Terafab - Growth Engine for SpaceX and Tesla

Kind: captions Language: en Hello, hello, hello. Hey, this is Warren Redley. Thank you so much for watching. You know, you think your comments don't matter, but I posted a video and somebody said, "Hey, there's a video about about Tesla and SpaceX potentially merging." And somebody said, "The irony talking about what you don't want to hear, what you don't want, I hear instead of Terraab." So, I should make a Terapab video. Good point. Down for it. So sick of the endless merger talks. This is your fault. I told them, "Hey, this is your fault. I'm making a video about Terraab." So, this is a video I really haven't done a good indepth this is why Terrafab matters. This is what's really going on. So, we're going to dive into that right now. Are you ready? Let's go. And where we're going to start, why why do Tesla and SpaceX need Terraab? We're going to talk later about this structure, like what's Tesla's role, what's SpaceX's role, but fundamentally like why do they need this? So if you understand the game plan, Elon wants Tesla to make, let's say, 100 million Optimus robots a year, maybe more. He wants SpaceX to be pumping out many gigawatts of orbital space compute. Many gigawatts, which require both of these require a lot of chips. Optimus requires a lot of AI5, AI6, AI7 chips. The orbital compute might be using AI6, AI7 chips. It might be using D3 chips. Maybe it'll be using Nvidia chips. There's all kinds of possibilities. And one of the fundamental problems what Elon sees, this is my big thing. What does Elon see that the rest of us don't see, which is a big question in everything and that in the world of Elon, he sees a future where Tesla and SpaceX need vastly more compute hardware than the semiconductor industry, the broader supply chain for AI is planning to produce. Like, they're not even close. They're not planning to scale up anywhere near enough to what Elon thinks Tesla and SpaceX need for their objectives for the future. Just check the chat really quick. Okay, just making sure we're we're on the same page here. So, and this isn't just like a TSMC problem, right? TSMC is not growing as fast as Elon wants. Elon also uses Samsung for its fabs to make um the AI5 chip is going to be uh the AI5 chip is going to be both Samsung and TSMC. Other hardware comes from different places. They're getting memory from Micron. The entire supply chain for AI depends for example on ASML which is a company in the Netherlands that makes electro extreme ultraviolet lithography machines. the fabs run on EUV machines and if you don't have enough EUV machines, you can't make enough chips. And so at many layers, Elon sees the industry just isn't scaling the way I think they need to scale for the things that we want to do at Tesla and SpaceX. ASML is an important bottleneck. Uh TSMC is a bottleneck. Samsung doesn't Samsung is a player in the fab world. Intel's a player in the fab world. So Tesla's demand comes from Optimus and not really from cars. the AI4 chip they're already making probably enough. Maybe they'll put AI5 chips in cars, maybe they won't. But by the time Terapab, this is Terapab, by the way, and behind me Terraab. By the time Terrafab really ramp, Terrafab really ramps up, they're probably going to be past the AI5. We'll talk about that. So, I see Optimus, let's say it's 100 million Optimus a year, right? I mean, they're going to make one million a year in Fremont. That's the plan for the Fremont uh facility. Then in Texas, they've got a a factory they're building that would be able to do 10 million a year. But long run, if you think the world needs billions of robots, then a 100 million, let's say the world needs 10 billion robots, and you're making a 100 million a year. It takes a 100 years to get enough robots. So you might be doing more than 100 million a year, but at 100 million a year, you need 100 million AI 6, AI7 chips a year. And the industry just isn't scaling to make that much. And then on the compute hardware side, on the terrestrial compute for um AI data centers on Earth and the orbital compute, you know, Elon's talking about terowatts, you know, maybe one terowatt a year, something like that. And the industry is not getting close to the amount of compute that you need for that. So I think the plan is the AI series of chips for inference compute that goes in Optimus and the D3 D4 whatever dojo series of chips that's going to go in the orbital compute. I think they're probably also going to put inference compute in the orbital hardware and the inference chips in the orbital compute as well. But we're going to see and that's one of those things that'll develop over time. the I think the SpaceX need I think Elon said was 75% of the what Terapab will produce and Tesla would use 25% of what Terrafab would produce. So Elon sees it's not simply getting better and it's not just about prices. It's like you can't make enough stuff if you don't have the stuff to make the stuff. It's a little bit of a twist on Elons. If you don't make stuff, there's no stuff. um they have to create the additional manufacturing capacity themselves. I think they still buy Nvidia hardware even though it's more expensive once Terraab is ramped because their demand is so great. They're still going to want more. So what are they going to make? Let's just check the chat again. How are we doing here? Not just AI6 chips. Oh, it's it's a bunch of stuff. It's um Sorry, let me I don't I'm going to get into what they're going to make now. So AI6 and successor inference chips AI, you know, AI5 in theory, but I think they'll be on to AI6 by the time they they get ramped up. Um, there'll be AI6, there'll be AI7, there'll be AI8. Future designs of chips that are inference chips that go in Optimus and maybe other friendly shaped robots or who knows um and maybe maybe it'll be in computer or orbital compute or other compute as well. Um, so that so optimist is like the huge volume application from Tesla. you have D3 successor data center chips. Um, and you know, you get some chips are used for training, some chips are used for inference. Elon said, you know, sometimes inference chips might not be ideal for training, but they cost a lot less and you have enough volume. They're still okay for training. Then in addition to the GPUs, right, you need CPUs which sort of coordinate how everything works, runs the agents, that kind of thing. You need controllers, you need networking silicon, you need other specialized chips, you need memory. Uh they he said that that Terafab is going to make memory as well because the memory industry is not scaling up as fast as he wants. Packaging, co-as, wafers, there's a lot of things that go into this. And if I understand correctly, the idea is Terafab is going to be basically one huge vertically inteerated operation that's going to make all of this stuff. even maybe replacing the EUV machines with uh free electron lasers. That's a whole another story. Where are we? Markplot in the house. Awesome. We have Markplot. Good morning, Mark Plot. Or good evening for me. So, uh we are we're not we're not at my bedtime yet. So, we're good there. So, how big is Terra Fab? I want to show you this image. Oh, before we get to that, just real quick, if you are a YouTube channel member or an ex-subscriber or a Patreon supporter, I've given you access to my latest video, which is this is Congressman Ro Cana. He's running for president on the Democratic ticket and he was interviewed by NPR. Um, he's pretty much taken shots at capitalism and capitalists and he denies being a communist, but he's obviously a communist if you listen to what he says. Almost comical in some cases. Some of the things he said, you're just like, "What? He he's not he's not making any sense. But so this video is now available if you are a YouTube channel member, if you are a Patreon supporter, if you are an ex-subscriber. So check that out. The video will be public in about 24 hours. So um here's Terapab. I'll play that video later. So you have Terraab. One of the things that's going on is there's a lot of people posting images of Terapab. You have this one and you have this one. And it looks like terrab is ridiculously huge. You know, Giga Texas is almost a mile long. And this thing looks like it's 10 miles long. And I discussed this at length with Gemini and Chat GPT. And I think this is a massive exaggeration of the size of Terafab. And it is huge. But you know, 100 million square feet versus 10 million square feet. Um, when you consider the width and everything, it might be 2 and a half miles long. This is making they're making it look like it's 10 miles long. So that's one little detail. just be aware of is they may be exaggerating the size of it a bit. So, um be skeptical of the 10 mile claims. Um but the the thing is it's enormous. Uh it may be very very large. The exact dimensions aren't completely clear, but it appears on the filings it's 2 and a half miles long. Um and those compar those graphics are interesting, but they're not really useful. So, you get to this question, how is Terraab going to be built? You can see there's four sections to Terraab and actually the filings that were submitted to Grimes County uh where it's being built had this idea of four phases to the project. So my sense is that what we're going to see is uh Terraab is going to be built in four there's you can see the four segments. So the first segment will be built first. Let's just check the chat. Where are we going? 14 forces. My only concern is having all of it in one area because bad weather can impact all operations at once. Tornadoes and flooding are events to worry about. I am pretty sure that this building will be tornado safe and they will have all kinds of things set up to protect it from flooding. And when you build a facility like this, you make sure it's solid. You make sure it's ready for anything. That's you. When you make an investment of this scale, there's things that you do to make sure that it's safe from the the the major threats, you know, the the hundred-year threats. So they'll likely build one section at a time. The four phases fits the visual design. I think they're going to build what what they'll do is when they build the first section, they'll build the shell quickly and then they'll finish the inside progressively. We've seen them do this with Giga Texas. Put up enormous walls roofing. I I don't mean they're going to build all four sections right away. I suspect that they'll build one section at a time. They might not put the fins on right away. You know, they'll the the final product might take years to get to the to the look that we see here. the road will probably be there because that's pretty straightforward. But, uh, they build the shell quickly and then they start finishing in the inside and filling in the machinery that goes in, installing tools, utility, production lines, and that takes longer. Now, they move fast, but there's a limit to how fast you can move. So, the structure itself is the easy part for for Elon, you know, for El team Elon that's easy. For the rest of us, that would be hard. compared to the other things, whether you're going to have clean rooms, not clear if they're going to have clean rooms, the utilities and the electrical wiring, the pipes and everything, the process equipment calibration yield. It's not just pouring concrete. A lot of complexity there. So, um, when they build the first section, my hunch is that they will use conventional extreme ultraviolet lithography, which by the way is very advanced technology today, but by the time Terapab is running, they're going to be working on something on something else. So, and I think we have it in one of these images. Let's see. In this image, you can sort of see in the center there's a circle. And some people think that's some kind of cyclron or something to accelerate electrons or protons or whatever for the free electron laser. Not not clear if that's really happening, but that that's one of the theories. But I don't think that they're gonna have the free electron laser thing ready to go on day one of Terraab. So they may stock up the first segment of Terraab with EUV machines. If they can get free electron lasers working that quickly, great. Free electron lasers um are the problem with E. There's a lot of problems with EUV and free electron lasers could end up being less expensive, more effective, less problems, um and you know, ability to do a lot more. We'll see if they go there. Elon did say free electron lasers for the win. FTW. So Elon seems to be on board that we're going to be doing that. Hello Carsten. Just checking the chat here. Terafab trucks with laser beams on heads. Sure, why not? One way they bring all the people on board is to build out so much solar and storage to provide cheap energy for residents. They are building solar, but that's not part of this project. So I'm I'm not going to get into the solar thing. That's a that's a whole different issue. I think that's in um that's in in Bastram, Texas near Austin. Um dis the the free electron systems can be very large. So, my gut is maybe it's underneath. It's not clear how they do it. I think that has to be it's it's a long tunnel. It's not the circular thing. It's a long tunnel. Um but we'll see. You know, Elon has big ideas. We'll see what Elon's able to figure out and they'll try a bunch of different things to see what they can make work. So, vertical econ vertical integration, everybody knows SpaceX and Tesla are big on vertical integration. It changes the economics. Just to be clear, when you're looking at the financials with a lot of things when I talk about Tesla and SpaceX, there's a lot of things where I get into the napkin math of how much revenue, what's the unit of the product. When I do napkin math, I think of what's the product that this represents or what's the service the company's going to offer. Maybe it's Starlink and you have customers, maybe it's robo taxi and you have either robo taxis themselves or paid rides or or paid miles. um you identify a unit, what's the revenue per unit, what's the cost per unit, and let's figure out what that's worth. Tarap, at least as we understand it so far, isn't really a revenue business. Terapab is an enabler. So that the revenue business, which is robo taxi and data centers, that that can scale and it's also a cost. It also lowers cost. So first thing is it removes the supply chain ceiling that happens if other companies don't scale like they need them to. And the second thing is there's a really really dramatic lowering of hardware cost. So I want to get into that point. The vertical integration removes layers of supplier gross margin from the cost stack. And this is a really big deal when you think Nvidia has a 75% gross margin and TSMC has a 50% gross margin and the memory companies have really substantial margins and the networking companies have really substantial margins. A lot of the stuff that goes into these chips isn't the cost of making it. It's the cost of paying margin to the company that does it. So, if Nvidia sells $100 of hardware at a roughly 75% gross margin, which was last quarter, only about $25 of that is Nvidia's cost of making the product. Um, and then so if Tesla and SpaceX are able to sidestep Nvidia, they save that 75% gross margin. TSMC 50%. If the Nvidia's cost, some of that 25% that Nvidia is paying is going to TSMC to make the hardware. So bringing that fabrication inside inside that that saves on that. You don't just add them, but it may be like an 80% or better lower cost per chip compared to paying Nvidia, paying TSMC, paying all these others. So it's not that owning a fab makes fabrication itself 80% cheaper. it's that it saves you on the mar on the gross margins of the companies that are supplying it to you. And then with custom A6 custo the the chips that Tesla makes like the AI4 chip in in my car, the AI5 and AI6 chips that are coming, the D3s that are coming, they're not general purpose GPUs like Nvidia makes. So you can specialize them and you can eliminate things that aren't as effective. And this is a tricky one because Elon really loves Nvidia's chips. So, I think Elon's going to continue to buy as many chips as Nvidia will sell them, even though they have to pay margin because that's going to end up being small on the scale of overall cost because they're going to get so much more from Terraab than they're going to be getting from Nvidia at that point. You got memory and packaging. There's all kinds of layers of, you know, what's going in here. And free electron lasers. Now, you're saving 50% margin on on ASML's EV machines. Just that's the that's the huge costsaving here is you've got all these companies whether it's Google or Amazon or Meta or Oracle the hyperscalers are building out these data centers and they're paying margin to a chip design team. They're Google does some of its own chip design. Meta does some of its own chip design, but most of that they're paying somebody like Broadcom, or you're paying AMD or you're paying Nvidia for chip design services. And then you're paying TSMC or Samsung for fab. And then you're paying Micron or Skhinx margin for memory. And you're paying this guy margin for that. And Kioia and SanDisk, you're paying for ND memory. And just go down the list. And there's some margin going to ASML and so on. So if you if you take all this out, your costs for the chips go down so much. And now we already know that these data centers are massively profitable, but imagine how much more profitable they are if you're able to eliminate all those costs. Like they've said that their payback on building out a data center is like less than one year. Well, if you lower the cost by 80%. Then your payback on the data center is like two months and then the data center runs for five, six, seven years. That's that that's why this is so so massively important. So, let me just check the chat again. If you're new here, I like checking the chat while I'm doing this. Uh, Markplot says, "I need Optimus today for maintenance of small chores." Yep, my wife wants Optimus right away. Packaging, there's all kinds of stuff that goes in around the chip. That's a little bit over my head, honestly. But there's co-as, there's wafers. Any comments on Ed Zitron's remarks on the AI market? Um, William, I haven't seen Ed Zitron's comments. There's a lot of people making a lot of comments on AI. I'll just I'll just I don't know what Ed Zitron said, but I'll just briefly address this. There's basically two attitudes towards AI on Wall Street today. The Wall Street attitude is this is growing too fast. The growth can't possibly sustain. The demand isn't there. And for those of us who use AI, and I would encourage anybody, by the way, whatever issue you're thinking about, whatever problem you think you have, have a conversation with an AI and say, "Hey, what could we build to fix that?" It could be something in the physical world, it could be a piece of software that you could build. And you might be shocked how much you can get done with AI. And I'm building stuff with AI. I have friends who are building stuff with AI. And it's going to get easier. It's going to get more powerful. and more and more people are going to build more and more stuff and then you're going to have real time video a real time video avatars and and real time video generation and we're going to have killer apps that nobody sees yet. There's all kinds of things that come when you have this kind of technology and you're you know agents are going to be interacting with each other. Robots are going to be interacting with agents. There's a massive volume of compute that's going to be needed. So either you believe the demand is coming or you don't. If you don't believe the demand is coming, then everything everybody's spending too much on AI and it's crazy and it's going to be it's going to blow up. If you see things the way I see them and a lot of other people who are into AI see them, the demand is growing so fast that supply has no chance of keeping up and that's why you need Terraab. Um, is it disclosed who all has invested in Terapab and how much? Um, it's not really disclosed who's invested in Terrafab directly. My understanding is roughly first of all Tesla is building the advanced research fab in Giga Texas. This is not a Giga Texas and SpaceX will be building this. SpaceX is going to be consuming 75% of the chips and Tesla will be consuming 25% of the chips are produced. So it's a safe guess that SpaceX will contribute more to the building of the Terraab than Tesla will. That's a reasonable guess. Um Intel is a partner. Is Intel chipping in? Will Nvidia invest in it? We don't know yet. Um, okay. Oh, and thank you for Mark Plot for answering. Packaging is capacitors, memory, rectifiers, etc. There's a whole lot of things that go into the these chips. That one's over. That's probably something I should learn more about because I hear packaging all the time. I should know the what's I need to learn what co-ass is. There's so there's so much like if you're watching this and like I'm not claiming I'm an expert on AI or I I'm I use it a lot. I use it for a lot of things. I understand it, I think, more than the average person. But really, this adventure, this AI stocks report that I'm doing is my effort to understand the the SEC sector better. And you guys follow me along for the ride and you guys learn along with me. And sometimes you guys teach me stuff and sometimes I teach you guys stuff. So, one thing to understand is tariff is going to be extremely expensive. The margins disappear into free money. Tesla and SpaceX and whoever invests in this thing are going to spend a lot of money building this thing out. Now, if this thing operates for a long period of time and you advertise it over 10 or 20 years, those costs get very small compared to what is produced. And if it's very efficient at producing, it produces in high volumes, the costs, the the capex, what happens is you invest a heck of a lot of money, I'm guessing at least hundred billion dollars to build the whole thing. But if it's producing a crazy, let's say it's producing a hundred billion chips a year, well then it's $1 a year. one you know the capex hundred billion dollars h 100red billion chips a year then the capex is spread out is fairly small it's probably not going to be that favorable but let's say it's a hundred billion to build it and you build a billion chips a year and they go 10 years that's only $10 a chip for the capex um if Musk only needs a normal normal quantities of chips terap's excessive if he really needs silicon for hundreds of millions of robots and many gigawatts of AI comput of AI compute the fixed costs are spread over an extraordinary volume. So it looks crazy under conventional semiconductor demand assumptions. It looks rational only if the demand projections which Elon has and the rest of us who who follow this have um if those if those projections are reasonable then the demand is reasonable and the the building is reasonable. So one question that pops in my head from time to time and people ask for you I'm doing better understanding AI than Don Lemon. I don't know if that's a compliment. Give us an example of something you've built using AI. I have a website called lcradio.com. lcratio.com. lc r a t io.com. I've been building that site. I am building apps to help me learn Japanese and Thai. So, I have a Japanese flashcard app, a a Thai flashcard app, and a Thai reader app. actually started with the tie reader app and then over the along the way I decided to add a flashc card feature to the tie reader app and then I decided the tie flashcard feature itself was its own app. So, I've been building those using primarily Grock and uh Codeex, Chetchi Codeex. Uh I used some Gemini and I use some Claude, but it's primarily those other two. Um and then I, you know, this this thumbnail, right, this thumbnail was enhanced by Chaty PT. No, I think um Grock did this one. Um so the thumbnails for my channel are primarily created with AI. um that you know I do it's a it's a mix. I use an app called I think it's Affinity. What is this app called? Affinity. I use an app called Affinity. Affinity. I used to use um Keynote, but Keynote doesn't do colors well. So um I use AI to help you know like the glow the glow on me and the glow on Terapab that was added by AI. Um and then I I have the AI help me. I don't script. I I hate script. I think there's a lot of content creators who are reading off scripts and I hate reading off scripts. So, I have outlines and I use AIS to help build an outline and I'm using I used uh chat tpt tonight to help me with an outline. I talk about what I want to talk about and the AI helps me shape an outline so that I can you know so that there's AI involved in this in this in this uh in this very live stream. I'm looking I've got chat GPT on the side here. I often use Grock. I often use Gemini. Gemini's you know because Gemini is owned by Google which runs um YouTube sometimes it has special insights on YouTube stuff. So then this question could terraab sell chips and other hardware to other companies. My sense is that the internal demand from Tesla and SpaceX is so large they will have priority and there won't be a lot of sales to other companies but excess capacity could be monetized and if that happens there's a revenue side of the story. Nvidia is potentially a customer. You know Jensen and Elon are tight. They text each other. Uh Elon just said really good things about Nvidia. Jensen's always saying really good things about Elon. It may be and Nvidia depends on TSMC to produce its chips. And it may be that they decide, hey, TSMC isn't growing fast enough to produce our chips. We're going to produce more chips with Terap. They they it seems unlikely in the short term, but at some point Nvidia's need to produce may be so great. And then Tesla partners with Nvidia. Nvidia cuts its margins a little bit of what's charging Tesla and they save margin on TSMC or Samsung for the fab and they save other margins and then Nvidia gets a lot of chips produced that they can sell make even more margin maybe from other customers. So um Micron you know memory maker they have enough fabs if this fab scales enough it may be that some others are going to be able to use this uh you know Google is a serious investor in terapab Google wants to launch orbital compute as well so it's possible they'll make orbital compute for Google as well um yeah so the the selling to others is to me optional if they if they have enough robots and they have enough orbital compute, they don't need to sell to anybody else and they won't sell to anybody else. But it may be that their scale is so good and their need is like maybe not, you know, maybe they only need 90% of it, they sell 10% of it and they make some revenue, but I don't think that's the important part of the story. Let's just check the chat here. Huah or Bangkok? I am in Krabby, Thailand. I have lived in Bangkok. I visited Haheen. Huahin Hua Hi Nuahin. I visited Huahin several times. Uh I lived in Bangkok for over a year. Um, yeah, I don't use an AI voice either. What would that sound like? I I thought I think Farzad's done that. I just it I probably should use AI scripts. I probably should use AI voice. I'd probably make more money. I'd be more successful. My channel would grow. I'm not doing this for that. I like talking to you guys. Could you imagine if I did a live stream and it wasn't me? I don't know. If you haven't noticed, sometimes on the live streams, not tonight. Sometimes the first minute or two is recorded. Um, but uh not tonight. So, I use Grock Build a lot. I use Grock Build a lot. I use Codeex a lot and I use Grock Build a lot. So, yes. Um, let's get back to where we were. So, Intel's role, Intel is partnering in some way. It's not really clear what Intel's role is, but Intel is a technologically a leading fab company. I think they're relatively small. Intel's had a lot of problems with their business in the past. a lot of bureaucracy, but maybe Intel's able to work through that and they're I think they're actually building some very advanced fab stuff. So, they're their technology going into terapab may be very helpful. Process knowledge, manufacturing expertise, a bridge into advanced node production, you're going down to 2 nanometer or smaller. Still not clear. Is Intel going to be investing money in terafab? We don't know. Um, so the other thing is the fin I don't know if we talked about this already, the financial side. Don't expect to see in SpaceX or Tesla quarterly reports or if they merge the collective report, don't expect to see a line for Terafab revenue. It may happen, it may not. Um, it enables Tesla to produce more Optimuses, lower robot and vehicle costs, higher growth. I don't I don't think I I don't think Terafab is going to make chips for the cars because the cars don't need such advanced chips. We'll see. Um, a lot of people think they're going to use the cars as like inference compute for everybody and I just think they're not going to bother because the cars aren't always connected. The cars are busy driving. I I don't know that that makes sense to use the cars for inference compute for, you know, others. We'll see. um leads to higher gross margins, lower selling prices, more AI comput, you know, the robo taxi ride gets cheaper if your cost of making the vehicle is cheaper. Um and if the chip is more efficient, then you use more less energy. So the the robo taxi rides, these are small changes, but they could help lower cost of robo taxi rides. SpaceX can build more terrestrial compute, much more orbital compute eventually building I think the chips that and the moon they build the structure of the satellite. Maybe they build the solar panels on the moon and they build the radiator on the moon and then they send chips up from Terraab to put them on the satellites that get launched from the moon. That's we're talking five 10 years from now. I don't think I don't think their moon's going to be launching satellites that quickly. Probably 10 years from now. Elon wants it to go faster. We'll see. You know, maybe a AI accelerates things. Um so you get lower cost per unit of AI capacity, much faster scaling. Capex comes first. Terapab could initially look financially ugly. We're seeing this right now in the markets. Wall Street doesn't see the long-term demand growth and they see all this capex and they say this is crazy. They're investing too much money and a lot of the the hyperscalers their stocks are low. Google's price is low. Nvidia, I think Nvidia is potentially one of the most underpriced stocks on the market. Google also very very underpriced. Um so that you know there's a lot of you know weird stuff going on with financials and mainly because Wall Street doesn't see what we see. um appreciation will matter, amortization, you know, there's a question, how long do the chips last? There was a sense that chips are only going to last three years, but H100s that are five or six years old, the price the price of renting the GPU is actually going up rather than down. So, it looks like the the demand is so high that even older chips will still be generating value, you know, five, six years after they're built. Um, so the question isn't uh how much revenue did Terarab generate. The question is how many robots does this enable the Tesla to build? How many gigawatts can SpaceX deploy? What happens to cost per unit? What happens to gross margins? That's the impact financially of Terrafab. Let's just check to see where we are in the chat. We're doing good. Terap is not the exportable product, right? I won't watch AI slop. Could SpaceX acquire I do think there's a chance that SpaceX acquires Intel. I just don't think it's likely. Hello, Phillip. Um, DEI killed Intel. I think bureaucracy in general killed Intel. The question is whether the new CEO can rescue Intel from all the problems. Rin, um, yeah, the government is highly invested Intel. Did you cover who pays for the chips? Who buys Terraab products? Tesla and SpaceX and Intel and military. Um, what I expect is that Tesla and SpaceX will use Terraab products. It's not clear this. I did another video about this. If it's a joint venture, if if the companies don't merge and Terrafab remains a joint venture, then my theory is that Tesla and SpaceX will get the chips without paying margin. And if they sell chips to somebody else, that somebody else will get uh the Terrafab company will make margins on what they sell to others. Um DoD will need custom chips for Star Shield. I don't think they're making chips for Starlink that I've heard of. It's not impossible that they'll make chips and hardware for Starlink and Star Shield, but I I haven't heard that yet. So, um I I don't think it's a question of selling Terraab products. It's a question of feeding Terapab chips and other hardware into Tesla and SpaceX for their products. I and I don't think there's a military customer that I'm aware of. Not to say it's impossible, but I don't think I haven't seen that suggested. You know, if Tesla starts making drones for the military or SpaceX starts making drones for the military, maybe. Um, but I I don't know that this hardware is necessary for that. Guess you would probably want AI6 and AI7 chips in that kind of drone hardware. So, that's that's an interesting thought, but that's not part of this video cuz I didn't think of that ahead of time. Um, so they need some mechanism for allocating depreciation, operating costs, R&D, and capital spending. Who gets how much of the capacity? All these things are agreement agreements that get nailed out between Tesla and SpaceX in advance. If they merge, then it's all internal and they figure it out. um during shortages the bigger question may be who gets like let's suppose there's a shortage along the way who has priority that's going to be a challenge but I think the plan is you know three quarters SpaceX one quarter Tesla and that's how they work through it um if there's a merger the merger simplifies tariff dramatically because one company can allocate capital and chip capacity internally so you don't have to negotiate between se two separate groups of shareholders um removes some related party complexity it doesn't make the merger automatically desirable. The merger itself creates valuation, governance, ownership, regulatory, financial reporting complications of its own. I would say if in 2028, if [clears throat] the Democrats win the next election, and I'll I'll remind you again, check out my video. If you're a YouTube channel member, an ex subscriber, or Patreon supporter, you have access to this video now, but it'll be public in about 24 hours. Um, this is one of the guys who's coming after Elon. [clears throat] If they can, they will arrest Elon. They will destroy the companies. They don't care about America's economic success. They don't care about jobs in America. They want to destroy Elon. They want to destroy capitalism. They want to destroy success. So, if [clears throat] the Democrats win in 2028, I'm less worried about 2026. But if the Democrats win in 2028 and you basically elect AOC as president or Gavin Newsome as president and they've decided to go to war with Elon, then a lot of things are at risk. But if the two companies become one, they're an even bigger target. I don't think that matters a whole lot because they're such a big target anyway. and Elon such a big target, but that's a big risk. Um, Terafab strengthens the industrial logic for integration. It's another example of Tesla and SpaceX increasingly sharing technology, suppliers, infrastructure, strategic problems, but it can work. This is the thing, it can work without a merger. It's pretty clear that it would work without a merger. It makes [clears throat] some things either easier along the way, but you have to deal with the merger itself along the way. So just to close, Terafab is not really we're going to talk about anthropic next by the way and we'll talk a couple of things we're going to talk about. Terafab is not really about making chips. It's about making sure that chips do not become the constraint that stops Tesla and SpaceX from growing especially robo taxi and especially orbital space compute or I say orbital compute. It could be orbital around earth or it could be orbital around the sun. Just to keep in mind, the Dyson swarm goes around the sun, not Earth. May generate literally no external revenue. It solves [clears throat] two problems. Get enough hardware and make the hardware cheaper. And the key question is, is Elon and the rest of us wildly overestimating future demand for what Terafab is going to produce? Or is the existing semiconductor industry wildly underbuilding for what comes next? A lot of us think pretty clearly they're underbuilding for what comes next. uh the the data center companies are investing as much as they can but particularly at the level of like ASML TSMC they're just not building even the memory chip companies which are building fabs fast they're not building fast enough so it's not excess capacity it's the growth engine for everything to succeed so I wanted to just see just a quick mention here this is a video that I'm thinking about if you look really close this is Tesla's Q1 quarterly report Tesla's Q2 2 quarterly report and you have the cortex numbers in the bottom and the cortex numbers changed from H100 equivalent to megawws. I think that's really interesting. There's a video coming on that just so you know. So the other thing to talk about that came up recently on on online I saw this I think it was in the all-in pod. Um, it seems likely that Anthropic is going to have their IPO. Anthropic, uh, the maker of Claude, one of the leading AI companies. This is the AI stocks report. If Anthropic IPOs, they will be a stock we will be covering. I cover Anthropic and OpenAI anyway because they're likely to become AI stocks and they're central to the story. They're talking about potentially a $2 trillion valuation at IPO. Now, my take and so I have friends who use Fable, friends who use whatever Claude's latest model is currently. They're using Fable and they swear by it. They're addicted to it. They love it. If you are doing really high value work and Fable is really better than everything else. If it's high value, you don't care about paying $50 a million tokens, right? That the value of the project is so much larger than the cost of the high-end Frontier model that you pay it, you don't worry about it. But there's a lot of other models that are very very effective. So I find Grock build and Grock 4.6 or Grock 4.5 or 4.3 before that. I think I was even using Grock 4.1 fast before that. I've been using various versions of codecs. I personally am not using the Chinese open source models or other open source models. One of my friends here in Thailand, he's French half French half Thai. He's using Kimmy K3. He says it's fantastic. He's getting a lot done with that. Um, at a certain level, the question is, does your project need to spend $50 a million tokens or $30 a million tokens when you can get the same work done or or comparable work done for three $3 per million tokens? So, Grock 4.6 is even Elon would say it's not as intelligent as Fable, but it's fast and fast is a measure of intelligence and it's inexpensive, which is a a value. So for a lot of work and I just what I see is you have this layer cake. You have the application layer at the top with Palunteer Tesla robots Tesla robo taxis. I think there's an open question whether coding is the application layer or the model layer but it seems so integrated to the models that we call that the model layer. So you have the model layer which is Fable Chat GPT and Codeex. Um Grock, you have Gemini. You have Meta has an AI now. Spark I think it's called Muse. Forget what that one's called. I think Spark is a Gemini version. Um and then you have Kimmy K3 and you have DeepSeek V4 or whatever. There's all these other models. So to me the model layer is extremely competitive. You have choices. You go to other layers of the stack and Nvidia is the 800 pound gorilla and you have choice. You can go with AMD, you can go with Broadcom, but they don't necessarily have the supply chain to build what you want at scale. If you're at the fab layer, TSMC is dominant. Yes, Samsung builds a little bit, Intel builds a little bit, but TSMC is dominant. So TSMC has a real huge advantage there. At the lithography layer, that the lithography machine layer, ASML is a monopoly. So there's a lot of points in the supply chain where there are companies that have a monopol and sorry the data center layer um there's a lot of competition but the demand is so great and this is a critical distinction between the data center layer and the model layer. You've got Google, you've got Amazon, you've got Microsoft, you've got Oracle, um, and you've got SpaceX, and then you have, uh, Neo, what do you call them? Neoclouds that are selling access to GPUs, access to compute. And there's some differentiation. You know, some of the providers are providing better quality than others, but fundamentally the demand, there's more demand than the data center companies can meet with the chip design level and with the model with the model level in particular. The model level is the one that matters. You just make another copy of your of your software. The the holdup on giving more access to your software isn't the software itself. It's the data center compute. So, that revenue probably goes to the data center. the that segment of the value in the chain is to my mind more likely to go to the data center than it is to the model layer. So, OpenAI is trying to build compute. Anthropopic is trying to build compute. As far as I know, they haven't succeeded so far. They don't have the the experience that, you know, their ability to build computed at scale. It's not clear whether they'll succeed where SpaceX and Google in particular are just really good at building comput at scale. Amazon's really good at this. So there's a challenge there about who's going to succeed. And I'm I'm skeptical about whether how much value stays in the model layer of the chain. If it comes out if it gets to a point where Anthropic can't charge more than $5 per million tokens instead of $30 per million tokens, then how much profit do they have? How much revenue do they have? You know, the game changes for that. M. Robinson MBA says, "So you think the Democrats and the Republicans are run by different people?" Genius. Um, I think they're both bad. I think the corruption problem is both. Uh, that's why Doge wasn't able to cut spending more. I think if you listen to the rhetoric, if you listen to JD Vance and Marco Rubio, and then you listen to AOC and Camala Harris and Gavin Newsome and Roa, then they're different. Their objectives are different. Their plans are different. If Camala Harris had won the 2024 election, Elon would have been arrested. Tesla and SpaceX would have been shut down. I They did. They're not really hiding it. I mean, Biden was suing SpaceX for not hiring illegal aliens to work making national security uh in you know, ITAR protected technology. It's batshit crazy. You know, Biden was very clearly anti- Elon and Camala Harris was worse and ALC is even worse than that. So, I I don't know. They they they very clearly want they want to tax um what is it called? Unrealized capital gains. Roana is like excited about it. He's I I didn't I should have screenshot that because I for my Roana video, but he's talking with Mark Cuban back and forth. He's basically advertising, oh, oh, yeah, if uh a founder builds a company and it gets to $10 billion in market cap and, you know, his his net worth becomes, you know, $5 billion, even though he doesn't have any cash, we're going to tax him 5%. So that's 200 $250 million, but he can take a loan from the government. What? You're going to make founders borrow money? Like, it's just nuts. Um, should we have military level defense installed around Terra Fab? I think that they will have very strong security measures around terapab. I don't know about military level. I don't know what that means. Military level defense isn't always that good. I would bet we're going to have better better defense around it than anybody else. Dems and Republicans are competing AI chat bots. I don't like Camala. You really think he she they said they wanted to arrest Elon? I I don't know what you what you're looking at. If Kamala won, we invade Iran. Same party. I agree that. No, I don't think that Camala would have invaded Iran. I think Camala would have invaded Israel. I mean, listen to what they say. I I don't understand. I agree that they're in a large ways they are the same, especially when it comes to spending money, especially when to come when it comes to things like NGO corruption. If you look at uh what's the guy's name? Peter Schweizer has a book has a series of books about political corruption. And my favorite is a book called Extortion. Um, some of his books are partisan, Republican, but in all of his books, he pretty much takes shots at both parties. And there's there's extensive corruption on both sides. But when it comes to agenda, what they're trying to do, Trump has been very like the NC NAACP tried to shut down um Colossus in Memphis. They tried to shut down the power generation for Colossus in Memphis. It's in the across the border in the next state. And the Trump administration came in and shut him down. said, "No, you think Camala would have shut down the NAACP when they tried to shut down, you know, uh, SpaceX's AI AI uh, data centers." Very clear difference. Um, I'm not saying they're all different. And again, I'm saying Trump's pardons are corrupt. Clearly corrupt. Some of some of Trump's pardons, uh, Trevor Milton, the the guy from Nicola Motors, I'm I'm not claiming the Democrats are perfect. 51 Industries says, "Wasn't this a video about tariff?" If you read the video description, I did say that I would be talking about anthropic and then this is a live stream and my regular live stream. We have a chat and we have a lively chat and we talk about whatever. We have covered the terapab part of the video. So if you only wanted to see about Terapab, if if a question comes up about Terapab, we'll talk about it. But um but it is common and it's a common complaint by the way that Warren, you're talking to the chat too much. Just stick with the topic. you know, stay in your lane, blah, blah, blah. Th this is this is the way the live streams go. We're 45 minutes in and, you know, once we finish covering the the central stuff, we go off on tangents left, right, and center. And Mark and 144 and Mark and Jim Whitehead when he's here and others, we have a conversation. This is a club. You're and you're welcome to join. You're welcome to to participate. You're welcome to become a member. Um, Mark Plat says, "BLM, Black Lives Matter, is a domestic terrorist organization." Yeah, there's a I mean, I think the perfect example of the difference between the Republicans and Democrats is the Democrats want to import tens of millions of people from Somalia and Afghanistan and Pakistan and give them freedom to steal money from the government and rape our daughters, right? I mean, it's pretty which is what they're doing in England, right? That that's that's their goal. And then Republicans want to say, you know, we only want high-skilled people to immigrate to our country. Some of the Republicans don't want anybody to immigrate to the country to be clear. But there's a perfect example. Um, thank you, B. Thank you, Mark. Um, so there's a perfect example. It's not that they're the that they're the same, but very clear differences in immigration policy after Trump took office. Very clear change. Very clear difference in support for companies like Elon. U Trump's tariff policies. I don't necessarily like Trump's tariff. I'm not saying he's wrong. My gut is I don't like Trump's tariff policies, but very different from what Biden's tariffs policies were. So I I'm not saying that they're they're completely different. Um I think the corruption is very common in both parties, but I think ideologically and some of the policy decisions are very very different. If they wanted to tax unrealized gains, they'd have to figure out how to return unrealized losses. No, they wouldn't. They take your money. They don't. There's just I was just watching a video today that in New York City when you figure out some some condo complexes that these condo buildings realize, wait a minute, um the the city has the number of square feet in our building wrong and we really have 20% fewer square feet. So, we should get a tax reduction and we should get a refund for the past 60 years. And the city won't give them the refund. 70% of the time they refuse to give the refund. Sometimes they give like two years of refund instead of six. government. They don't want to give your money back. They never want to give your money back. They want to take your money and they want to hold it. Um and and I'm not again I'm not saying the Republicans are perfect. Um the there's also the the brighter chat. I I'm never I'm never in those videos. Um not never rarely. The the Israel policy, you know, Trump supports Israel. JD Vance supports Israel. Rubio supports Israel. Camala Harris very clearly did not support Israel. Um if if the next if if ALC gets elected then they will be supporting Hamas. They will be supporting Hezbollah. They will be supporting the the the Islami