a good point that you know curs giant curs is also deep here's another thing why c I mean this is just a complete win because you have to like what we are building out in pioneer lens I call it the knowledge graph I first called it the wisdom graph but that sounds stupid so it's the knowledge graph which basically graphs out all the core uh components of this AGI revolution space infinite energy compute then AGI core the AGI capability and then what does it do it does three things What is the business model of the whole thing? The business model is to replace human labor, drive down costs to 10% of what it is now, and then scale labor up infin in like infinitely because everyone needs more labor. And that falls into three categories.

[00:14:48]Digital human, the digital human, which is the first one to go, which is all the white collar. Then drivers, which is a separate category because it's so huge as jobs. And the third one of course is uh embodied AI. It's like robots. But it happens in that order. It happens with the digital human first. And people have completely ignored this. Everyone talks about robots taking order. Dude, this takes 10 years before this happens at scale. The digital human starts now.

[00:15:14]>> Yeah.

[00:15:15]>> Mass replacement of all. And now in order to roll out the digital human, you first need it. Okay, that is what we just discussed. It's the scaffolding, it's the LLMs, and it's getting to critical capability which enthropic already has. That's already AGI economic AGI they now now SpaceX will catch up very quickly in the next six months I guess they will be well in the next three months they will be where anthropic is now anthropic will then be ahead but they both crack it and open probably too then the next question once you have the capability replace every lawyer doctor like everyone like everyone you still need to replace them it's not enough to have the capability you need deployment power what is deployment power it means you need to be in every enterprise S&P 5 fortune 500 midsize that is why we have these private equity deals now where the private equity companies partner up with OpenAI and entropic because they own all of middle America uh so you have enterprise then you have private equity and that's all you need plus venture capital then you can go hardcore after everyone it's a little scary but this is going to happen >> my I predict that Elon will partner with uh either Salesforce force or Oracle, somebody's already embedded in enterprise.

[00:16:33]>> But Herbert, here's the trick. He doesn't need that anymore because who has the absolute lead in forward deployed engineers across Fortune 500?

[00:16:42]Cursor. Every single >> Oh, yeah. That's true. Developers have already adopted cursor first.

[00:16:48]>> True.

[00:16:48]>> Yeah. Every single Fortune 500 company has cursor.

[00:16:51]>> Yeah. The developer.

[00:16:52]>> So, it is also a complete victory on the enterprise sales side. He already has them. He doesn't even need to sell them anymore. They already are locked in.

[00:17:01]They might also >> he'll pay them 60 billion own cursor for all the reasons you said. Um yeah, >> but what let's talk about the rental part of it. Uh so so you know the ju just to finish off the story here.

[00:17:12]You've got Grock Foundation. Remember how he's got multiple models running at the same time because he has so much compute. He's able to rent some out and but he has uh multiple. Here's the 1.5 trillion parameter one. He says that one is finished training. Evaluations look good. A lot of cursor data was added in supplementary training and there's more to come. Fine-tuning is underway.

[00:17:37]Reinforcement learning begins in a few days, two to three weeks, the public release. And then the.5 trillion model, that one is currently serves all Grock production traffic, especially for difficult coding tasks. We will open a 0.5 trillion model towards the end of this year. It should still be quite useful.

[00:17:53]>> By the way, it's not it's also not just cursor data. People have to it's not just getting data. It's it's also the question what is that data? So cursor is not just data. It's also aligning this entire development operation to where it matters most to the Fortune 500 teams.

[00:18:11]>> They they actual value that is already being generated. So >> if it's already integrated Yeah. Right?

[00:18:17]You you deploy the power of of these uh.5 and 1.5 trillion parameter models of Grock now fine-tuned and trained towards the highest value you can generate because of cursor. So cursor is very important.

[00:18:30]>> Wow. So so just a reminder everybody most companies out there they have one of their frontier LLMs they're working on and then they're training the next generation one and their advantage is that there's recursive learning. Now that that one that they created is teaching is building the second the new one. But these guys are building multiple models. They have multiple shots on gold. They have different um goals for each one of them. The.5 trillion probably answering quick questions. The 1.5 trillion who knows maybe it's what you're saying the cursor one that's enterprise level. They have multiple because they have so much power. Um, and then when it comes to how the rental, you guys obviously remember Claude uh, Anthropic partnered with SpaceX and this guy calculated and he said, "Okay, if assuming he just did his napkin math, assuming Colossus one has 220,000 GPUs and then he broke it down to H100s, H200s, and GB200s. This is the typical pricing assumptions. He's figuring out that basically they're they'll pay5 $5 billion of annual revenue to XAI just for the Colossus one. But but uh that filling $5 billion to XAI Anthropic is paying them. But if you remember Dario, the CEO of Anthropic was on the Dark Cash podcast and he said that whenever you have compute you have to make a tough decision. Do I take my compute to build the model to make it smarter or do I take the compute and make it inference? so that I can charge customers and have more customers to use it more and make more money. Well, that $5 billion that he's paying XAI's inference, that will translate to $15 billion a year in revenue to Anthropic.

[00:20:13]So, it's a 60 to 70% gross margin. So, for Anthropic, it's a it's a win for them. It's a win for XAI.

[00:20:22]>> That by the way, >> what's that?

[00:20:26]>> No, exactly. So it's 15 billion they're paying not 5 billion.

[00:20:30]>> Yes. Well, so yeah, Freedom Dwan calculated that Colossus one is $6 billion. Okay, that's true. But SpaceX is or anthropics paying them $15 billion a year. So that 9 billion is on the Colossus 2 part. And she just calculated thinking that actually that's only onethird of what Colossus 2 actually has because Colossus 2 has 550,000 GB200s.

[00:20:53]And this is it implies that Anthropic is renting just about 150 to 200,000 GP200 which is like a third of that which means that Elon still has 2/3 either for his own use or to rent it out even more.

[00:21:09]So this $15 billion a year is just the beginning. It's not full yet at full capacity. So he thinks she thinks that you know it's like oneird yeah of total compute. Well, the long story short is there's infinite demand for compute because every piece of compute you buy, you can sell for every dollar computer you buy can sell for $3 effectively. So, how much of that would you buy?

[00:21:35]Unlimited obviously.

[00:21:36]>> Keep buying. So, >> yeah. And you can use it yourself too.

[00:21:39]That's the point.

[00:21:40]>> I mean, if you were the seller, but >> Yeah.

[00:21:43]>> Yeah. This strategy by Elon is as brilliant as Amazon's AWS, right? you you need it for your own services and so they built out this data center for their own e-commerce services because in December they have huge spikes so they overbuild but then they don't need all of that during the year and so then they start renting it out to businesses who don't need to build that because they built it and then they make it pays for it pays for you to build even more and then it pays you to even build even more >> only I have to interrupt here because that's not even the strategy because I wouldn't call that a strategy That's like saying, you know, your strategy is to mine for gold. Like, obviously, that's a good strategy if there's a lot of gold. So, that's a no-brainer. The real strategy is not a strategy. The real strength of Elon is he can build these data centers because it's not that no one else is trying.

[00:22:37]They're all trying. Stargate, everyone is trying to build them. The difference is that is Elon's core strength. He just builds them in two months. The others don't build them in two years. So this is a pure execution advantage. It's not that anyone else doesn't want to build all these data centers. Elon built them and this ability to actually build them and deploy them which is a core competency. Even the biggest Elon critics admit well that's the one thing he knows how to do. He gets it done.

[00:23:06]Giant complicated engineering projects.

[00:23:09]And same with space. That's not an ingenious strategy. It's a no-brainer.

[00:23:14]The problem is no one can do it. So, I mean, you know, that's that's the best type of business. A total it's like it's like eternal life. It's not a great strategy to develop eternal life. It's it's the unique ability to do it. It's not, oh, I have an eenous idea. What if there's a pill and you don't die anymore? Oh, Joe, you're so smart. It's like, no, that's not the smart part. The smart part is to build the pill. So, so that's pretty much what's going on here.

[00:23:43]>> But it's a flywheel is what I'm saying.

[00:23:45]Because the money that you spent building it and then you want to use it for yourself, but then in the meantime you're renting it out. That gives you more money to build even the next one for yourself too and rent it out and build build. That's why scale is >> but the key is you need to be able to build it >> like the Stargate people wanted to do the same only that yeah >> it didn't build. Why is it not built?

[00:24:07]And this is it's very important because you know this is always the magic thing.

[00:24:12]You need very deep strategy, first principal thinking in the beginning, but then you need to execute. And execution is like this. That's where the magic happens because you can have 10 people with the same brilliant idea and then they just all fail except for one. And look at Stargate. They they did all this stuff. They started digging out all kinds of things. What TSMC in Arizona is always the same story. It's like, oh, and then it takes five years. And when Elon does it literally like Yens No said that N was like this was the insaneest thing I've ever seen. They just built this biggest thing ever in two months and and Yens was completely blown away because he knows that stuff. And so I think this is exactly why we pay Elon a trillion because it's just okay. This is not even strategy. This is just you literally cannot do this if you're not Elon. Uh yeah, apparently uh when it comes to the the the solar pan the the AI sat many satellites by the way uh I think the number the date I heard is 2028. So I don't think we should be trying to expect uh data AI data centers coming out sooner than 2020. I think even this X1 the S1 filing said that. So uh but he said of Pier Farerugu asked the question hey on the um the picture below the AI satellite mini seems to have three to 400 kilowatts of solar panels for 100 kilowatts of compute the pee of 1.2 2 35% panel decay over the lifetime of the bird barely get me to 200 kilowatts. What am I missing? And then Elon said, I will present a more detailed AI satellite design in a few weeks. So they're working on it and he's confident he'll be able to redesign.

[00:25:53]It's it's it sounds very similar to, you know, how when he was working on the on the chips, he had to redesign the chips and he spent weekends. Sounds like that.

[00:26:02]That's what he's doing now. The AI AI satellite is now >> totally hit this out of the park. This is like if you know Elon and what he does best and then you understand the engineering challenge of the satellite which is exactly in his ballpark. It's not Einstein level. You there's everything is known about it but the total complexity of is like medium to high. Uh when it comes to commercial scale it's not like Starship is not medium to high. Starship is like extreme right and FSD is mega extreme and AI is like infinitely extreme. So you can't force it. It's like decades and then at some time you get lucky. Here we're talking medium to high which is Elon's sweet spot. The data centers are I would say medium also medium to high probably.

[00:26:46]Uh but this is absolutely doable but it has a high level of complexity and you need to just nail it and I have 100% confidence that they will totally blow this out of the park and win. And I said 2027 will be the first deployments of like proto like >> maybe one or two get to stress or more than that maybe hundreds or something to just test them.

[00:27:10]>> He said 2028. Okay.

[00:27:13]>> Well 2028 I think will be deployment.

[00:27:15]That's where you have actual data centers. I I think you need to have like a few dozen or something in space just to measure stuff like you know decay all kinds of stuff. So I think that will happen in 2027. I'm pretty sure. I'm pretty sure we'll have orbital AI space satellites prototyped in space next year.

[00:27:35]>> Yeah, I think prototypes will be there.

[00:27:37]He actually said just a few days ago, he said it's a lot easier than people think. AI data centers is a lot easier than people think. So that confidence and then he's designing it. He's a more detailed satellite design in a few weeks means that you know it's uh they've been they feel confident that they've got a design >> like out of the risk and complexity the AI satellites are 5% 95% is Starship because Starship we are close but this thing is insanely comp >> I was going to ask you that uh so Friday huge success Starship 12th launch uh so you were saying just like you were thing with FSD does it work. This thing rests on this Starship flying and it was a pretty well a 100 almost pretty close to 100% success. It >> well I would say Starship is actually more is for me more transparent and predictable than FSD used to be. It's still not transparent to me. FSD is just works somehow. But the the Starship situation is very clear to me. I did the math on it. You basically have all these different conditions. I don't know if you saw the video where I I was bad in screen sharing so I didn't share enough of the model I think but this is basically the the majority of risk lies in launch explosion because there is where the money is and that's where you that kills you on the insurance side even selfinsured. So if you blow up this thing with the cargo in it it's super bad not just for that flight but the next 20 flights down because you have to pay for that in anticipation you have to reserve them is bad. So the most important thing is don't blow it up especially with test cargo like this time. So that was total green. It was not total green. They had like a engine failure. One out of 33 or something but doesn't seem to be risky.

[00:29:26]It doesn't. It's not great. There's definitely like a little problem there but it deployed. Number two does it deploy the cargo? Because if it doesn't deploy the cargo it's also a total failure, right? It deployed the cargo very successfully. That's very good.

[00:29:40]then um the booster rellanding failed.

[00:29:43]But this is much less dangerous because this just hits your economics, right? If you if you launch a booster, you fail to catch it. As long as it doesn't destroy the turret, it's a little expensive. But if just the booster explodes, who cares, right? If you use a booster 10 times, you divide the cost by 10, it's roughly, what is it, 80 million or something per booster. So it goes down to 8 million.

[00:30:05]If it blows up after five times, well, then then it's like, you know, 16 million. So, it's not totally changing the equation. If this thing blows up at launch with a billion dollars of cargo, you have a problem because you're increasing the next 10 launches by 100 million each. So, that's like a little bit the math. It's the most important thing is don't blow it up at launch. The next most important thing is catch the booster. And we count how often. Once you go above 10, it doesn't even matter, right? Because 10 reuses is 8 million each flight. 20 reuses 4 million. So you basically if you fail at the 10th relaunch, you just basically double your from from 4 million to 8 million the cost. And then the last thing is technically the most challenging but also the least important. It's catching Starship. It is very important long run to really get to very low. So, it needs to be done, but it's much less critical.

[00:31:02]After I ran the math, much less critical than anything else because Starship only costs 40 million or something.

[00:31:10]So, if you don't if you can't reuse Starship because you can't catch it. It can't reland. It doesn't even matter.

[00:31:15]Then you just pay 40 million per launch at 150 tons versus 70 million launch cost at 15 tons right now.

[00:31:24]I mean, that's ridiculous, right? So I people have to understand Starship doesn't even need to be reused for orbital AI to become economically feasible. That was my biggest revelation when I modeled it out. Plus it's pretty safe to say they will reanded like by next year or something. So but it doesn't matter if it blows up every second time. No one cares. Like as long as it doesn't blow up up, that's a huge difference. But on the way down, if you fail to uh reuse it every fourth time, >> it's not a big deal, >> right? Yeah, >> it's really not a big deal.

[00:31:57]>> I'm curious when Elon said that. So, first of all, he said, "I will never give up. We will have the best AI." So, that was great to hear him, you know, double down on that. Then he said that, let's see where things stand 3 years from now. So, three years from now would be 2029. So if 2028 now you're saying you know pro AI data centers out there 2028 2027 but 2028 for sure pretty confident especially with the success of the starship this Friday and Navy he's got 10 more lined up by 2028 you know what I mean that's when they'll be at least once a month a spa a starship if not more >> more definitely more >> can you believe that and and then the payload of how many AI data centers will be out there. I think that that's the scale that he's thinking by that point.

[00:32:50]>> It's kind of hard to beat us.

[00:32:52]>> My prediction is it's always important to understand when is Elon over over confident with his timelines when he's late >> and when he's not late. When it comes to the three years on the digital human on the AGI thing, >> I think Elon is off. Elon is totally overestimating the time because here we are in a situation where you have brilliant teams. He knows how to build out this compute. The Grock team is decent but the cursor team is brilliant and the cursor team can handle this whole thing because the cursor team was missing that compute and the LLM access.

[00:33:30]Now they just have a I would say a second tier team. I don't want to be too mean, but it's a 1.5 tier team on the LLM side, but Kursa is the best team in the world for scaffolding, which is much more valuable, better than anthropic, more experienced than anthropic, and deeper embedded in the entire enterprise world of the world.

[00:33:49]>> They are going to take charge. That's also what Elon likes about them. They're very young, by the way, 25. This guy is probably the youngest, richest man ever.

[00:33:58]25, 60 billion by. But good for him. So they are very smart cookies and they have that under control and the leap to overtake Anthropic I think within months they will be on their heels like seriously on their heels. So I'm very bullish on uh Space X AI now becoming a really top tier model together better than much better than Google of course but as open AI an entropic by end of the year this will be a tough competition >> like beating that's a whole different topic but they will be very very close enough that no one cares. Yeah, I think so. If Elon was so concerned that XAI has run away with the race and have won and are now going to get to AGI and ASI, he would not have helped them. He there's huge value for them to get the 5 billion so that their IPO goes successful. So, you can kind of like think about that. But ASI AGI, no, I think he's he realizes you're not going to get there until three years from now when you >> I think I know trick is like an amazing story. But I know that trick because they also leaked their main thing. I think they were just good in prompting.

[00:35:04]That's what they are. They just figured something out and that's it.

[00:35:06]>> Oh, yeah. No, for sure. I mean, you can, you know, you compare it to Codeex, which is OpenAI's coding system, and you can see there's a huge difference between personality. I mean, you know, you know, I'm using OpenClaw and I was using Claude and it was brilliant, but it wasn't just the intelligence as you said. It's a lot about how it answers and how it communicates with you and makes you like it won't give up and it'll try to solve a problem versus codeex would just stop working and I think or it'll lie to you. It'll say oh it's doing something I'll take care of it and then like later you go did you do it? No, I didn't do it like but you said you do you said you did it.

[00:35:45]>> And I remember this the throwing was totally behind in the beginning. I think the LLM kind of sucks. It's not very good in my opinion. Now we maybe it gets better but but the way to see that there's not a direct analogy we have to think in basic fundamental principles more but these LLMs they are like raw intelligence they're like your frontal cortex >> right that you can get a highest IQ person but doesn't mean that they're going to be the best whatever because they're just just pure IQ >> and it doesn't even come down to a person you are talking about giant virtual organizations right you have I see that when we work on our stuff here with our agent in pioneer lands pine lands daily on new publishing thing this is I don't even fully I have no transparency no one probably has how many agents are working on that but the agents spin of other agents we give them instructions about how to you know uh curate news and everything and rewrite stuff and do videos and it's the same thing with these companies like anthropic has probably it's hard to tell how many millions of these agents organized somehow that they don't even fully understand and that constellation of prompts and agents working behind the scenes makes actually the thing and the question whether OpenAI has 155 IQ and Enthropic 145 completely irrelevant. If they deploy it in two million wellorganized brands and they deployed in 10,000 badly organized brands, your 10 IQ points don't help you. So >> I think that's why cursor they are basically the magic organizational superpower without these IQ points or they're using Kimmy. give me is probably 120 whatever.

[00:37:21]>> Oh yeah. The other point I want to make is that uh what Elon is renting is the inference which is allowing anthropic to charge to make money but not to use it to build smarter models. That is reserved although they I think onethird of Colossus 2 allows them to build smarter models maybe not inference. So that's just to >> yeah remember that that's the >> like no one even fully understands any of this including Elon in my opinion.

[00:37:45]Like no one truly understands the exact setup of this thing. It's more trial and error and you know they are LLMs with IQ and there's this weird unknown scaffolding super structure uh and the truth will be in between and once you can deploy the super structure plus the LLM to somehow recursively improve but I think no one really figured that out yet it's only little pieces >> pieces >> then you're dealing with something huge so that's all happening and that gives you know it's it's literally we have to think about this AI thing really like giant civilization ations not like a giant super brain it's not one super brain it's millions yeah that help each other agents >> and what are they doing and how are they organized and just if you imagine I give you a million people I give me a million people even if view yours are five IQ points higher if I am better in organizing I >> better system >> right and so what is that system and that's no one really you know this is like vague because it changes by the minute because they're spinning off more like this is a complicated question.

[00:38:49]>> Yeah.

[00:38:49]>> But that's how to think about that's why you can beat >> well the best system and I know you've been thinking about this with pioneer lands is model it after democracy and agency right you give your citizens agency to create businesses be be pro-commercial businesses you know reward everybody properly that's how you do it. So if agents are able to decide on their own that they're going to do something then and then you reward it properly maybe that's how you build the best >> exactly and that's how we have to also think about ASI it won't be a super brain it will be kind of a civilization and it can and the difference to humans is if I give you a million humans to make two million humans takes 18 years and some dating apps but with these agents it takes a minute and they can decide is is two million actually better or is one million this is extremely amorphous and like it's a new paradigm but that's how I think about it because that's like the most helpful you know juristic >> okay always fun to talk to you thank you so much Joe if you guys love this conversation check out pioneerlands.org This is what they do. This is what they talk about all day long and they plan the future. Thanks everybody.

[00:40:00]>> Thank you.

[00:40:00]>> I've created a website that is the most comprehensive resource for the Tesla investor. Please check it out. Simply go to my website at herbalm.com.