# The TRUTH about Coreweave: Your ULTIMATE Guide to AI’s Least Understood Business

**Channel:** The Wall Street Skinny
**Source:** https://www.youtube.com/watch?v=nbjo1cwtZFE
**Transcript page:** https://www.withtranscript.ai/video/nbjo1cwtZFE

## Chapters

- 0:00 — Market Highlights and Media Bias Analysis
- 6:28 — Cloud Fundamentals and Crypto Origins
- 15:16 — Neo Cloud Financing Models Explained
- 25:12 — Nvidia’s Role, Risks, and Closing Outlook

## Transcript

**[0:00] Speaker A:** You're talking to two women who started our careers when shit hit the fan. You started at Lehman Brothers, Rip and I started in CDO structuring in 2007 and watched my group go from 40 people down to three people in the span of about a couple months. My group just started to implode. Leverage can cause problems in the short term, even if the long.

**[0:19] Speaker B:** And in the long term, leverage is.

**[0:21] Speaker A:** Leverage is always scary.

**[0:27] Speaker B:** We've got a lot to cover today. We're going to do an entire segment on the NIO clouds. But first there were a couple of big stories that both roiled the markets and surprised them. The first that I want to talk about is this news about OpenAI. Did you see this headline, Kristen? I'm actually going to pull up our favorite news source, Ground News. Tech stocks drop after report that OpenAI's revenue is lower than expected. And it's really interesting because on ground news you can see this bias comparison function here. That left leaning coverage framed OpenAI's roughly $15 billion ANN annualized revenue as an AI bubble warning because it was lower than expected using terms like plummet panic and massive financial failure. Okay. Whereas right leading coverage is apparently highlighting the disappointing number and scares but emphasizing the accounting differences, continued growth and possible Future targets near $70 billion.

**[1:18] Speaker B:** So there is a little bit more of a positive spin on things from one side of the news cycle and a little bit more of a doom and gloom very negative spin that we can see. And this is one of our favorite things about ground news. Go to gr groundnews.com skinny and you'll save 40% off your subscription. This is the best way to stay on top of the markets and get this alignment reading on all of the different coverage that you see on the biggest market moving headlines. Again, that's ground news.com skinny and you'll save 40%.

**[1:44] Speaker A:** And I want to just like emphasize why that's so important, especially if you're investing. If you're doom and gloom, like I happen to have a negative bias, like, but like if you're doom and gloom, you really should also consider what could go right. And by the way, if you're so bullish, you should probably also consider what can go wrong. And I do think if you're in an echo chamber where you know, if you're reading more left leaning sources like you're going to probably be more worried that oh my God, the world, you know, that this is a bubble, the sky is going to fall. Whereas if you're reading more right, you Might be more inclined to invest. And I do think one thing that I've said this before, I think one of the most like potentially costly mistakes that I have made was actually like kind of being too bearish.

**[2:22] Speaker A:** Like there is. There is a case where you can be too much to one direction. And so I think just like going in with as much of an open mind and like researching, but like also making sure you know both sides of it is really important.

**[2:35] Speaker A:** And I was brought up in with this kind of mindset that like you should be conservative and that conservative is basically like how you win. Conservative is also how you can like lose too. So I just want to be. And by the way, this is not like right or left conservative. Just like with me. And I was like, invest your money and like put it in the bank account. Like never touch it. Like that is so damaging. So anyway, I just like, I want to reinforce that. I think it's. We are so easily swayed as individuals by like these subtle like cues with how things are positioned. And it's like it's super minor, but it's a. It's like. It's like a microaggression. But it's like not. You know what I mean? It's like the micro. This is how you shift people's.

**[3:11] Speaker B:** It's micro influence. The language that we use influences your perception of things.

**[3:15] Speaker B:** So you have the same story, but one is fears of a bubble and the other is like ah, temporary hiccup with some accounting stuff. But like everything's gonna be fin. Yeah. And then there was another wild headline that I saw which was that Starbucks and Chipotle might be merging. The funny headline that I saw was like, who feels like a pumpkin spice burrito?

**[3:33] Speaker B:** Kristen, as someone who is a Starbucks former employee and aficionado, how do you feel about this?

**[3:39] Speaker A:** Dude, I would be excited mostly to talk about it from the M and A standpoint. Now do. Would I be excited if it went through? I don't know if I'd care.

**[3:49] Speaker A:** Again, I like, it's not like you have some mom and pop. Like this is Chipotle. So I don't know.

**[3:55] Speaker B:** Does Bill Ackman have a pretty key stake in Chipotle?

**[3:59] Speaker A:** That's a good question.

**[4:00] Speaker A:** I don't know. I wouldn't be surprised. Tim Hortons or. No, it was who. Who did the recent deal. And it wasn't even recent. Oh my God, it was so long ago. I think it was Burger king bought Tim hortons back like 10 years ago. This is a whole again. I know. Way Too much about like Burger King and the history. But they were bought, I think it was by 3G private equity firm Bob Burger King. And then Burger King bought Tim Hortons to do. It's called like a tax inversion based in Canada that is no longer legal anymore. But I don't know what the end result was for Tim Hortons, but it was the reverse. It was like the like a, you know, hamburger company buying a, buying a coffee company versus.

**[4:40] Speaker B:** Oh, I see what you're saying. Yeah, yeah, yeah, yeah. Well, I just feel like we haven't had that much M and A activity in this like food and beverage retail space. Right. I feel like the Kroger's Albertsons deal like two years ago, it was the last big headline we had that I can remember along these lines.

**[4:57] Speaker A:** And that was struck down, which I did think was an interesting choice because back to the, like they allowed Amazon to buy Whole Foods. Like Walmart's a huge grocer. So that was an interesting choice to strike that down. You know, obviously I love M and A and I was telling you I am so excited because we do these like hand drawn videos and I am so, so pumped to do something where we get like. So we're going to talk about neoclass a little bit and Oracle is technically like, it's funny, they kind of bounce back and forth between like are they hyperscaler, are they a neo cloud? But the, you know, Ellison's own or it's Larry Ellison who is the, was the founder of Oracle and he has, he is bankrolling his son's acquisition of Paramount. Sorry, no, of Warner Brothers. So Paramount's by Warner Brothers.

**[5:38] Speaker A:** So I'm like super excited to do, to do a deep dive into that again. M and A is so much fun. So like I said, I'm here for it because the more things we can talk about the better. But if I was a Chipotle eater, again, I don't know that it's necessarily necessarily a bad thing. Although it depends if they're going to like close down stores. Are they going to like, you know, because when, when you do an M and A deal, it's about synergies. It's about like where are the oftentimes cost savings sometimes revenue upside. But you know, maybe it's bad for Chipotle so I don't know. We'll have to see.

**[6:08] Speaker B:** I haven't eaten a Chipotle ever since there was like a very high profile string of food poisoning incidents like eight years ago and I haven't eaten there since so I don't know anything about them.

**[6:18] Speaker A:** Well, when I used to travel, I was looking for a chain that I recognized, so I usually would go and eat. And again, I actually like Chipotle, but I just haven't eaten there in a while, so.

**[6:27] Speaker B:** Yeah.

**[6:28] Speaker B:** All right, well, let's get to the meat and potatoes if you will. I guess the, the interior of our burrito today, which we'll be talking about the Neo clouds. So before we, we get into the exciting stuff that's going on with the Neo clouds right now, I think a lot of people take for granted that they know what a Neo cloud is or even what a cloud is and like where this whole business model originated and what it originally was and what's different about it. Now and then we can get into all the fun and sexy stuff. But I mean cloud services are something that has been around and been ubiquitous for companies for a very long time now. When you think about the traditional clouds, like the aws, the Google cloud service providers, the Azures, right. What these guys do is instead of taking like, I'm going to use my phone as an example, but like the physical hard drive, right.

**[7:14] Speaker B:** If you're a big business, if you're an investment bank and you have all this data and you have all these files, you don't want like giant hard drives filling up your offices.

**[7:21] Speaker B:** So you store a lot of your files and your data on these external servers and you rent this cloud time effectively from these companies and their storage and all of that is basically what makes the world turn. You know, I have like a Google Cloud subscription myself and I think they charged me like $2 a month or something like that. Yeah, I don't know, something like that. But like that's not how they're making their money. It's not off of retail customers like me, it's off of these giant enterprise level relationships with like those cloud storage.

**[7:51] Speaker A:** No, totally. It's just funny because I have a Dropbox subscription. I do Apple, Apple storage again, the cloud storage and I spend, I feel like I spend a decent amount. That's why I'm like only two. Like mine is much higher. But I also have, I have a lot of store, like I store a lot of stuff.

**[8:08] Speaker B:** Yeah, I keep a lot of my stuff on external hard drives. So like I, you know, most of our stuff is on my external hard drives, not on Dropbox or anything like that. Whereas I know you use Dropbox more

**[8:19] Speaker A:** like your hard drive.

**[8:21] Speaker A:** Yeah.

**[8:22] Speaker B:** And so those business models have been around for a very long time. They have their own exist preexisting data centers that they use then kind of.

**[8:31] Speaker A:** Can we talk about how you used to live next to like a giant data center in Tribeca?

**[8:34] Speaker B:** Data center in Tribeca, like right near some of the most expensive real estate in Manhattan.

**[8:39] Speaker B:** So for everyone who's like, oh, no, they're building a data center nearby house, not to discount your fears and worries, but like, I lived next to a data center for a very long time that was probably being used for cloud storage. And I mean, this is an ancient one. It probably was like a telecom data center back in the day that they've like repurposed over the years. So who knows what's going on inside there. But we used to, we used to joke that we thought it was like a CIA black box site. And then like somebody burst the bubble that it was a data center. I was like, oh, that's exciting. Who knows what goes on in there? But anyways, you know, then after the, after the emergence of bitcoin and kind of the proliferation of crypto, you then had a lot of these crypto mining companies.

**[9:19] Speaker B:** And crypto mining is something we've talked about this before. Crypto is not my strong suit. I like have a little bit of a mental block on crypto. Some of the language is really confusing to me. Like, it's just not something that I'm native to and it takes me a long time to understand it. So we're not going to get into the entire business of crypto mining because that is something that I am not an expert in. But there were a lot of these companies that did engage in this business of crypto mining. And in order to do so, you need massive amounts of compute. You need asics, right? These specialized chips and their entire job, they built these warehouses full of ASICs. And the whole job of those asics was to solve compute problems.

**[10:00] Speaker B:** And so they needed massive amounts of energy. They needed all of this compute. And they put them to the task of basically. If you guys aren't familiar with blockchain, I'm going to try to simplify this down as much as possible. You have to, like, solve a problem in order to. And the problems are geared towards the difficulty so that you can generate basically like one new Bitcoin every 10 minutes. And if the problems are too hard and you're not generating one new Bitcoin every 10 minutes, they make the problems easier. And if the problems are too easy and they're generating too many new bitcoins, then you need to.

**[10:29] Speaker A:** It's like taking the GMATs?

**[10:31] Speaker B:** Yeah, kind of.

**[10:33] Speaker B:** And so as a reward for solving these compute problems, these bitcoin miners get paid. Right. But the way that this was all set up and engineered from the job is there would be diminishing returns to keep this scarcity of bitcoin over time. So there was like a built in death of this business model that will, that is known, has been known since its inception in like 2009 and has been on the horizon this whole time. So as that, I think the original payment or whatever, the original reward for bitcoin miners when it first started was like 50 new bitcoin per block.

**[11:08] Speaker B:** Now they get six and a quarter, that's going to drop to three and an eighth and get halved. Right. So eventually it's like, you know, it's never zero, but it gets infinitely close to zero. So I'm no mathematician, but like, doesn't seem like there's a lot of joy left in that business of bitcoin mining. But there were all these companies.

**[11:25] Speaker B:** One of the biggest ones was Atlantic Crypto and there were all of these companies that had this business of doing this bitcoin mining. Like I said, they had relationships with computer chip providers, they had relationships with power providers and they had a lot of the physical infrastructure. So I'm just setting the background here of like there was this whole world of these bitcoin miners with this failing business model and they suddenly realized that they basically had all the raw ingredients to do what we are now doing with AI. And so in 2024 we emerged, companies like Atlantic Crypto rebranded themselves into core weave and are now all of a sudden the Neo clouds that we're talking about today. So that's kind of the backdrop and I want to admit I just wanted to make the distinction. No, I love that traditional cloud service providers, Neo clouds.

**[12:15] Speaker B:** And we'll explain exactly what these guys are doing and how that business model differs and then the origin of these and why it makes sense that a lot of these started as crypto miners and why a lot of the Neo clouds that started up that that didn't have that infrastructure already from being crypto miners failed and have failed. So with that, Kristen, I know you've done a lot of research on this. Where should we begin?

**[12:36] Speaker A:** Well, I want to go one step further, which is there's this other bifurcation, you know, you were talking about like the Amazon's, like Amazon AWS everything.

**[12:45] Speaker A:** You know, the other term people throw out is hyperscaler. And one of the big things that at least the way that I was sort of trying to understand because Oracle I mentioned is this one company that depending on who you're talking to and like, I don't know, like they sometimes call them a Neo. Like people will be like, oh, they want to be a Neo cloud or oh, they want to be. Or they're a hyperscaler. And I think Core we've actually wants to be called a hyperscaler versus a Neo cloud. But the traditional core we've done Oracle core weave.

**[13:16] Speaker B:** Oh, interesting.

**[13:17] Speaker A:** Core weave I think has, from what I have read, they want to be called the hyperscaler. Yeah, they, they call themselves because it's like they're, you know, they're scaling at, they're scaling massively.

**[13:27] Speaker B:** They are scaling, they're scaling cloud service.

**[13:30] Speaker A:** Yes. So the key distinction kind of goes back to like how do we get here? And so, and kind of like what is funding a lot of the spend and the capex. So your hyperscalers, if you think about them, right, it's your, to your point, your meta, your Google, your Space X, your whoever who has a profitable business, SpaceX and like that's a whole other can of worms. But again like Space X is sending rockets to space.

**[13:56] Speaker A:** So you have a company that is actually producing money. Again, Google is like a cash printing machine, Microsoft is printing cash. All these companies, they have wildly profitable businesses and so as they were kind of building up the, their AI capabilities and by the way, a lot of the hyperscalers do have again capabilities in not just like the infrastructure part of the layer. Like they kind of, some of them obviously have their own models, some of them have their own applications, like they'll have potentially some of them are making chips, so they oftentimes are doing other things as well. But the key difference, the way that like I think about it is they, because by the way, Oracle has none of that, they just have the infrastructure, but it's like do they have another profitable business that they can use to actually subsidize the massive spend?

**[14:39] Speaker A:** And so for a lot of these companies, whether again it was Google or whoever, they were able, because they just had so much money, they could actually spend and build out that infrastructure with the cash flows from their other businesses. Whereas your Neo clouds, especially as they started to scale, like I was shocked, I think that Core, we've only had like 10 data centers or something a few years ago and like in terms of their compute capacity, it's just, it's gone parabolic. But they're, they don't have Those cash flows to finance the construction of like what they need to do now what they do have to your point is a lot of them have the powered shells.

**[15:16] Speaker A:** Like in some cases, either you have certain companies that actually own the power shells. And before actually talking about power shells, let me back up.

**[15:22] Speaker A:** Another framework I found helpful was Jensen Huang's discussion of like AI factories. So if you think about like what you need to produce compute, you need the chips, you need the energy, you need the powered shell, which is just literally as a building that is, has, you know, the ability to plug in and has power that is able to at the right amount so that you can produce the compute. And again, power is one of the. It's actually in the US like the thing that we are the most constrained by.

**[15:48] Speaker B:** Yes.

**[15:49] Speaker A:** And so a lot of these companies have the access to those powered shells, whether it is because they own it. So there was Applied Digital and Core Scientific. They own the powered shells now. Fun fact, Core Scientific actually let me talk about Applied Digital. Applied Digital was in the business of trying to sell compute. They were like, we don't want this business. We're getting out of this. Like it's too risky with the compute. Like we're out of that. We are just a landlord company. We are just renting to the core Weaves, the Lambdas, the whoever. So in some cases you might have again, they were former crypto miners as well. They had again, they had the buildings and they want to own the infrastructure there. Other players, again Crusoe's kind of your classic. They actually started producing power and so they became a vertically integrated player from that direction.

**[16:36] Speaker A:** And they again are selling compute. They are again, they have the power, they have the chips, they have the infrastructure. But what I think is fascinating about three of these companies in particular, Core Weave, Lambda and Nadias, they don't own, I mean they're starting to own more, but historically they didn't own any of it. They did have the lease, like they did have the space to your point, because they were using this to mine crypto. So they actually had the leases and so they.

**[17:02] Speaker A:** But what they did not have is money to buy the chips. And so that was where you started to see in like I think it was 20, 24, people came up with a structure for financing the growth of these companies by basically using asset backed lending. Or you can think of it as like project financing a little bit, but basically it's this like shipped back financing. And so what the banks said is they were like, okay, Core Weave, Lambda, Nebbyous N scale you can use, you have contracts like you have a contract with Meta or with Microsoft or with whoever, right. With Anthropic they need the compute and they have an agreement to take or pay for that capacity for some time period. It's usually like a five year contract if you will. And and so we can use that promise of those cash flows and we can look at what your, what the cash flows are going to be that you're going to get.

**[17:56] Speaker A:** We kind of haircut it a bit using what's called the debt service coverage ratio. We you know obviously take out the expenses and all that and we can raise, I'm making up a number, right, $100 million worth of worth of debt backed by those cash flows. And then when you buy the chips you can actually those chips and the contract are kind of part of an SPD and the lenders, that's their collateral.

**[18:16] Speaker A:** And so you started to get again core weave for example I think like half of their debt, they have significant amounts of debt and half of their debt is this chip back financing actually used to be a higher percentage. They've started to do converts and like other financings as well. But like the point is a lot of the growth is funded by these contracts. And so what is interesting about this whole thing is like I will say again it looks a little bit as someone who has come again like I said from a background where I started in CDO structuring, I did work in converts for a bit. I worked in project finance. Like I've seen, you know, I worked in financial sponsors. We were doing leverage buyouts where you're looking at how companies get bought and what the requirements are for the purchaser if you will to have skin in the game.

**[19:01] Speaker A:** I did think it was interesting that for these chip back financings, what's called the advance rate meaning how much money can you borrow? If you need to buy a hundred million dollars worth of chips, how much money can you actually lend? And with some of these like really high quality off takers like a Meta or a Microsoft you can borrow almost, I mean like 95%. So like you could borrow $95 and go spend the, you know what I mean? So like you will need to put in a teeny, teeny amount of, of equity of skin in the game. And again as someone who used to work in financial sponsors that is crazy.

**[19:34] Speaker B:** Like usually the think about buying a house, right? Yeah, 20% down. If you're a normal person there are programs where you can put 3% down, 5% down 0% down if you're a veteran or you know doctor's loans give you like 2% whatever it is. But like that's unusual. Most people want to see a much higher down payment. So it's that, it's that ratio that is pretty alarming for a business model like this to be so high.

**[19:59] Speaker A:** Yes. And then the other piece is that there's like the mismatch between the timing of or I should say it's not actually a mismatch. Usually in like a power purchase agreement in the power and energy space you would have very long contracts. So there's called like a power purchase agreement and it can be like 15, 20 years, maybe longer. Obviously if you're building a power plant those things last a really really long time and the debt is going to be for a significantly shorter duration. However with the chip back financing the contracts in some cases they're as short as three years. Like in other cases the contracts for

**[20:29] Speaker B:** that compute are relatively short.

**[20:31] Speaker A:** Right. So it's like if you think about a power purchase agreement like this is how you oftentimes will get a power plant built is you have the agreement from some utility or from whoever who's going to buy the power. You know, you have that locked in contract and so then you can size the debt based on that promise that you're going to get from some again high quality off taker if you will.

**[20:49] Speaker A:** But with this particular scenario you usually have shorter contracts. Now apparently they are extending like there were some. A few years ago there were only three. Now they're a little bit longer but the contracts can be with like your anthropic or your open AI and we just again there was that article about how open AI's cash, sorry their revenues, their ER was like 20 billion what they thought it would be. And with all the emergence of the open weight models and like this, you know, concerted potential effort to slow things down and people calling what the end of humanity, you know are these, are these frontier model companies going to be able to make money enough to like satisfy the obligations that they have?

**[21:30] Speaker A:** This was another big thing that kind of spooked the markets a few. I think it was last year with Oracle because again they had a contract with OpenAI and people like OpenAI is not profitable. How are they going to be spending all this money? Like it's basically on a hope and a prayer. So there is a little bit of concern there. Now what I will say is to play a little bit of devil's advocate on the other end These contracts are also right now typically for a price that is below market because there is such a high demand for compute. So when these contracts wear off, these new clouds are able to rent out the GPUs on an hourly basis for more money. So again that is kind of like +1 in the compute column category if you will. But the cash, I mean look the, the, these companies are not profitable because so much if you think about like bottom line, like we think about what it does it take to actually like earn net income.

**[22:20] Speaker A:** But we're not just talking revenues, we're talking about like bottom line. And because you have these assets that depreciate quickly, that's been a huge obviously point of contention. We've talked about the useful life for GPUs and all that. Now currently these hyperscalers and neoclouds use from an accounting standpoint like five to six years. Apparently a lot of hyperscalers are using chips that have, that have been around for longer than that. So like it is possible that maybe they can be used longer. There is also the Michael Burry allegation that they should be shorter and that like because of the advancements in technology that they are usually going, they're going to only be like two to three years. So there is a little bit of debate there. But let's assume the actual depreciation that they state which is five to six years. You still have a significant amount of depreciation which hits your net income and a significant amount of interest because so much of your financing is debt funded and with rates, right?

**[23:14] Speaker A:** Historically we had like the Zerk error where the rates were nothing, but now a significant amount of revenues ultimately get eaten up by your interest expense. And rising rates are obviously very negative for this type of structure, if you will.

**[23:28] Speaker B:** One other interesting thing to talk about is kind of the Nvidia of it all. So when we talked about the original chips that these guys had lying around, if you will, they were Asics, right? They were not Nvidia's GPUs. So in order to do the, and again the diff, I just want to be distinct that like the service that these guys are providing relative to the og, like cloud classic, if you will, the NEO clouds, what they're doing, it's just AI models, right? They are a one trick pony. Each of these different companies, the three that you mentioned I think are probably the three that are dominating the NEO cloud landscape. You've got your core weave, your Nebbys and your lambda. They do have distinct business models. Within them, both in terms of how like their relationships to the equipment and stuff like that, but also the service that they provide in terms of like what you actually get when you sign up for their services.

**[24:18] Speaker B:** But just to keep it really simple and bare bones, let's just say that they're all using Nvidia GPUs. And the reason that they're using Nvidia is because a lot of the training software that you plug and play onto it, rather than having to say like, here you go, why don't you every time you want to train something, write that from scratch and figure it out. It's like, no, all of the stuff that tells people how to train models was designed for the Nvidia GPUs, which just makes it easier. This is not all NEO Clouds everywhere, all AI training, to be clear. But let's keep things really, really basic and just talk about the fact that the vast majority of the GPUs being used by these Neo Cloud providers are Nvidia GPUs.

**[24:58] Speaker B:** It makes the whole process easier or whatever. How is Nvidia then involved? Because Nvidia touches all the different parts of these guys and it's part of our broader conversation that we're having kind of on a thematic basis here about circular finance.

**[25:12] Speaker A:** Well, yeah, I mean, and back to your point, I do, I will say one other kind of thing in the Pro Nvidia column is they are general purpose chips. So I will say, like, it does give you a little more leeway to kind of repurpose what the chips are being used for.

**[25:25] Speaker A:** But yeah, to your point, Nvidia is invested in Core Weave. It is invested in Nebius.

**[25:32] Speaker B:** It's definitely invested in Lambda.

**[25:34] Speaker A:** Yeah, so they have an investment in basically like a lot of these Neo Clouds, but they're structured differently. Oh, I actually didn't look into the exact structure. So if you happen to know that,

**[25:44] Speaker B:** well, some are equity investments, some are debt guarantees, some are the guarantee to then like lease that check.

**[25:51] Speaker A:** Sorry. Okay, so let me, let me talk specifically before we. Because in Core Weave they do have a 11 ownership stake and then they also do have an agreement to backstop unleashed capacity.

**[26:02] Speaker A:** So I think it's up to 6 billion, $6.3 billion if I remember correctly, that they will pay. If you have for example, a contract like someone rolls off and they.

**[26:11] Speaker B:** With Meta or whatever.

**[26:12] Speaker A:** With Meta and they don't renew, then Nvidia will step in and will essentially like pay them at unused capacity if they can't find somebody else to. To do it. So like again, there is significant demand for compute. So that is potentially not a problem like right now. But if for whatever reason, like you couldn't actually find a buyer, they will step in and they will backstop that. So that's what we mean by they will backstop that capacity. And then to your point again, you have done, like I said, a lot more on some of the other ways, like Corey is the only one that I looked into. So if you want to, well, just

**[26:42] Speaker B:** know, I mean, honestly I think that's sufficient for the purposes of our conversation. Just understanding that they do have a direct equity stake in some of these. With some of these they do have an agreement to like you said, step in and be kind of like the buyer of last resort, if you will, for like using some of this compute. And then with others they're actually using their own credit rating to back the borrowing of these companies who otherwise wouldn't be able to go out into the market and tap them at investment grade levels because the business model looks so shaky, if you will.

**[27:13] Speaker B:** And well, and I want to, like you said, the mismatch of the lease timing, that effectively they're lending them their credit rating so that they can borrow at levels that make this whole business make sense or else it doesn't make sense.

**[27:23] Speaker A:** Well, yeah, and just to kind of reinforce that, the way that the lenders and the rating agencies look at these contracts, if you had for example, the Meta agreement, they look at Meta's credit rating to come up with what is the interest rate that you should be using to size the debt. So core weave is like a thing. It's like a triple C rated company. They have again so much. I want to look up their leverage. Their total debt to EBITDA is 9.2 times, which as someone who used to work, like I said in financial sponsors, like you would not see leverage go above 7.

**[27:58] Speaker B:** Right. For LB, we're talking about like leveraged buyouts, we're talking about. But again that's usually my ton of debt on companies.

**[28:05] Speaker B:** And that was like, that's like the ratio you'd seen.

**[28:07] Speaker A:** Yeah, that is a very, very aggressive like. So basically what they're saying is it's going to take them nine years with their current level of operating cash flows to pay it on the debt. Like that's how you can think about that leverage number. But nine times is insanely high. So the point is that like they are, they are a junk rated company from a risk standpoint. That's just like how we tend to Talk about ratings. It's like you have your triple A, you have your double A, your B, whatever. But if you look at the companies that are agreeing to pay them, I mean, Microsoft has some agreements. Microsoft is literally rated triple A. They are rated higher than the US Government. There are two companies that are aaa. Microsoft is one of them. So the way that they then look at, okay, how should we think about sizing this debt?

**[28:45] Speaker A:** Because they are getting payments from Microsoft, so they care about can Microsoft make the payment?

**[28:50] Speaker A:** Not so much. Core Weave. So that's why using the special structure where you have the contract and the chips in this spv and so you're basically looking at can Microsoft make those payments. And if you feel good about that, then it doesn't matter what Core Weave's bar like total debt is. That's like elsewhere on the balance sheet, like, whatever. So you borrow off of that. And then to your point, right, there are other companies out there that have much lower credit ratings than these companies. Like again, an anthropic and an open AI, like they are not going to be AAA rated. I mean, we don't even know if they're going to be able to be profit. Like they are not right now.

**[29:23] Speaker A:** So if you have a company like that, it might be helpful to have Nvidia say, hey, I'm going to backstop the capacity just in case. So then to your point, you're borrowing and the lenders care about can Nvidia make those payments? Not can, like again, anthropic OpenAI whoever. So yeah, just wanted to like double click on that because I do think it's a really interesting point that the rate that you borrow at is based on the person who's going to be paying for the compute, not Core Weave itself.

**[29:50] Speaker B:** And so one of the things, you know, the Nvidia of it all that makes it so interesting is when we look back at prior examples like Lucent, right, where there was so much risk embedded in the system because of all of these investments that are. When we talk about kind of like the nature of circular financing, we talk about the risks associated with it. You know, when we look back to prior examples in history and we look at companies where it's like, hey, listen, I, I'm going to keep making money like gangbusters as long as all of these people are making money to pay me. And by the way, I'm invested in them and the value of my investment goes down. Nvidia is stretched more thin than Lucent was at the time, which was like kind of held up as like one of the big, you know, dot com bubble bursting.

**[30:37] Speaker A:** I am curious and I'd love to and I haven't looked into this but like kind of looking at what are the obligations that Nvidia has relative to how much in cash flow they're earning because they are earning a billion dollars a day. I mean they are from a. Sorry I should say their revenues are a billion dollars a day. They are earning so much money and they do just kind of keep getting themselves embedded more and more into the AI ecosystem in a way that I actually think is like quite savvy. With that said, I almost like do wonder it's stepping into the open weight models like which helps them in some, some areas of their business might potentially hurt them in other areas. Like I said, if again you have an anthropic and an open AI who like now can't actually figure out how to monetize the way that everyone kind of believes that they should or will be able to that they can't fulfill those obligations, like does that then hurt them?

**[31:24] Speaker A:** Someone literally messaged us or commented on a reel I did about Anthropic's upcoming ipo and this is before. I mean look, we, we don't have, we have a cash number for them. That was from the Leaks prospectus. I don't believe there's a debt number, but we had talked to some of the last episode that like these take or pay obligations that your neoclouds literally think are secure enough that you can lend so you can borrow significant amounts of money against them, do not show up as debt as a liability on the balance sheet of these companies. So these are off balance sheet liabilities which I think is like it's something that people don't necessarily understand. It was something I was surprised that they wouldn't be on balance sheet but they're, they're not until I think like the capacity has started to be delivered. So that makes again like as we're thinking about your, your hyperscalers because I was like oh, how much money are they borrowing?

**[32:12] Speaker A:** Because they are going out and they are starting to tap the debt markets. Like I mean obviously SpaceX has tapped the debt markets. We've seen Google tap the debt markets. We've seen a lot of these guys actually raise debt because they actually need to use it to fund the capex versus just using their cash flows. But when you look at what their leverage is, I think I looked it up and like Claude said that their leverage is only like 0.4 times again debt to even duh. But remember that does not that does these take or pay commitments which by the way even if they were a liability they might not. They wouldn't necessarily be counted as debt. But it's just something to kind of be cognizant of that like there is a lot of like there's a lot of money that's on the hook. Like people owe a lot of people a lot of things that I don't think are being captured.

**[32:56] Speaker A:** And someone also messaged on one of our things being like, you know, how do you guys see the world? And we were like look, you're talking to two women who kind of had started our careers when shit hit the fan again. You started at Lehman Brothers RIP and I started in CDO structuring in 2007 and watched my group go from 40 people down to three people in the span of about a couple months. My group just started to implode.

**[33:17] Speaker A:** So I am always like very bearish even though I do think there's potential optimism and I am optimistic long term and like where things go. But we just saw what happened with situational awareness Leverage can cause problems in the short term even if the long term. Leverage leverage is always scary.

**[33:38] Speaker B:** Yeah and you know you mentioned anthropic stuff coming IPO but Lambda is likely going to IPO as well. So one of the three, I mean core weave is already long since public but Lambda is going to join the ranks of publicly traded companies, you know with not the world's most favorable coverage around Neo clouds. And I think it's going to be really interesting to see how that how

**[34:00] Speaker A:** that goes N scale is another one. They are, I think that they're targeting a 35 billion dollar valuation and like there was some going concern language that management.

**[34:14] Speaker B:** So anyway, anything else that we need to cover about these neoclass? I think this has been a pretty good question crash course in them and you guys are going to hear this name a lot in the overall market moving news. What do we miss?

**[34:25] Speaker A:** Nothing. I think that just giving you guys a sneak peek at what we have coming next. So we actually went back and went and read the smartest guys in the room. Mostly because it's a story about Enron.

**[34:38] Speaker A:** Because everyone who's throwing out allegations of circular financing are like Enron, Enron. And I want to be very clear, Enron was like outright fraud. Like there might have been some similar accounting things that were used but that's not why they got in trouble and nor was why it went bankrupt. It went bankrupt because of like actual fraud. So we are going to talk a little bit about that.

**[34:56] Speaker A:** There are some really, I think you and I have been like chatting non stop over this past week about like our thoughts and all of the lessons and some interesting things related to like actually the AI era in a way that I don't think people have been talking about. But we're really excited. So if you guys want to like read along, you can get caught up. So next week when we publish that episode, you will be all caught up. So yeah.

**[35:18] Speaker B:** Awesome. Well, thank you guys so much. It's always a pleasure. And if you will, please share the show with a friend and give us a five star rating wherever you're listening.

**[35:27] Speaker A:** Thanks so much.

**[35:27] Speaker B:** How do we stay on top of all the news that's moving markets? We use Ground News. Ground News is basically the Spotify of news. So it first lets you look through all the coverage, then it tells you to how who funds each news outlet, where it leans politically and how factual it is. My favorite feature is the blind spot feed where they share important stories that are being ignored by one side of the political spectrum or the other.

**[35:48] Speaker B:** Go to groundnews.com skinny to save 40% off when you subscribe. That's groundnews.com skinny and save 40%.