CoreWeave PANIC Is Hiding a MASSIVE Opportunity

Fear is taking over two of the biggest names in the AI infrastructure space, Cororeweave and Nebus. See, investors are worried about the debt, massive spending, and whether these companies can actually deliver on their promises. And while some of those concerns are legitimate, I think that the market could create a massive opportunity for investors willing to look past the panic. But not all opportunities are created equal. And while I hold both of these stocks, I'm only buying one of them aggressively at these prices. So, let's break down what's really going on, what the market is getting wrong, and where I see the biggest opportunity. Because when stocks sell off, this is when we really have to pay attention and look for buying opportunities. We're going to go through top to bottom what the heck is happening, what I'm doing with my money, and where we go from here. Let's talk about the core financing problems. I wanted to start with Michael Intrader. He goes through some of the talk about actually how they set up their financing, specifically GPUbacked financing and why whenever people bring out these poly markets like coreweave is going to go bankrupt. It's just completely not true and in fact coreweave is not even liable for a lot of this debt, it's actually a lot of the GPUs that are could potentially be stripped away. But let's listen to how Coreweave actually gets paid because it's similar to a payroll loan rather than an unsecured loan like people might believe. Great. We're going to sign a contract. Once I have a contract in hand, then what I do is I create something. It's not a particularly creative name. It's called the box. >> Yeah. >> Right. And what I do with the box is I take my contract with Microsoft and I put it in the box. I go to Jensen and I buy the GPUs. I put it in the box. I take my data center contract, I put it in the box and now the box governs cash flow >> and it has a waterfall of cash flow that comes into it and goes out of it. And so the way it works is then I build the compute and then I deliver the compute to Microsoft and they pay the box. They don't pay me, >> right? >> It goes into the box and the first thing it does is it pays the data center. It pays the power bill. It pays the interest and the principal and then whatever's left flows back to us. >> There's no way to lose if those debts remained the amount that you actually took them out for. The problem is is some of these are shorter term loans. If they wanted more debts, then those new debts would come in at higher interest rates because interest rates are starting to climb. These are essentially some of those boxes that they were looking at. and that within those is the debt that you need to actually finance the amount of money that you're making in the contracts in gray. But the most recent one and the one that people are bringing up right now, which is roughly about $1 plus billion, is that the actual debt that's due is starting to skyrocket because bond holders don't want to hold it. And so interest rates are spiking up. And that's exactly what Social Capital highlighted here. Now, this is only on yield of $1.25 billion. This is not their entire book of all debts. The problem is though is that if they took out new debts, this would be the cost. Let's listen to Chimath talk about this. The most important thing that you should always look at is how the sharps, the super smart money, prices, risk, and when you're dealing with companies like Cororeweave or Nebus or Iran, etc. that have to borrow a lot to fund these boxes. The smart money is trading what's called a credit derivative swap. Okay, this is default insurance and these default swaps are meant to ensure the bond holders against a potential default. That risk spiked this week to about 953 basis points, which it doesn't matter what that number is, but as an imputed risk, it's about a 16% chance of a default in any given year. And over a 5-year duration, what that math implies is about a 55% chance of a default. So, it's a coin flip whether they default or not. That's not what I'm saying. That's what the insurance market was saying about their debt. And I think that's something to look into because those guys I think are the smartest money in in all of this. There is a very very high chance what the market's actually saying and this actually spiked up to 65% chance of default over that 5-year term. I think it's actually come down a little bit since then, but it's very very high. Now, this doesn't mean Coree goes bankrupt. This means one of those boxes may and that you're essentially underwater on the amount that you're going to pay versus how much you're receiving after you've paid like you said the electricity bill, you paid the banks, the interest, the principal, all of that. You could potentially still have a negative number on the other side if interest rates are spiking up. Right now, Core Weave's fine. The way that we were going before is that interest payments were actually getting lower and lower. I mean, total interest paid was going higher as they took out more debt, but the amount that they paid on their debt was going down until this sudden spike up from worried bond holders or whichever bank or financial institution was dumping their debt. So, the problem is, and you can see it here in the purple at the bottom, what lenders would charge for today if they ended up taking out new debt. If we end up looking back to Core's backlog, they could essentially just for all the backlog that they have today, they have their debt that's assigned to that backlog. new backlog needs to find new debt in order to actually finance them on time. The problem is we already know that Coreweave has over $2 billion that's about to be assigned in Q3. We know that they are going to end up getting more contracts and that the amount of demand that's coming online for artificial intelligence is continuing to climb. This is the problem. So, it's not a matter of whether or not Coree is a platinum neocloud. We already know that. We know that they have some of the best deals. We know that they have some of the best pricing. We know that they can set up clusters faster than anyone. These are all great signs for Cororeweave. And it's not that they won't deliver compute on time. It's that maybe spot prices have gotten so high recently. And we'll talk about this a lot more later on, but essentially the way that people are looking at pricing some of these data centers or some of these deals is the amount of millions of dollars per megawatt. Thousand megawatts in a gigawatt. A gigawatt takes roughly 50 maybe $55 billion to get set up if it's fully on Nvidia hardware. Exactly what Coree does. So we know that Coree they receive $20 million per megawatt for a lot of their longerterm deals. This is like averaged out across all of the deals that they've signed today. A lot of shorter term deals are going for as much as 50 if not a little bit higher in terms of millions per megawatt. And David Saxs will talk about this later on, but payback period is extremely important. The faster you can get rid of your debt and then go out and find new customers, the better it is. Coreweave has gotten a little ahead of itself and sold all of their initial capacity before spot prices started to climb up. And this is the problem. >> Any business model that thrives on a financing spread gets cornholdled. Any and every all of them. It just it will it will not matter when the 10 years at 6%. It won't matter. Doesn't matter what you do. It doesn't matter what you claim to do. You're going to get run over. You're going to be roadkilled. >> This is across QXO or SoFi or Coreweave or any of these companies. If the cost of their debt goes out of whack, then that spread between the risk-free rate and how much Coree bond holders want to hold is going to be affected. And all of the math that you were going to do at Coreweave to make sure that the debt was perfectly financable through these GPUbacked asset financing, the math just goes completely out the window. If what you were doing before was 8% and now what you're doing is closer to 16%. That math is hard to figure out. This is not a demand problem. So is there still money to be made? If coreweave can figure this out, hyperscaler cloud backlog and this is only amongst the largest companies. SpaceX is not even in this list and they have massive amounts of backlog. There's a lot of businesses nscale and lambda and together AI and MRO and all of these businesses are not included in here. And yet these businesses have over $2.4 billion dollars amongst these five names worth of backlog in the last quarter. And every single quarter that ticks up by hundreds of billions of dollars. Even though we're getting new GPUs online, even though we're servicing them, we're getting further away from solving this problem. TSMC, which is sort of the foundation of all of this back in Q2 of 2026, said, "I believe from this day on all the way to probably 2029 to 2030, the demand is very strong. Whether in between there's a dip or not, I'm not very sure. The trend is so robust that I believe that we are witnessing a kind of new industry. I would say the new industry maybe calling it the AI industry. They then went on to say next quarter they said the demand is actually still far above our capacity today to supply. They can't even get through their own backlog. We know this from even Nvidia. They said next year's overall sales were supposed to be essentially they could do 100% year-over-year, but it's supply constraint, so they're only going to do 70% year-over-year. Still extremely high, but impacted by this. Lummentum CEO said, "We are practically sold out all the way through 2029. We can only supply roughly 30% of what we are being asked to supply." Let's go on to Ben Horowitz, which was also on the podcast. Ben talked about this as well. He spoke with Andy Jasse about some of the concerns. That's the current CEO of Amazon. Let's have a listen. >> This shortage is so bad. You know, I was speaking with Andy Jasse who runs Amazon yesterday and I said, "Yeah, like we're out of everything, you know, till 2028." He's like, "Oh, it's beyond 2028 bit. Like, I don't think we get stuff back like end of 2028 maybe." Uh, so everything else is sold out. So he he has the advantage of that he's got supply in a great shortage which is extremely different than the internet bubble. The internet bubble we didn't have enough demand because we had to build the internet people had to get broadband you know you needed uh all this stuff people had to buy you know originally they had to get their browser app fries all this kind of thing. There weren't enough consumers on the internet to absorb all the bandwidth that had gotten built out and they had a massive amount of debt, not unlike uh we're building today, although not as big. Um but but the lack of demand killed it. We don't really have a lack of demand in AI. So I think the dynamics will be different. Uh I do think like the for sure if you get into a pathological debt situation that can wreck anybody. Uh it's why um by the way we're not invested in any of these uh Neo clouds either for that reason. >> See Ben Horowitz which is A16Z they're probably the largest venture capital fund in the world right now and they're not even investing in the Neocloud space because of what they said about this almost parasitic debt situation that can happen where every single time that you are potentially about to get compute online and then that money is coming back to pay off your current debts. you say, "Okay, well, we're not actually going to pay that off right now because we are going to focus on taking out new debt to sign up another data center." And if we were doing this one by one, one data center, then creates the next data center, then creates the next data center, it would be perfectly fine. But Coree wants to go from 1 to 5 to 15 to 45 to whatever the list is after that. And because you're keeping up that exponential curve that essentially you never end up getting to a healthier financial state. If they can slow down on some of that debt and potentially not risk having to give up market share, that's the really good sign. And that's essentially what Jensen Wong is hoping that we are going to see that AI market turmoil end up passing and that people are still going to end up spending and that we will get past this debt situation. Regardless, this is leading to some pretty interesting opportunities. Coreweave over the last two weeks is down 32.3% and then Nebius also down 32.2%. Almost identical in the last 9 days. The real concern now is will data center development slow. Maybe Cororeweave and Nebius and many of these companies will be fine but we just won't be able to build as fast as what we wanted to previously. This is essentially highlighted from Hawk 10, the CEO of Broadcom that said capital, not chips, memory, power, but capital might become the biggest AI constraint. Then Intel CEO Lip Bhutan said the exact same thing. Capital in the cost of capital is the frontline challenge for a lot of these data centers and hyperscalers. They said neither sees demand cooling soon. and Hawk 10 said compute demand outstrips supply and could stay that way through 2029 possibly into 2030. So demand is super high, but if we can't finance these data centers, maybe it ends up slowing down how many data centers actually get built. Microsoft probably being the greatest example of this just cancelled a 50 megawatt data center leased out to Nscale that was expected to go in Essex UK and then found out that they won't receive grid power until after 2030. So they said no thank you. We're actually done dealing with this and we'll move on to some projects that might end up going a little bit quicker. Maybe an interesting time for Bloom Energy as a stock, but also Bloom Energy needs those data centers getting permitted, needing chips. It's not one individual bottleneck, let's say power, memory, or optics that's stopping it. It's sort of the combination of all three and whether or not banks believe that you're actually going to receive that money. If we end up looking out to Urkott, this is the Energy Reliability Council of Texas where they said we are seeing essentially 438 gawatt worth of interconnection requests in which the last latest quarter said we brought on 4 gawatt that expects to be energized in Q4 of 2026. Currently, the entire grid has 91 gawatt and you have over 440 gawatt worth of requests. We're never going to get ahead of how much actual demand there truly is. But what happens during these moments, data centers and GPU spot pricing actually skyrockets. And most interesting phenomenon because people realize that okay, if we can't get new data centers online, if cost of capital is skyrocketing, we need to pass that on to the consumer. Doesn't mean that the consumer actually accepts those new prices. But in this case, they are. And even if you look over the last let's say 17 days where this brings the Blackwell 300s up to another all-time high and now we have Vera Rubins coming online which will be even priced much higher because they're more efficient and then that millions per megawatt. This is another pretty interesting stat. Well, that is the sort of metric that we look to to see whether or not companies are doing well. Megawws is the energy but not the compute. People are actually renting compute, but we're measuring it through energy. So, as we bring on new chips over time, the overall amount of millions per megawatt across all neoclouds and hyperscalers is going to trend upward as we use less A100s and we use more Vera Rubins. But let's go ahead and listen to Chimoth really quickly talk about some of this slowdown and the impact that this could have on businesses. >> So, let's just say you're a big mainline, you know, GI bank, right? JP Morgan, City Bank. Could you imagine that the reason why you can't service your customers is because the underlying cloud that you use can't actually secure and grow enough power? And so, back to the other moat that I think is really valuable is to own the land, the power, and the shell because it's such a constrained asset. Back to what Andy told you, Ben, it is so critical and it will stop the best experiences on a dime. So if you're a startup, I actually think the bigger problem is you'll go to these clouds, you'll be like great, I'm ready to launch this whisbang product and they're like, "Suck it. We have no capacity for you." >> It is one of the more interesting things and I think actually a very big bull case to where spot price is going over time is because capacity is going to get harder to put online and that companies like Cororeweave that actually do own the data center or at least colllocation. They co-own the data centers and are leasing them out. That's the interesting thing about Nebius. They own a lot of their data centers. They do collocations as well, but they own a lot of their data centers. They own the power. They own the land, the shell, all this stuff. As we go towards more of this constraint, it's actually going to change the pricing per chip and could actually lead to faster payback periods than before to make sure that we can set up these new data centers as long as we can afford it. As long as the task that people are using with artificial intelligence is worth paying a higher rental price. I'm choosing to buy Nebus here. Like I said, they own their own data centers. There's other things that I wanted to talk about that's actually making me realize that this core financing situation is much more bullish for Nebius than not. Even though the selloff is proportionate and they both dropped roughly 30%. A little over I think they were both 32%. Let's go ahead and look through why I like Nebius a little bit more in aggressively buying this company. They said, "We build the basic layers starting from ground, land, power, shell, and they design and build their own data centers and fill them with their own design racks." There's a lot of companies, and I won't call out individual names, that are looking to lease out the land, are looking to lease out the data centers, are working with Dell, SMCI, HP to help with the racks, are trying to find any way to increase the speed to get to that data center online because they know over time the payback periods are so high that they're going to make their money back working with a company like Dell or SMCI. And Nebius does, too. But the interesting part is that they have the engineering expertise to actually do all of this themselves which very few do it from top to bottom and are the only ones that own their actual full stack. They said on top of that we have a product which is our own multi-tenant cloud. This is the same kind of cloud as the hyperscalers have. You know the big names AWS, Azure, GCP these are examples of multi-tenant clouds. We have this full stack as well. Coreweave does not have this full stack. A lot of them do not have this multi-tenant cloud. In fact, a lot of them sell bare metal or are actually uploading a lot of the software directly onto the GPUs. They're not creating virtual machines to run this, which can actually help with security. It can help with latency, but whenever you're trying to deal with smaller customers, they might not book out an entire Vera Rubin. And so, you have to kind of split it up like what you're seeing from Nebas. On the other hand, both of these companies whenever people say, "Hey, but Cororeweave has been massively backed by Nvidia." And that's true. And more so than Nebius, they also have a $6.3 billion back stop and an additional $900 million they've invested all the way back in 2023. So, Cororeweave is their number one. But the fact that they've also invested $2 billion in Nebus gives that same sort of credence that they also do not want Nebus to fail in any sort of measure at all. They have the stamp of approval from the biggest name in the space. And if we go back to the cluster max ranking, we also just found out that Nebius is now the second and only company in the mix to hit that platinum status ranking for the best manage platforms. Corweave's there, Nebius is there, but they're both up there. But now we go to the main part. A lot of those is what are they similar to in terms of the backing from Nvidia or platinum status and exemplar clouds on all of their different products. We know this. This is where they really differ. Nebus' cash position is somewhere around $8 billion. And in terms of long-term debt, which is a lot of it, although core, we've actually pioneered this box strategy of putting in GPUs and making sure that every component gets paid before they do, which was a really great way of seeing lower cost of financing over time. Nebius also adopted this similar strategy. While they pioneered it, it's not exclusive to them. There's no copyrights on a debt financing strategy. So, Nebius also does this. However, they don't need as much debt. 50 to 60% of their capex is prepayment from their customers. I don't know how they do it, but they do it and their customers pay them upfront to set up the data centers ahead of time. Customers like Microsoft, customers like Meta. Corweave, on the other hand, does have 5.5 billion dollars in cash and they've got tens and tens of billions of dollars. This is just long-term debt. This is not also including the many other components that could lead you all the way up into the 50 plus billion dollar range in terms of the debt that they actually have. Coreweave is a different beast whenever it comes to the debt problem than Nebus. Now there's still worrisome problems within Nebius, but there's also interesting parts as well. They also have over $3.75 billion in equity for a company called Click House, a company that is expected to do an up round that might be 100 or 150% higher than where we are today, which would make Nebus' stake even higher. They have a what they said worth billions of dollars, a majority stake in AVID. This is a self-driving company that they've also seen investments from Uber and Nebus has made a lot of money on this as well. There's really not strategic investments that you can point to from Coree to say, well, we also have to factor this into their overall valuation. On top of that, it's just Nebius has been seeing insiders buy shares of their stock, planned purchases. These are small and planned purchases happen more than once. So, I'm not worried about the actual amount, $50,000 of Nebia shares, whatever, few hundred shares, that part, I don't care. It's just the direction of where insiders are going for Nebas. A lot of these people like Arotti, he's a billionaire, right? Yandex had already made him a very very successful entrepreneur and so he doesn't need to sell Nebus this early. Michael Nrader on the other side and in fact a lot of these co-founders and participants within Cororeweave have only just recently IPOed have recently seen their shares unlocked. They've not had access to this level of wealth before and so they are pulling out what is a combined just amongst the co-founders $3.3 billion worth of stock. Amongst some of the other large beneficiaries, it's now up to roughly 5.5 billion of core stock since IPO selling no buying activity. So the comparison there just gives me a little bit of insight showing that Nebius is continuously buying. Why is Nebius falling alongside Core if they are so different? And it comes down to this. The secured debt facility we raised in July, it was $750 million was on the back of one of these deals. And with $40 billion in contracted backlog, we will do more of this. They're telling you we are going to do a lot more raising of debt as we turn more into coreweave because we're running so fast. We have to. >> So there's one metric that the more I think about it, the the one metric I would really keep an eye on is payback period. To me, that's like the northstar metric here. So, Cororeweave says that its payback period is 2.5 years. Their typical deal is five years. They pay back within 2.5 years. SpaceX, they pay back in roughly one year. That makes sense because I think that SpaceX is monetizing at about 50 billion per gawatt and core at about 20 billion per gigawatt. And the reason for that is because SpaceX is basically getting spot. You know, they're doing these deals that can be cancelled by either side within 90 days. So they're basically getting spot minus 90 whereas Coreweave is signing up for 5-year commitments. So it's you know contracted way in advance and they're kind of getting this contracted price and what's happened I think over the past year is that spot has moved way up >> way up >> because again of the scarcity right and the demand is so off the charts. So in any event, payback period is really critical and as long as the payback periods are really short, I don't see how these companies can lose or at least you'd have to be really incompetent to screw it up, right? This is an extremely important part and especially as data centers are potentially built out a little bit slower as we find harder access to memory, to power, to just the land in general. These sort of constraints are going to lead for spot prices to rise even faster. I don't think I have the clip of it, but Chimoth talked about how smart it was of Elon to give these deals for 90 days out, not to take advantage of Enthropic, but to make sure that they're protected because $50 million a megawatt might end up being the low end of the deal. Cororeweave is here stuck at $20 million and they could be stuck for multiple multiple years because some of these contracts were signed a long time ago. They essentially signed like five plus year deals with Microsoft, with Meta, with OpenAI, with Enthropic. And if spot prices keep rising, then you're essentially missing out on all of that premium and doesn't mean that you're going to go bankrupt taking that smaller amount, but it is going to affect your payback period. As we look to Nebius, this was a part of their most recent highlight. They said payback period on deals that were closed in Q2. So this is their Q2 deals. This is greater than $20 million is where they're sitting right now on Q2 deals. Payback period, one year and 10 months. A lot of this also has to do with the cost of their overall data center which is also much lower than their competitors because they build their own racks because they own their own land because a lot of those margins do not have to go to collocations and leases and all of these things because they own the full stack their payback period is potentially faster even if they are at the same level as cororeweave but that's not true. See, if we look at Q3 short-term capacity deals, they're already above 40. And Arcotti has gone on to say some of them have been above $50 million in their most recent fireside chat, which could lead to payback periods of under one year in length. This is an extremely important case is that Nebus has so much capacity for higher amounts of overall pricing closer to spot than Corewave. Cororeweave doesn't have a gigawatt that's free to be able to sell on spot prices. Nebus, I mean, they don't have a gigawatt either, but they have way more as a percentage of their overall mix that can be sold short. Way, way closer than what Core has cuz the bread and butter of what Nebius does are deals that are set between one year and 3 years in length, not the 5-year deals that you can get caught up in. I'm buying Nebus here. I'm extremely excited about the flexibility that this company has. There's going to be some new data centers that come online. We're looking at NScale potentially doing a down round. They just signed a fiveyear deal that investors got excited about about selling Vera Rubins to Enthropic. $45 billion deal for a company that's worth potentially around $20 billion. Sounds exciting until you find out that the per megawatt pricing was roughly 16 million. So, they're signing up these deals to make major headlines ahead of their IPOs that are going to turn out to burn them in the long term because those are five-year deals. If you believe that artificial intelligence is this long-term buildout that we're going to have constraints and new models that end up pushing these constraints even longer, Google is talking about signing long-term deals for memory all the way out to 2035. The only reason why they're doing that is to try to secure some sort of ceiling pricing on memory all the way out until 2035. That's how long they believe some of these constraints will last. If Nebus can have these fast turnaround periods, they can get capacity up without having to raise a lot of debt, high amounts, 50 60% of capex is self-funded via prepayment. And then they can turn around faster than the companies like Cororeweave. I like Cororeweave here. I'm a shareholder. I don't want to sell. It's just if there is a concern, if let's say, as Chamath said, the US tenure goes up to like 6%. Cororeweave is going to have a very very significant problem and likely Nebius will fall alongside that but it doesn't actually mean that their business is impacted nearly as much because they have the cash available to pay off all their debts. And then we go to the things like the flexibility of the way that they can make revenue. Coreweave doesn't have an asset light model. Nebus does. Nebius realized that their constraint a long time ago was capital. And so they actually created a new business line where they can go to other data center providers and say, "Hey, you have capital, you have GPUs, we have connections to Nvidia if you don't. We have the software if you don't. We have the ability to have customers because people are going to nebus.com to look at our GPU prices. We're on Palunteer. We're on Nebus' referred clouds. We're on all of these different sites. If you're a brand new company, you have none of that. And so we can actually set up a data center for you, get paid in cash without having to put any of our capital up just using our expertise. If Cororeweave can try to replicate those models, and I think you actually will see that because everyone looks at a business almost like it's a static company like what would this company look like if nothing changed from their end rather than what looking at them being essentially very active in trying to change this. So in the future, I think what the market's not pricing in in is is Nvidia coming in and making any sort of deal with Cororeweave is Cororee coming in and making any sort of change to their product, the way that it's priced, needing higher prepayment, any of these sort of deals going forward or or even as Shimoth and them said in their podcast, the ability for Cororeef still to potentially cancel one of their large deals, pay that penalty, but then have potentially like a gigawatt worth of a data center that could be used for selling that spot that could they potentially face a small penalty to see a huge upside and then what would happen to that stock later on. I think that the market is just generally over reacting so much to a short-term story without seeing the flexibility that these neoclouds have for having the actual land the chips and the customers that is very very hard to find anywhere else. So, I'm bullish, but I am focusing my new purchases on Nebus, not Cororeweave, but not selling any core weave either.