Jensen Just Solved Elon’s $119 Billion Problem

Kind: captions Language: en Have you asked yourself how Elon Musk is going to find the $119 billion he needs to fund his terapab chip factory? Well, Jensen Hong of Nvidia just explained it in a new article Jensen wrote. He outlines how he just recruited the biggest money managers on Earth. Apollo, Black Rockck, Blackstone, [music] Goldman, KKR to turn AI compute into something you can finance like an airport. Over $500 billion of outside capital to start. [music] If you own Tesla or SpaceX stock, this changes the math on everything. Elon is building. Certain Basher's translation is the one to read. [music] He says compute is becoming like commercial aviation. Airlines don't own every plane they fly. Leasing companies do. An investor owns the AI factory. The AI company rents the [music] compute and a $40,000 GPU gets valued on what it earns instead of [music] what it costs. In Jensen's words, in AI, compute is revenue. Now, hold that next to Terraab. CERN went through Tesla's filings and job postings, over 100 million square feet, more than 10 gawatts of power. That's New York City's draw. Around $119 billion of investment, and a hiring list full of people who build chips atom by atom. The wall that used to stop plans like this was capital. Wall Street just started tearing it down. And Tesla is building the factory on the other side. Sir Basher is a chartered financial analyst running his own investment advisory firm called Brilliant Advice providing wealth management services. [music] Welcome sir. >> Hi Herbert. >> Well tell us what this is about. I you saw Jensen Hong's post and I think you have extrapolated that to what you now think is what's happening with how Elon is going to fund Terra Fab. Jensen Hang has seemed to have changed his strategy recently where he is investing in every AI company out there and but I think that he's striking deals where he's going to get a revenue share of the product that those companies create. So not just selling you a chip but that too. Can you tell us what you're thinking here? >> Yeah, this is I think a very significant development. Uh Jensen actually created this problem for himself because he's been so successful. So if you think about the early days of AI, uh the companies that built these AI data centers funded it out of their free cash flow. No problem. They had plenty of free cash flow. If you look at Microsoft, if you look at Google and so on, those those companies, no problem funding this. And then they got to the point where they were increasing their capex and their free cash flow diminishes greatly. Uh in some cases even negative. And then they're also raising debt. they're borrowing money to fund these purchases or these these expansions and and that's all great and then you get to the point where Nvidia perhaps is helping them out in some ways in some cases uh taking equity in some of the smaller companies and you have the situation where people worried about this idea of circular financing where Nvidia is taking money from a company but then basically giving that money back through you know buying equity in the company and so a lot of people became very concerned about that so in my mind this is the next major evolution of how to fund the buildout of AI and that is to get the capital markets involved and this is really a necessity. Uh we do this for literally everything else in the world whether you get a mortgage as you mentioned in the intro whether we build an airport um you know buying a car for most people you finance it literally everything that people buy or companies buy can be financed in one way or another. Now, it may not make economic sense to get a subprime auto loan, but that is available for people. In this case, we are looking at probably the one of the biggest capital investments that this country's ever seen. Uh probably greater than the railroads. And so, it makes sense to finance this through large players like these large finance companies, these investment firms, and so on. And basically what we'll end up with at the end of the day is uh large pools of capital like pension plans uh and other large capital pools uh you know pension plans on the state level company pension plans etc insurance company portfolios uh investing in basically what will become a new asset class and that is AI compute. So I think this development actually is extremely important. We talk a lot about the technology and how the technology evolves, but this is an evolution in the financial part of it, which is also critical. If you can't get the money, you can't build these data centers. I know people like to sort of dismiss Wall Street, but this this is actually they're playing a critical role here now. And I think this is a a big deal. >> Yeah. So, uh the key thing is Tesla and SpaceX. Uh are they going to be one of the winners? It's a massive race not only with companies against companies, mag seven hyperscalers against each other, but countries. This is a race and who can raise the most money and then fund it themselves. So, at one point people were saying, you know, Google, Amazon, they've got a um, you know, an advantage because they they're just breaking in cash and that's why they're able to fund $20 billion of $200 billion of capex this year. you and I you've done a show with me where you pointed out that in the future there's likely going to be 10 titans you called them uh because it's really the biggest companies are going to win the biggest share of a lot of things and so it's a race so the question Tesla investors have SpaceX investors is how is Elon going to raise $119 billion are we going to be even more diluted capex spending where's that money going to come from some people are saying SpaceX is going to use Tesla's robo taxi cash flow to fund its ambitions and the ambitions are sky-high literally. It's never ending. It It will never end. There's always going to be a new thing, a new thing, a new thing. Uh once it's after Mars, we're going to go to Nept, you know, just it's never going to stop. So, uh but Jensen and interestingly, Jensen and Elon have partnered together and you can see this. Larry wrote a great article about this and it was and I think you're just what you both are coming at the same idea in a separate time. Uh so let's let's start it off here with Jensen Wong and his article was Nvidia AI factory compute is becoming an investable asset class. I'm not going to show his article anymore but uh can you just before we get to your points can you just summarize what he's saying? Yeah, and I think in my article I really tried to summarize and expand a little bit about what he was saying, but the idea is that this is an infrastructure investment and any major infrastructure in the world, whether it is an airport, a road, um just a large project of any kind, a tunnel, you know, you think about the tunnel in Boston and so on, those are often many times um uh public uh but also public private. In this case, what we're doing now is private companies are building out these AI facettories. Now, we do have some nation states doing it as well. So, this is also becoming, I think, a public private partnership in many ways. And you've got to find the financing for that. Um, and and this is what Wall Street does all day every day is is create these large pools of money, find the investors to fund them. They package up the securities in many different ways. you could create uh you know a super safe investment that was backed by these AI factories that paid you know a modest yield over what you could get from treasuries for example and that would be very interesting for pension plans and insurance companies. You could package up another kind of security that had a bit more risk. You could package up another security that had a lot more risk that had almost like equity-like upside to it and everything in between. Think about uh the mortgage market that we have all the different classifications of mortgages. It really could be the same kind of thing for financing in this AI facetry race. >> Gotcha. I think I'm understanding because what he's saying is that AI data centers is almost a given highly stable, highly proof upside is intelligence that if you build it, they will come, you will sell it. And so people who invest in it, you actually have a good stable income coming back to pay you back much more. It's like a house, right? I think people are realizing this is um a utility not just a night you know a startup trying to do something. Um so >> I think that the important distinction here just really quickly is is that if you were financing you know a particular AI model that's different than financing an AI data center an AI fat tree where you could use any model or number of models or combination of that's a very different kind of investment. And it's much less risky that as as as Jensen has said, AI compute is revenue. So you want to build these big AI data centers, create the compute, what models you put through them, that remains to be seen. It doesn't really matter from that perspective. >> It's almost a new currency. Instead of the US dollar, it's how many tokens can can you give me tokens? Looking for tokens. Trade you for tokens. That's right. >> AI factories are the industrial infrastructure of the AI era. Inference is the workload. Tokens are the new commodity. There you go. Compute is revenue. So, Nvidia just unlocked trillions of dollars of capital. There you go. All right. Walk us through. >> Yeah. Now, I know in in Jensen's piece he said something about 500 billion, but if you can do 500 billion, then you're you're on your way to trillions. So, I've I've taken liberty to expand the scope of this a little bit. Um, but if you think about, you know, if you wanted my example here, if you wanted to invest, you know, five billion worth of computing infrastructure, you need to find the money from somewhere. And this has been the issue with a lot of companies this summer is their capex budgets are going up and up and up and Wall Street is saying where are you getting this money from? You're using all your cash flow. Your free cash flow negative. You're going into debt. Some cases are selling equity, right? Or all of the above. And then the question is, is there another financing mechanism? And the answer is yes. This is what Jensen's working on here. Mhm. >> So the idea is that uh in the middle paragraph there, the new model separates the user of compute from the owner of the underlying infrastructure. Right? Again, think about when you drive on a road, you don't own part of that road. You don't, you know, to drive on that road, it's not something you own. In this case, in many in most cases, it's a public uh it's publicly owned. In some cases, we do have private roads where you pay a toll, >> right? But you've separate the owner separate the owner of the road from the users of the road. And we're doing the same thing here with compute. >> So yeah. So instead instead of um SpaceX or Tesla buying their own and spending their own cash to buy it, you can go and get capital from an institutional investor, they own it. Yeah. >> SpaceX maybe has, you know, skin in the game like I'll just make up a number 25% of their own money and they raise 75% from other investors. And that gives those investors too a lot of comfort that SpaceX has their own skin in the game that SpaceX is just not taking their money and and not making a good investment. If SpaceX also has significant capital at risk, you know, that's fine. But in that case, SpaceX could invest $50 billion and raise 150 billion from other other investors and that that would make a lot of sense. sern uh now that SpaceX has announced that our return on investment for these invest uh data centers is a year or even less than a year if they can prove it out and then others can't when they then get AAA rating and people or equity like all the everybody wants to just throw money at SpaceX and Tesla. >> Yeah. And this is why I think that this is um initially a big funding vehicle for SpaceX. >> Yeah. Obviously, Jensen and and and Elon have had a lot of conversations about Elon's need for chips. Elon's announcement that they're going with Nvidia as their sole supplier, right? So, there's a high level of discussion that's happened here at many many different points or different levels. Elon is the from what we can see right now the fastest builder of AI data centers in the world. They have told us that their payback period is less than a year as it stands right now on these factories. So, if I'm Blackstone or Black Rockck or Apollo or these other firms, I would want to put my money with the firm that is the best that's doing this the fastest, most efficient way. And right now, that appears to be SpaceX. >> So, Elon can raise money whenever he wants. But, you know what my theory is? My theory is that Nvidia, they haven't completed their announcement. So, Elon comes out and says, "We've committed or we have uh partnered with Nvidia. they're going to give us as many of their chips as we kind of want or we got first dibs at least. Nvidia says the same thing. But what they haven't said is Nvidia probably is going to share in the revenue that is generated from the leasing or even Terraab like Laric suspects that the partnership terab will be funded by Nvidia like there will be much more of an intertwined relationship. If Nvidia is smart, and I believe they are, >> it would be in their interest to be as involved in terabb as they can possibly be. Absolutely. So therefore, I expect that in some in some fashion. This also may be something, you know, where Elon says to Jensen, I'll take as many chips as you can give me. >> And that's great, but it also creates a bit of a risk for Jensen if Elon becomes his only customer. Of course, that's not the case. I'm just exaggerating. So Jensen also wants to try to find a way for financing for all the other players in the space. He loves Elon. He wants to give Elon as many chips as possible, but then you've got a business risk if you've got one customer that's taking call it, you know, 25% of your of your chips. So it's in Jensen's best interest to find financing for everybody else. To the extent that Microsoft and Meta and maybe some of these other smaller firms, some of the Neoclouds need financing to get the chips, Jensen wants to try to make that available. So it's a very smart move on his part. >> Okay. >> Yeah. >> So, so this is the this idea that compute is it becomes a productive asset. Uh compute is revenue. That's simple fact. Now the nature of that revenue stream may change over time. How that's generated, the chips that are involved and and all that may change over time, but the fact remains is that the more compute you have um allows you to do all kinds of things that generate revenue. It's that simple. So with this one, long-term compute contracts could resemble energy offtake agreements. So if you think about, you know, in the energy market, um, a customer might agree in advance to purchase electricity, natural gas or some other, you know, oil or something like that for a specified period. So if you get somebody that's kind of willing to underwrite that, um, then that makes it easier to finance it, right? when you've got a kind of a a I forget what the word is, but a primary customer uh that's going to take most of the compute in this case. Um again, to the to the degree that Wall Street has a lot of comfort in terms of where this revenue is coming from, they will be more than happy to finance it. >> Are you referring to like an anchor customer, anchor tenants? >> Yeah, that's a good way to put it. Yep. >> Long-term compute contrast could resemble energy off agreements. Yeah, it's already example cuz compute is basically like you you're now saying it's it is energy, it is oxygen, it is water, it is uh electricity. Yeah. >> And I think what we've seen now with SpaceX is these agreements are fairly short-term in nature. They have a 90-day out clause for both parties, >> right? But if you're Wall Street financing a project that's going to cost, let's say, hundred billion dollars, you want to make sure that the customers are are lined up and locked up for a long period of time, at least through the period that you need to get your money back. >> Mhm. >> So, I mentioned this earlier that that once you've got these revenue streams and it's predictable in some way, then you can finance this in any creative way that you can think of, right? So, you don't have to finance all of this with equity. You can finance it with senior debt. So this is debt that gets paid off first if there's some kind of bankruptcy situation. So that might be a portion of the factory. You might have some other layers of debt that are subordinate to the senior level that have higher returns but higher risk and that appeals to you know private credit investors or other kind of high yield type investors. And then you might have a layer that's pure equity and that might come from some kind of infrastructure fund. And all of those firms that were named by Jensen have these different kinds of vehicles. um infrastructure funds and private equity and private debt funds and all kinds of stuff. This is what these companies do. >> So CERN, uh we've heard Tesla CFO come out and say that this year we're going to not only increase our cap spend to 25 billion, but we're also going to go to the debt market and raise 30 billion because we're about to launch Robo Taxi and Optimus, but right now that's Tesla having to raise that debt themselves, right? And it's not Robo Taxi. I'm just assuming Robbo taxi is not anywhere like AI compute yet. It's not going to be seen by these massive uh money managers as AI as as you one of the words you use is predictable. Predictable >> but once it becomes predictable then it becomes easier and easier to get the money to flow in and build out and Tesla does not have to use their own cash. Is that right? >> That's right. There is a point at which something that goes from development that's introduced to the market that scales it gets to a certain level where Wall Street will now look at that and say okay this has been going on for a period of time we're comfortable with the cash flow situation here we we know this is predictable essentially allow us to build some financial products around this and I think and I' I've written about this a few months ago that robo taxi will be a new asset class as well >> just like AI compute is a new asset class >> that's what I was trying to get to you've already written about it it's a new asset class which means like yeah okay I I can't think of it but it's basically transportation and so then you can invest in transportation and then you'll get a return you put in the money that helps fund it fund the growth and all that >> and this is this has been a question for Tesla for I want to say a couple three quarters ago >> where one of the Wall Street analysts have asked about how they would fund the massive expansion of robo taxi and I believe they indicated that they would look potentially to Wall Street to help fund it. Um, it could be through debt or it could be through some other creative structures. One of these companies like Blackstone, for example, could put together a robo taxi infrastructure fund. >> Sir, okay, [laughter] that just made me think of something pretty big. So, for SpaceX to acquire Tesla, they need to get the investors to say yes at the annual shareholder meeting. The investors, unfortunately, to retail shareholders, retail shareholders only own 22% of the Tesla stock at this point. The vast majority are Black Rockck. They've been loading up this entire week. I did a bunch of shows showing how much money they've been loading up. >> Yeah. >> In exchange for saying yes. Could they have negotiated with Elon and say, "Look, I'll say yes. I own like I don't know what the number is, you know, billions of Tesla stock. I'll say yes, but you need to give me first dibs just like they would when you go when you're going to go IPO." they raise their hands and goes, "I want to be one of your, you know, IPO guys." And uh these guys will say that, you know, like I get to fund the robo taxi roll out, you think? >> Well, absolutely. And I think that if they could show Elon that they supported him. >> Yeah. >> Right. And then when this opportunity comes around, which is not too far away, >> Yeah. >> he's might look more favorably on the opportunity to work with them. Right. >> Uh this is the nature of how the financial world works at this level. Um, and we've seen that with the SpaceX IPO. >> Yes. >> The analysts have written some pretty nice glowing reports about SpaceX. Why did they do that? >> Yeah. >> Is there a Chinese wall between the investment banking side and the research side? Well, there's supposed to be, >> but still, you're not going to write a negative report about SpaceX. And none of them did. So, this is the same game I think that's being played at this level. whether they, you know, whether it's been agreed to or even discussed or not, everybody understands what the game is and you want to put your firm in the right position in the right light so that when these opportunities come along and you know that you you can capitalize on it and so in this case what you're saying is that these companies are acquiring Tesla stock so that they can vote and and and approve this merger because it's going to financially benefit them. Absolutely. AI data center doing the purch AI data center's cash flow. Robotex is going to be cash flow. Robots's going to be cash flow. We better like get in good terms. The other thing on the other side is retail. That's you know Elon and Tesla and SpaceX are going to use other people's money. And so this idea like you know you and I have debated about why I I kept saying why would Tesla sell cyber cabs to customers? Why would you sell Cyber Cab? You can make all that money. Of course, one of the reason is you don't want to put they don't want to put up the 30,000 bucks to finance sub despite the fact they'll be paid off in six months, I don't know, in a year. Uh but the one of the ideas is let a retail a fleet manager buy thousand cyber cabs and then that's just another way to use their money to fund the growth. >> Yeah. I think my strongest argument for why Tesla would want to involve others, whether it's individuals, small business, or large businesses, get involved in this is the political side of things. >> Mhm. >> If Tesla destroys the transportation world and replaces it with autonomous transportation that you don't own, that you just use on an hourly basis, you want to have, I think, many as many players as possible in your camp. And so, if you've got the big financial players in your camp, >> they're pretty influential in political circles. If you've got mom and pop involved in it, that also helps from a voting perspective. Yeah. So, that just seems smart to me from that perspective, not so much a pure financial one. From a pure financial one, it makes sense for Tesla to do everything themselves, but I don't think the world is quite that simple. >> And Elon himself has mentioned that in the past that he would uh, you know, sell cyber cabs. And I I I believe that where he's coming from is an understanding of, you know, how the political world works. and you want as many people on your side as possible. >> Share the wealth. Okay. So, institutional capital can dramatically expand the AI buildout. So, it's gonna even more money can be flowing in. >> And this is the opportunity that that Jensen's been talking about now for a few years is these AI factories are just going to be, you know, this massive super important to the economy. It's going to need a lot of capital. And so you better get access to where the money is. And that's these global pools of capital that are locked up in pension plans. It's trillions of dollars, insurance companies, sovereign wealth funds, private credit funds, infrastructure investors, right? And then just other asset managers that manage mutual funds or other pools of capital that the money is out there and they're looking for opportunities. And so why not create it? And this is I think exactly what Jensen's trying to do here. And that's why I think 500 billion is just the starting number. It's going to be in the trillions. >> And this uh translates specifically to SpaceX and Tesla because the same thing. They're the ones building the AI data center. SpaceX is anyways. Tesla's building robo taxi and robots. So uh all that money will flow there as well. >> Yeah. Now the one of the criticisms is is this. I've said the first post I made on this was that basically this is like aircraft leasing. Uh in the aircraft world, most airlines don't own uh all of their airplanes. They may only own a portion of them. There are leasing companies that are set up to buy the airplanes from Boeing and then they lease them to Delta, United, and other airlines. >> Mhm. [clears throat] >> And the airlines operate their airplanes, but they don't own them. And so you could do the same thing. You could have SpaceX, you could have Microsoft or other companies leasing the data centers from these pools of of capital that might own them. Again, I think they'll have some skin in the game. I think Microsoft and and SpaceX would own, you know, a percentage and there'd be other partners involved. So, one of the criticisms is, well, airplanes retain their value, but data centers don't. So, Cernin, this is a terrible analogy. You've got it completely wrong. >> What? But actually, I do think the analogy applies because the airplane itself, the body of the airplane could last 25, 30 years. And that's like the data center building that's going to last 25, 30 plus years. What what doesn't last on an airplane is the engines. They require a lot of maintenance, a lot of parts, and that's kind of like the GPUs. The GPUs may only last for a few years. So, actually, this analogy, I think, to aircraft leasing actually is a very good one. you've got you've got a portion of it that retains its value and holds its value for a long period of time and another portion that's highly consumable and doesn't last very long. But what we're seeing on the AI side right now is some of the older Nvidia chips, they still have a lot of life left in them and they're making money. So, it may actually be even better than aircraft leasing in that sense. >> I was going to say I just saw Jensen Hong say that A100s are still being used. Anybody who's jumps up to the others, you'll need uh uh inference. So, not training. You're not using it for training anymore because it's old, but you'll use it for inference. Somebody ask a question, you get a, you know, answer back. They'll use the 80100s. >> Yeah. >> Now, the other big thing that this leads to is is if compute becomes kind of like a commodity. In that case, then what you get is what you would call a spot market. So if you need compute today, what's the current price, right? I need I need compute for the next 24 hours. What am I going to pay? But then you can also create what you would call like a futures market. I need compute 6 months from now and I'm going to buy a contract to lock in that price six months from now. So you get an even another another layer of financialization of this. >> Okay. >> And and we have this today in oil. We have it today in lots of commodities, wheat and other commodities, metals. Um we have it in financial products uh the futures and options market right when you buy an option on a stock you're betting on the increase or decrease in the price depending on what option you buy and how you structure it. You could also do the same thing with compute. And so you can imagine given the size of AI in the future just how big this options and futures market could be for AI compute. So that's also very interesting to think about and we're seeing maybe the beginning of that. >> Wow. Do you know uh right now I we we were saying that there's 10 titans all the big boys trying to create their frontier LLMs and they're fighting each other but the reality of where this could go and very likely is that AI becomes so distributed every single company has to have AI they're doing it now in their own you know uh walled AI data centers with their own uh trained LLMs running their business because they wouldn't want to you know outsource that because guess what's going to I'm a law firm. I use anthropic. Another says, "Hey, I'm a law firm now. See you later." >> This is one of the big risks. Yes. >> So, in fact, they that massive law firm is now going to go, "Hey, I need AI compute." And so, there's going to be many many players. It's going to be like every single company in the on earth will need spot market. Some will need it now more than others. Some will need it for sure, like you said, six months from now. Yeah, the market is huge for this. Now there will be some differences. If you think about oil, uh not all oil is the same. >> Right. >> Right. There's light sweet crude, there's heavier crude, there's oil priced in different locations around the world. Different again different types of oil. And AI probably too is probably quite similar to that. Not all AI is the same. Not all AI computers the same. a cluster of you know GB300s is very different than a cluster of A100s obviously right the capability of those systems is very different so we'll probably see a a futures market that reflects that as well just like we do in other commodities today um and this one actually could be even more even more complex and that would make sense >> wow yeah >> but the question is I have in my mind is there still a premium for a company perhaps like SpaceX that says Okay, not only do we stand up data centers faster than anybody else, but we've optimized this data center in certain ways to make the compute for you customer the most advantageous and cost effective possible. It's not just, oh, we've got a gigawatt of compute that you can buy from us. It's we've got this gigawatt of compute that gives you the the most efficiency, the greatest optimization you could possibly imagine. And here are all the ways that we've done that. So that would kind of go against a little bit about the financialization of some of this if if there is still those highly unique differences. So it's going to be interesting I think to see how this all plays out. Um you know the cost of compute from SpaceX may command a premium for example in that scenario if they can offer something to their customers that's quite unique because of what they do not only being fast but just being highly efficient and highly customized in some way. Um, also a data center with, you know, a whole array of mega packs is a lot more reliable than a data center without that. And I think we've seen some stories about data centers that didn't use Tesla mega packs that maybe it didn't work out so well for them cuz the battery backup systems that they used to sworn as good. So there's definitely going to be some differences in a gigawatt of compute depending on where it is. >> You can already see it right now. Tes SpaceX and Tesla's supercomputers >> have mega packs. They will have solar because both companies are building their own solar factories. They will have communication through SpaceX Starlink. >> Yeah. >> So when you are a customer, you get a lot more from that than you do the others. >> Uh right. >> Wow. And you can see where this is going to go. I mean Elon can bundle so many things just like Nvidia bundle CUDA and that's what makes their chips so different from anyone else. U SpaceX will AXAI will bundle uh Grockbot they'll bundle Grock they'll bundle digital optimists if you're a factor you might get along with it. Yeah. >> Yeah. Just like Tesla uh provides software updates to its cars to make them better over time, Nvidia is effectively doing that with their chips using CUDA. So, it's a very similar thing. You can think of of sort of FSD for Tesla and CUDA for Nvidia is kind of very similar in that regard. So, as I mentioned in my article, the residual value of these GPUs also becomes very important because as an investor, if you get more life out of that investment, then your returns are going to be that much greater. And so you're going to want to invest in Nvidia chips. If they if those get more life, then that becomes kind of the the prime market and maybe AMD's chips which maybe get less life become kind of subprime. You you demand a higher return to invest in that. So you get that kind of risk based pricing as well. Um so the residual value of these things becomes pretty important over time. >> Elon is just so brilliant. I just can't imagine this. I mean, he's I mean, the whole thing about we need to build our own lithium refinery to build our own batteries to have our own uh mega packs. [clears throat] We need to build our own solar farm, solar panels, terraact. We're going to build our own chips. We're going to have our own memory. We're going to have our own packaging, TSMC. So it's like [laughter] it's it's it's it's like then we can build the data centers of course and then we're going to go to orbit because there's not no space here. [laughter] Just oh my god now you now you see the money because it's like uh you know the question is always are there customers to these data centers and is it unlimited? One of the big question mark for SpaceX is okay, you say you have hundred billion dollars of ARR this December. Fine, but it's an N and day out. Maybe there's going to be an AI boom, a bust, and if it is, nobody's going to rent them next year. So, you can't just take that number and extrapolate and keep adding more to it over the years. Many of us think that that is what's going to happen. AI intelligence is unlimited. We're haven't seen anything yet. Others saying it's a bubble. we've we've hit the max, you know, they'll they'll they'll be able to slow down next year and so this is the bet. >> Okay. >> So that that remains to be seen. Um and on that contract out it wasn't so much that you know Google or Anthropic asked for that. Elon said he asked for that >> because he can get spot pricing I think. >> Yes, >> he he can get spot pricing which is like somebody wants to pay him more uh three months or he will use it. Yeah. >> Or he will use it. I think that for me is the key thing. I think he's looking at the cursor acquisition and saying, "Okay, I think it's going to close." Close today, by the way. Um, if it does close, I could see an explosion in demand for what we're doing with them. And so, I may need this compute myself, and so I want to give myself window. Yeah, this is the biggest un >> accounted for revenue for SpaceX >> is nobody has yet said oh by the way cursor themselves can become a competitor to Anthropic who is making hundred billion a year. Curser only made $8 billion only made eight billion but you know compared to uh OpenAI and Anthropic they're nowhere near what those guys can do. But once you have the backing of all this data center compute, you've got the Grock that's so incredible and you saw Grock bot which is created by Cursor and if Cursor's LM competes with Anthropic, they can become a hundred million per year business just that alone and nobody has calculated that. I'm not going to do it yet until I see because it's too much of a leap. But boy, it's it's there. Could be there. >> It's it's a nice call option that I don't think's being priced in very well. Nice. >> The similarity I can think of is it's like you buy a car from Tesla and you get to use it and that's cool. But right now, you know, they're using that they're they're leasing that compute to others and that's what you could do. You could you could say, "Okay, with my car, I'm going to put in the robo taxi. I'm going to make some money off of it." That's kind of the the compute rental business. But in this case, the upside is not that. The upside is using it themselves. That's like you finding a way to make your car that much more useful and valuable to you by, you know, starting a delivery service or whatever. I mean, you suddenly can generate revenue from this car, not just using it to commute back and forth. This is the opportunity that SpaceX has with Cursor and this the scale of it is absolutely enormous is from what we've seen from the likes of Open AI and Anthropic and their ability to scale their revenue. Yeah. >> More money than the leasing, data leasing. That's why you're doing this much more. All right. So then what's this? utilization becomes the equivalent of hotel occupancy. >> Yeah. I mean, this is if you're investing in a data center, the utilization obviously is important. If you're only leasing out, you know, 20%, that's a certain amount of revenue, but if you're leasing out 80%, obviously that's going to be much more. This is a I think a pretty obvious statement. Um, and so these I think my point with this is that Wall Street investors are very used to looking at occupancy in commercial real estate or hotel investments, whatever they're investing in. this is a very sort of common thing for them to factor in. So this is not going to be a difficult stretch for them to to analyze this. Certain why don't you and I start a company um we'll call it um Datapedia I don't know Xedia you know so all the hotels are out there now there's all these data centers and we become the one-stop shop where you come in you see all the data centers you see occupancy right here full this one has vacancy you can see the rooms that they could charge >> marketplace or >> yeah it's more like Airbnb for data centers perhaps is what you're talking about um Yeah. >> Yeah. >> Could be it could be real. >> Yeah. No, I think we'll see a lot of that. I think the question though remains is like real estate. There's obviously a thousand thousands hundreds of thousands of players in real estate and the question is how many players will there be in the data center game. To me what what we've seen from Elon building standing up a data center is actually pretty complex. >> True. >> So it tells me there's probably going to be fewer than than than more than more players like this. There's only a few companies that can do this. Uh Elon seems to be head and shoulders above literally everybody else. >> Yeah. I'm not even sure if anybody has actually made 100,000 GPUs coherent yet. >> Yeah. And there's there's reports too like crazy things like the number of workers that that are working at these data center sites to build them. >> In Elon's case, it's like 3,000 at the Colossus, you know, work site. And in a similar size data center with other companies, it's more like 10,000 workers show up there every day to build the thing. So he's building it faster with one-third less workers, one-third of the workers. Like it's just on every level, it's just astounding. >> Okay, I think you've already covered this one. Residual value of GPUs becomes financially important. Uh used compute market could develop. Interesting. Yeah, this is maybe secondary, but at some point, you know, you can see these A100s getting taken out of the data center and, you know, can you get any more life out of them? Well, somebody might pick them up for pennies on the dollar and build a business around that. >> Uh, if there is a if if if prices of comput skyrockets and the demand skyrockets, then there's supply issue, then the use market will be very valuable. >> Yeah, you talked about this already. CUD is already doing an economic life for the asset. Yeah. Uh media may establish a residual value floor. What's this? >> Yeah. I mean if you are you know initially for example lenders or participants here may be worried about well I don't know what the future of AI is going to bring. Nvidia will you backs stop us? Will you ensure that the bottom 25% is covered? in case things completely fall out of bed here for some reason, right? Like they're they're they're backstopping this to some degree and Nvidia may or may not be willing to do this. We'll have to see if if if this is even required, but this is another way just to provide some financial protection to those investors that are getting involved in these in these kinds of deals. Um I don't think this will be needed. Um you know the comment here I says NVIDIA does not need to finance the entire project. it only needs to reduce enough uncertainty to help institutional investors become comfortable financing the remainder. So that's the whole idea with the back stop. Yeah. >> Okay. So let's uh switch gears here and talk terra fab. So what you've discussed so far is uh you know all this AI compute and then of course beyond AI data centers this also applies to chips even more so. Uh Terafab is a $119 billion investment. Where's Tesla going to get that from? Very concerned. SpaceX uh investing in this. It's going to take years, decades maybe some people think. So you did a incredible article and you titled it, how many bathrooms does Terra Fab have? Great job, CERN. Who cares how many bathroom does Terapab have? You funny dude. Well, that was the point and that really wasn't the focus of the article, although I did discuss it in there, >> but it was kind of kind of all things Terrab if you think about, okay, let's let's think a bit more deeply about the structure. And a lot of people had a lot of fun, I think, trying to show how big this is. >> Um, you know, compared to Giga Texas and the Pentagon and Apple Park and Mall of America and all these other structures. And I included some of those images in my post just for fun. Yeah. uh if you stuck at you know end to end in the Grand Canyon, how tall would it be and that kind of thing. Um and there was a second post here, Herbert, about the job openings. I think we'll talk about that in a second or are we are we talking about that now? >> We can talk about this now because I think I've got my slides mixed up. >> Okay, so we'll come back to the terapab itself in a second. The job openings one I think is actually really interesting and didn't get nearly as much attention as the other article about the bathroom. So it's interesting people more interested about bathrooms than they are about jobs. [laughter] Um I think it also probably was related to the fact that that Elon uh you know liked one and not the other. >> Yeah. Um, okay. So, in terms of, uh, the job openings, this is actually really interesting because they are already hiring now for, I think, 34 different roles. >> Mhm. >> And, and we've got, you know, some screen grabs here in some of these some of these jobs. But if you look at it, the way to summarize it is it runs the gamut across the entire semiconductor manufacturing process from building a factory and and coordinating what goes on inside the factory to all the way down to like the subatomic level of how do you construct chips >> at the two nanometer level and all these different like frontier levels. We can go into detail if you want on some of these roles, but and they're very interesting, but they they cover the complete I I would say from soup to nuts from start to finish of the semiconductor manufacturing process. And the obvious thing may be to say, well, yeah, that's what Terapab is. Terap is combining building semiconductors, building memory, packaging it all together. This is the purpose of terap and and yes, this is what the purpose of terapab is. And I'm I'm here to say that Tesla now is looking to hire people that cover that full gamut. Now the location that shows for all these is Austin, Texas. I think there's one in Palo Alto of the 35 postings and 34 in Austin. So clearly initially they're going to be at the um the advanced >> advanc which is the small version of the Terapab that is in Giga, Texas. All of these are being hired there first and then they will build out the tariff app at Grimes County >> which is 400 miles away from Kika, Texas. >> Yeah. I'm not sure it's 400 miles. 200 maybe 200 miles. >> Oh yeah. Yeah. Okay. Gotcha. >> Maybe less. I think it's like two and a half hours away. Herbert. It's a little bit closer but yeah. >> Okay. Are you sure? >> I thought I read four hours of a drive which is >> okay. >> Yeah. It's sort of between Austin and Houston. Not directly between, but um yeah, it's it's about two to two and a half hours away from Austin, I believe. >> Dude, why don't you go and do the ride because you live in Austin. Go ride there and tell me. [laughter] >> Well, it's funny. I woke up yesterday morning thinking that I wonder when Joe Techmire is going to take his drone out there and get some video. And that's exactly what he was doing when when I was thinking about it. >> You should go with him. Should have gone. >> I should have I wish I I wish I would have had the chance, but he he went out there and got some great shots. He did a great job. So, I mean, I don't know if it's worth us reading it because I don't know if we'll even understand a word of any of this. This is highly advanced. Uh, >> well, let me let me just summarize some of this for you. I've got some notes in front of me. So, you've got uh people that are focused on um is is it an industrial hygienist for example, [laughter] right? So, you've got to make sure that the factory doesn't slowly poison people, injure or whatever. So, that's that's that role. You've got an environmental health and safety person, right? You need to make sure that terap is safe for employees and everybody that's there. You've got an environmental engineer that's focused on air quality. Uh because terap will release into the air uh certain chemicals, all right, within the legal limits. You've got to make sure that you're consistent with that. Um you've got other roles on the safety side. You've got uh construction safety specialists to basically walk the site and make sure everybody's following all the safety rules. I think there's two roles with that. Uh these are senior roles. You've got an environmental engineer that's uh looking at uh waste water, storm water, hazardous waste, all that stuff. Okay, that's that's kind of broadly at the factory. Then you've got the people that are in charge of operating the machines, right? So you've got equipment engineers. Um so one of the the things when you conduct uh when you build semiconductors is that you're repeatedly adding and removing microscopic layers of material. So the wafer has to be polished to an extremely flat level in order to put another layer on. Think about if you're building a multi-story building and you build the first level and the first level isn't quite straight. Well, with a semiconductor you've got to make sure it's perfectly flat. And that's what they're doing level by level. That's a chemical mechanical planerization role. They're hiring for that. Um, you've got somebody that is uh they're going to build a lot of custom machines and they're hiring for a role that basically solves problems as it relates to the custom machines. A senior mechanical engineer. You've got a senior technical program engineer who's basically responsible for keeping this giant puzzle rolling at all times. right? Like if you think Lars has a a challenging job, this some of the people in these roles are going to be pretty senior, pretty pretty important people. Um you've got to predict um how different materials are going to behave when they're mixed, when they're heated, when they're cooled. And so there's a materials engineer that's going to use computer models to to predict that. Okay. Hopefully Grock uh is helpful in that regard as well. You've got um a person to to figure out how to um you know they use extremely thin layers of material to make these chips. You've got a person whose role it is to figure out how to deposit them what amount and then to solve the failures in in you know involving all these different materials. You've got somebody uh who's looking at this atomic scale using computer simulation simulation. It's density, functional theory, and molecular dynamics. Okay. To simulate atoms and predict how materials may behave. That's the level at which they're hiring for right now. Okay. You've got somebody in the senior materials engineer, uh, transmission electron microscopy. Okay. So, this person operates incredibly powerful microscopes capable of examining semiconductor structures at nearly some resolution. >> That's crazy. >> Okay. they're hiring for these people right now. I know I just said that a minute ago, but just to reinforce. Um, and so on. This is this is just, you know, half of the roles I think that I've discussed. It just goes on. It's really impressive in terms of who they're looking for. And some of these people, they're looking to hire people that have been in the industry for a long time. >> Yeah. This this is it. This is uh the universe. This is physics. This is >> Yeah. >> So, the