The $500 Billion AI Debt Machine & The Great CapEx Test

Hello and welcome back to the Market Maker podcast and Piers I was going through some of the analytics and looking at the performance of some of the recent episodes and and just as you might expect anything with AI in it seems to be outperforming and I don't want to be one who's like chasing clicks but if that's what the people want then who am I to deny them of that so >> Give the masses what they crave. >> [laughter] >> So in this episode we would normally do a bit of review of the week so what I've tried to do here from the research perspective for this this episode is blend together some of the news that's been happening this week with the dominant kind of narratives that of AI and on macro on inflation US CPI which came out yesterday from when we're recording this so it's kind of framed around this shift we spoken about a few times in recent episodes between who builds the best AI model to now who has the trillions of dollars needed to build the data centers to run them and that's distinctly tied to a headline of which we'll cover shortly. >> Yep. Absolutely and like just to look at the numbers hitting the tape this week so as a pure play AI cloud provider uh who you'll know the name of when we get to it but basically they've been showing a staggering 104.2 billion dollar revenue backlog. I know that's a bit of a mouthful but the order book what I don't think in the history of mankind there's ever been an order book that looks like this um anyway we'll get on to that the next up is then we'll Nvidia teaming up with six of Wall Street's heaviest hitters um to basically mobilize a new 500 billion dollar financing machine And then we've got global AI capital expenditures pacing towards one trillion dollars annually, that is um in terms of the if you like the hyperscaler uh spending splurge that seems to be continuing to accelerate. >> Yeah, so we'll we'll we'll try and go through this in the next kind of 30 40 minutes. Been loving the comments by the way. Recent videos has really take taken off. People sharing their ideas, >> Mhm. >> agreeing, criticizing. Actually, you know, if you do like the conversations in the show, do that, please, because it really help the the show perform on the different different platforms. So, we'd love to hear from you. So, really dissecting then the financial plumbing, we can call it, behind this AI supercycle. When it comes to that company you mentioned, that company is Coreweave, and Coreweave had an earnings blowout. I think their shares went up like almost 20% uh when I was watching Bloomberg yesterday. Brings up the question then, what are neoclouds, and why their revenue is doubling, and how they manage their massive debt loads. I mean, the numbers are quite staggering, actually, which I'm sure you're going to unpack in a moment. With Nvidia, their 500 billion Wall Street consortium, so the names involved in this, the ones that everyone likes to hear about, the Blackstones, the Apollos, and I guess trying to explain how they're transforming GPUs into a brand new financial asset class, uh which we were just talking offline about, which is which is really interesting. And I think a good point to note, particularly if you're a student going into some of these conversations, that application season coming up. And then the CAPEX, of course, ongoing tug-of-war. This massive hardware spending is triggering volatility in tech stocks and driving a bit of a rotation some of the broader market sectors. So, let's dive into this theme, CoreWeave's Q2 earnings report. Uh I I went on the about section of CoreWeave and I was like, "Okay, what's the easiest way to try and explain this business?" So, so here here here's the best uh way of doing it. And the reason why a lot of people look at it, it's become like this ultimate bellwether for AI hardware demand. And we'll talk about circular financing in a moment, but it's really important within that ecosystem. So, what is it? Specialized cloud computing provider that builds data centers packed with these Nvidia graphics processing units to rent high-performance compute power to AI developers and major tech companies. So, that's what it is. Top-line revenue surged 112% year-over-year to 2.58 billion. They also raised their full-year guidance range to around 12 and 1/2 to 13.2 billion. So, 112% year-over-year revenue. You can't top that, surely. You haven't you haven't got any numbers that can get better than that. >> Well, I can, cuz that wasn't the most impressive thing in the report. Um even though that in of itself is just outrageous. Um the most crazy crazy number was as I mentioned, this 104.2 billion um revenue backlog. So, I mean, what is that? So, firstly, right? So, they made 2.58 billion. That was our revenue. That's for the quarter that's behind us, okay? They've given guidance for the full year that we're in to say that on this full year that we're currently in, we might get to 13 billion, okay? So, that I think about those for some numbers. Even if you take the 13 billion and they hit that this year, 13 across a whole 12 months, their revenue backlog, that is basically book their order book. That is orders committed is 104 billion. Which >> What what's the legal um what's the legal tying for that? Are they Surely that could change though. That number could could flex or I guess they'd they'd probably have to pay a penalty, right? If you pull the order. >> Yeah, absolutely. Look, some the bears out there will talk of this circular financing or put in a easier way to understand, the house of cards. And whether all that that's going to come collapsing down. We're going to talk about that in a minute, all right? But let's just let's just bask in the glow of this unbelievable number. So look, that that order book um backlog, 104 billion, right? That's up just the backlog now has increased by 246% year-over-year. Okay, so this is all it's like in the last 12-months and well, it's been in the last 24-months, but at the exponentiality of the way this is ramping is just staggering, right? And by the way, then that being that, right? Right after the quarter ended, so not in these figures, they just landed another 25 billion dollar in net new customer commitments. 20 another 20 another 25. So that their backlogs now basically 130 billion, which is 10 times their forward-looking revenue for the for the year that we're in now. Um I mean, words just fail me at this point. But look, what you know, so here we're talking about the long-term contracts from the big guns, all right? So it's your Metas, your OpenAIs, your Microsofts, okay? And obviously you can split and you might say, "Well, okay, that sounds all amazing, but when might this revenue you know, when when are they committing to this revenue?" And actually, so 40% of that 104 billion is committed in the next 24 months. You then got another 39% that's committed 25 to 48 months, and then 21% the remainder is committed for 48 months plus. So, we're we're obviously talking about a very long, you know, runway um of commitments here, which is awesome for this business. I mean, look, you know, we're talking about valuations and so on. Like, your order but your forward-looking order book is is everything, right? Cuz if you're buying shares in this company now, well, of course, you're buying future growth. And I mean, they've got committed growth that's just in insane. >> Maybe we could take a step back for a second, and you know, this essentially is cloud, but when people think of that, a lot of people think traditional cloud providers, so Azure from Microsoft or Amazon AWS, Google Cloud. Yeah. So, maybe we could just explain for a moment, what is a What is this? What is a neo cloud rather than traditional cloud providers? >> Okay. So, your traditional hyperscalers, so they're, you know, they those big the big three as you've mentioned, right? So, Amazon's AWS cloud, you've got Microsoft's Azure, and you've got Google Cloud. So, they're built I mean, they've been building those platforms for decades, and they are general-purpose computing. All right, we're talking web servers, whatever, enterprise databases, app backends. I mean, we are a big user of AWS, like Amplify um ourselves. That's our cloud provider that we use, right? But, it's general-purpose. The Neo Cloud, well, we're talking specialization here. These are specialized providers built ground up exclusively for artificial intelligence. They They are optimized for AI, right? So, Coreweave uses things like liquid-cooled racks, custom high-speed inter- interconnect networks. Um they use bare-metal GPU deployment. >> So, outside of computing, you can kind of think of it as uh Volkswagen and Tesla both make cars, but Volkswagen will try and use the traditional uh manufacturing plant and tweak it a little bit to make the electric vehicles, whereas Tesla's built ground up. Is that a similar kind of comparable top level? >> You might say, yeah, Volkswagen's mass-market, trying to build cars for everyone in the entire system. And maybe Ferrari, who're trying to build a car that can race around a track the fastest. And so, it's that specialization that has attracted all the big guns, because they're they're very specific spend from these hyperscalers, their very very very specific spend is for this AI. >> Is that why people buy Ferraris? >> It's apparently. >> [laughter] >> I don't think many people buy Ferraris to race it round tracks. >> No, but from an engineering point of view, >> Of course. I know you're such an engineering purist. >> And that's what attracts people to buy them, cuz they're buying into that specialization and the and the thought of it, yeah. Anyway. >> Okay. So, then, let's talk about the flip side of the the hypergrowth then, the cost structure of this. Cuz giving you a couple more numbers from their earnings release that we had this week. So, they reported an adjusted net loss of $567 million for the quarter. Their GAAP net loss was $1.14 per share. Check this figure out, though. Cuz everyone goes a bit CapEx crazy when they're talking about the fragility of the sort of stock market rally we've had based on the spend. So, the capital expenditures hit 9.4 billion in Q2 alone. So, the company's spending 9.4 billion in the quarter. They're just net loss of half a billion on the CapEx spend. What I thought I'd do is look for a comparable, and I know there's a bit of a defensive nature of Apple, and they're not the biggest spenders on CapEx, but I just thought from a magnitude of company, given that not many people even know CoreWeave exists in the kind of public domain. So, that 9.4 billion, Apple's CapEx in Q2 2.5 billion. I mean, that's just insane. >> Yeah. So, look, whilst those revenue figures, the revenue growth awesome, the revenue backlog, i.e. the order book looking forward, is disgustingly unbelievable in a positive way. They are racing to try and keep up with demand to the point where they're having to spend and spend and spend to try and keep up, but the spending rate is right now obviously much greater. Like that 9.4 billion spend in Q2, their revenue was only 2.5. So, they're spending three and a half, whatever that multiple is, three and a half, nearly four times their revenues, right? So, obviously they're making a loss, but the point is that they can't capture that 104 billion order book. They can't capture the money that's been committed without investing in their growth now, right? So, it's about rapid expansion. So, they're taking tens of billions in debt to buy Nvidia chips and build out mega facilities. And in fact, you know, quarterly interest expense alone surged to $640 million. So, just to service the interest on the debt they're having to take on to try and build out capacity to try and capture this revenue commitment in the future, right? So, a couple of more stats then. Power capacity um So, CoreWeave's active power footprint at the moment is 1.5 gigawatts, all right? Well, that's what they're monetizing. Um they've got contracted power, so what they've committed to to build out is to reach 4.2 gigawatts at the moment. So, they're at 1.5 gig and they've got a pathway to 4.2, right? Just to put that in perspective, 1 gigawatt powers about 750,000 homes, right? So, if you think about CoreWeave's contracted power pipeline, it basically matches the electricity consumption of a city of 3 million households. So, that's the kind of power roadmap that they're trying to fund. And right now, the revenue's not quite caught up. So, they're still losing money. >> So, if you want to know why you're sat in England right now and it's 38° outside, uh you could probably pin some of the blame on these these guys uh contributing to that. But, so, the the demand pipeline definitely is legit. I mean, looking at those percentage splits and it seems very weighted to the near term, which is a good thing. So, 104 billion backlog. Is this like you've you often talk about this when you and I talk about more traditional markets, um, this priced for perfection idea? So, is this an execution risk scenario? >> Um, for sure, it's an execution risk, but I I would say more than anything, it's a it's a funding risk. Um, and we'll come on to it's the house of cards risk, or maybe it's a solid foundation. It's kind of all intertwined, and people use this word circular financing. But I I would say it's a funding risk above everything else. >> Okay, well, on that point then, look, that ties into that other big story we've had of the week. Uh, and that was Nvidia announced strategic partnerships with six of the world's premier private capital institutions. Those being Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR. They've got all of their hands in on this. >> It's It's It's the big guns. They're basically What's the mandate? Basically, they're mobilizing over $500 billion of dollars of third-party capital over time to underwrite this this whole AI infrastructure buildout. >> Yeah, and thinking about the sheer scale of this, historically, chip makers sold hardware to companies that funded purchases out of corporate cash flow or standard corporate bonds. But to quote the main man in the center of all of this action, uh, Jensen Huang of Nvidia, the CEO, he said, "In AI, compute is revenue." In his words, "We're helping create a new class of productive investable infrastructure, i.e., these AI factories. So, I thought that's quite a good then segue from what you said as to trying to then unpack this idea of the business model and this advent of a new asset class in terms of how these big institutional pockets of money >> Yeah. >> Blackstones and Apollos and KKR's, how are they engineering this? These are clever people. There's definite risk here. So, how have they done it in a way to mitigate that risk to maximize the opportunity? >> Well, I mean, let's try and break it down. You know, how is this financial engineering actually working? So, like historically, tech I mean, if you think back pre-AI, tech hardware used to depreciate very quickly. That's just because, you know, the new version is going to be I next year, right? So, in terms of in terms of what value can you extract from an asset? You know, what's its lifetime worth to you as a business? And if you buy it this year, and if you're going to buy something else and replace it next year, well, obviously you've got a super short one-year time frame. And so, you can't really, you know, you you have to write off all that cost in this year. Number one, and you can't use that as a tangible asset to borrow against because next year it won't have any value, right? That's how it used to work. But, the game in town now is basically Nvidia's CUDA, right? That's their compute unified device architecture. Um their their kind of software ecosystem gives these GPUs a much longer flexible software life across different models and operators. Okay, so there's this thing called the value cascade concept that they've come up with to try and give this a a kind of basically a new category, but basically the core argument is that the GPU stays cutting edge for 6 years. So even though I think about Nvidia and we talked about it on previous episodes, their engine engineering kind of cycle is 12 months. So each 12 months they're trying to bring out the latest chip and the the latest chip is always, you know, several X factor better than the previous one. So isn't it the same? If you bought a chip last year, well, it's worth nothing this year. Well, the answer is no. So they're basically in this in this CUDA system, these older chips stay valuable. So chips basically move down a hierarchy of workloads as it ages and at each stage it still generates revenue. So years three to four we call this supporting the secondary life, real-time inference, right? And then you got years five to six, there you support tertiary life. It's more like batch inference and analytics workloads. I mean, what does that mean? You know, when you're on Claude at the moment or chat GPT or whatever, you can choose which of its systems to use, right? Depending on the job you want it to have. Um and basically think about it like that. I don't want to use the Ferrari every single day, you know, just to drive to Sainsbury's. >> [laughter] >> Okay? So I want to use my my Volkswagen Golf to go to Sainsbury's. So even though the Ferrari's much much better still going to use the Golf and it's going to give me value. Basically, these companies can continue to generate revenue >> Mhm. >> for five, six, I don't know. People are talking 10 years, right? And obviously we don't know yet about the 10, but do we know about the five or six? What I will say is Michael Burry, the famous dude from The Big Short, who nailed the financial crisis and made an absolute fortune. He's closed his hedge fund. Closed it. Um, I think it was at the end of last year. Closed it. Why? Performance is shocking. Why? Cuz he was taking the other side of this argument. This This precise argument. He was saying, "These chips only have a 12-month lifetime. These companies are buying them, and then they're spreading and capitalizing that cost and spreading the cost over 6-7 years, and this artificially inflates their profit." And so you're trading these stocks at valuations on the idea they're making these profits. He was saying they're false. However, Michael Burry's wrong. And he's now out of business cuz we're already seeing this this cuz we were talking, when was it, last week? OpenAI launched its kind of consumer-facing product in 2022. We're almost four years in now, and these chips are still producing revenue. So it looks like the companies are right and Michael Burry is wrong. >> I'm reading a book at the moment called Superforecasting, >> Mhm. >> which is this research theory about trying to determine really who has an ability to forecast things in the future. And why this is particularly relevant in a financial context is like trying to identify, you know, who can predict what's going to happen in a certain event. And so therefore you have to think about multiple different layers in order to arrive at that decision. One of the qualifying factors early in the chapters is about when you have these people, particularly in finance, talking heads, people who go on Bloomberg, sell-side institutions, Michael Burry. >> Yeah. >> Never, ever will explicitly put a timeline on the forecast, and that null and voids the forecast in which they're making. Because if you're going to say this is an AI bubble, that's a house of cards, but you don't define the timeline, well, then basically it's an open-ended statement that has zero commitment and zero predictability power behind it. >> So, if it does fail, you can expect Michael Burry to be like, "Right, Netflix, where's my you know, the big short part two? Because I want to get paid out now, cuz I was right about everything, like I'm right about before." >> Well, of course, that's what happened in the financial crisis. He he was so early on that trade, the trade being that the housing market's going to collapse. He was so early, he was almost too early. He was like starting to put these trades on 2005, 2006. We got into 2007, and it was still going against him and against him, and he was getting margin calls from the banks who he'd done over-the-counter options and kind of derivatives deals with to kind of position himself for this strategy, and he almost killed him. Before then, just just in time, he was right, and fine, made a fortune. So, maybe there's maybe there's an argument there that he's right, but not for 5 years. >> Yeah, the other one is just not right. >> The other one was Ray Dalio. There's a book called The Fund, if you read that one, and he was back in the '80s talking about all this negative stuff, and very doom and gloom. But, yeah, I mean, he he ends up being right, but like a decade or two off the off the uh off the needle. But, okay, so look, summarizing then what you've said, so because these GPU clusters generate predictable, long-duration cash flows via these you know, when you're dealing with these hyperscalers, these are long-term contracts off of very established, deep-capital-pocketed companies who are spending big time. So, we even when you think about the others like Anthropic, but then the traditional ones like Meta. So, Wall Street is treating them right to say like real estate or aircraft fleets, or or probably the more correct definition, infrastructure assets. Is that how you'd see it? >> Yeah, absolutely. Yeah, it's that classic you know, if you want to lend someone money well, then as the lender, you know, you're obviously going to assess the credit risk and you're going to assess well, what's the opportunity for the person I'm borrowing, or sorry, lending to here? And is there any tangible asset that I can under or they can underwrite a loan with? So, it is if if it goes wrong and the wheels come off, what am I left with as the lender of value that I can extract at least some of my money and get it back. And yeah, so, you know, a very old-school way of financing like a manufacturing business was you would lease your you would basically sell your machinery you know, on your production line and and basically lease it back. Or or you would borrow money with the machinery being the collateral to underwrite the loan. You know, it's these physical, tangible things from which you are able to generate value and generate the revenue that you do, right? And so, that's it. These chips or these data centers, if you want to kind of just zoom out a bit, these are the infrastructure plays of our time. >> Hm. It's interesting you said that, cuz I know your your brother's in a bit of this space with the manufacturing plants. >> Yeah. >> think of it like that. So, actually, this isn't a new model. It's a different type of machinery, I eat the technology, chips, but the model is a traditional one. >> This this model is, oh my god, it's decades decades decades decades old. >> Mhm. >> Yeah. >> Okay. Well, look, that leads us on then to the the somewhat elephant in the room, which is that word, the circular financing. The when you go on YouTube, it's like that's that's the buzzword because everyone wants the house of cards because we're human. We love to see the world burn. It seems to be the way that people like to look at these things. So, yeah, let let's dive into a little bit about when what happened to Nvidia when some of this news was breaking with this funding because you would think, hang about, they're all they're they're forming a consortium, they're going to they're going to give you a half a trillion dollars in cash to fuel the dream, and then your shares fall. Like >> Yeah. >> how does that work? >> Well, so, on the news that broke earlier this week about this $500 billion debt debt deal that Nvidia have spun up, their stock their share price dropped 2.9%. And like for them, that's a lot when you kind of spin it into market cap terms, that's 60 billion market cap just just vanishes, right? Just in that 2.9% drop. And this is because you know, critics point to the circular nature of these deals, right? So, Nvidia invests in or helps structure debt for its clients, the Neo clients, CoreWeave, for example, right? Those clients, they then use that borrowed cash to buy Nvidia GPUs. Nvidia records that as revenue. That's your circular bit, right? Let let's kind of let's say Go on. >> Hold on. Given what you also previously said, are they not also like a seed investor in open AI and therefore telling open AI you should have multiple series of models that users can deploy so that we can extend the lifetime value of the chip clusters. >> So, this is it, right? This is the house of cards. So, basically Nvidia they're using their balance sheet strength. Obviously, monster balance sheet they've got cash flow that's just unbelievable, okay? Massively strong balance sheet and they're using that and their reputation to back its key customers. So, how are they doing this? Like CoreWeave, they'll they'll go and buy an equity stake, right? And then in return, they're going to get purchase commitments from CoreWeave. And basically, from the lenders' point of view, you know, if you're Apollo, why would I lend into this this circular machine? It's because basically Nvidia is the backstop guarantor for the loan, right? It's not a straightforward loan, this. You've got the payer of last resort is Nvidia, the biggest company on the planet who's got hundreds of billions of cash on their balance sheet, right? So, these arrangements they typically come with strings, though. So, as we've alluded to, CoreWeave commits to buying and deploying Nvidia's latest chips. So, when you think about CoreWeave's 104 billion, you know, revenue backlog, well, well, you know, some of that is from obviously Nvidia and this loan, but anyway, we'll come back to that. So, the actual debt financing um that these build-outs comes mainly from the private credit and the kind of banks and the insurers, right? So, Nvidia's backing is part of what makes that debt investment grade. So, investment grade is important here cuz CoreWeave have issued their own debt. They issued an 8.5 bits. This is separate to the Nvidia 500 bill thing, right? They issued a corporate bond for 8.5 billion. Um and it was rated A3 by Moody's. This is March of this year. Um and that is the first GPU-backed loan to reach investment grade status. And the re- and the only reason it's investment grade. So, investment grade status just means it's cheaper for CoreWeave to borrow cuz the credit risk is lower. If you like, the risk to the lender is lower, and it's all lower and a lower interest rate as a result because Nvidia and their role in that debt stack is basically It's so it's less lender and it's more the entity who's backstop makes the debt investment grade in the first place. So, because Nvidia are propping this all up, it makes it cheaper for their customers to borrow to then buy Nvidia chips. That's your That's your kind of circular nature, right? But to kind of finish this off, as long as CoreWeave can build that compute, which is dependent on hyperscalers keeping their commitments to buy that compute, then fine, revenues flow and the debt gets serviced, okay? The moment that Basically, the moment the hyperscalers stop spending or they're not They're not going to stop. Currently, there's an exponential acceleration in spending. This The moment we get evidence of a deceleration in spe- It only needs to be a deceleration. Worse would be flattening. So, we're going to carry on spending, but only at the same rate. Even worse would be we're going to start to reduce our spending. Right, you've got different grades of problem here. Even worse, I'm in the absolute Armageddon, is we're going to stop spending, right? Which isn't really going to happen, but but just to finish, if they stop spending, then obviously, there's there's multiple Nvidia exposures here. There's revenue, like Nvidia's revenue drops, their equity stakes in all of the the pies they've got their fingers in, well, that devalues and and basically then the backstop obligations become mount up and become maybe unserviceable and and you get hit simultaneously, which is then that systemic concentration risk that analysts are flagging. >> Yeah, and and >> Just on that point of the concentration risk, uh I read a a research paper from Columbia University, and I know it's just a neat way of just summarizing that in three categories. >> Mhm. >> So, one that you've just been talking about is Nvidia's near um mon- monopolistic position in AI chips compounds the concentration risk. >> Yeah. >> I mean, it blows my mind that the authorities can let this happen. I mean, it's just kind of cops and robbers, I guess. When when there's money to be made, you can move as fast as you can. Uh the reg- regulators are too busy trying to work out who's who's got power rather than sort out this sort of stuff. But the entire ecosystem's stability depending on one company's financial health, strategic decisions. So, the monopoly that Nvidia have, they're so they're like the the center of the universe. They're the sun, if you like, in this instance. >> Yeah. >> Then you've got the geographic in infrastructure concentration. So, you remember several months ago, this is the additional layered risk. If you remember the word Stargate, that was the one you remember when all of the all of the AI bros were with Trump in the Oval Office and they were like, "Right, yeah." And then there was the SoftBank was there as well, Oracle. >> Mhm. >> And they were talking about >> SoftBank man. >> Oh, yeah. They're all there. And they were talking about a multiple hundred billions in cumulative spending across five US sites. So, that in itself creates a concentrated exposure to specific data center locations and power infrastructure. I mean, such as everything diversification is key and this is not that. And then you've got the customer concentration, which you've kind of alluded to. Amplifying, compounding the systemic nature of the risk. So, your Microsoft, Alphabet, your Amazon, your Meta's. So, >> Yeah. And maybe maybe like Nvidia's the sun, your hyperscalers, that's the fuel that's powering that sun, right? As long as Microsoft and Alphabet and Amazon and Meta carry on spending, shoveling in the fuel, then we're fine, okay? There is another risk, which is more of a regulatory risk for the US, I would say, because you got those massive commitments. Oh, we're going to build data centers here and there and whatever and all these states, there's a real regulatory lots of regulatory hurdles slowing everything down. So, actually in the end, so you've got a you've got a compute capacity problem. You just can't build the stuff quick enough partially because the regulators get in the way and block it. Then you've got a power problem, which is are there enough electrons in the world to to actually power all of this kind of planned build out. And that that's also back to regulatory stuff in the US, you know, trying to build more nuclear reactors. I mean, basically takes 10 to 15 years to from start to finish to get a to spin up a nuclear reactor, right? I think China do it in 12 months. So, I think from a from a geopolitical race perspective, yeah, the US they want to be careful they don't score an own goal here by regulation regulating their way, you know, out of the front position that they're currently in. >> Well, let let let's step back a bit and let's tie this into some of the macro context. Firstly, of how does this fit then within the investment thesis or or strategy, so to speak? >> So, the fuel that they're shoveling into the sun. So, AI hyperscalers, their CapEx, this is an estimate for 2026, so a 12-month period, they'll spend 730 to 800 billion. That's just the four hypers or the four or so biggest companies, right? If you take everything, it's thought that we're going to break a trillion dollars this year, right? Um so, like if you think about Goldman's and JP, so their projected global AI, so if you if you're trying to forecast this out, as you said, it's very difficult. Um but Morgan Stanley estimates that the total cumulative hyperscaler spending will cross 3.5 trillion over the next 3 years. 3.5 trillion. So, if Morgan Stanley are right, if these hyperscalers carry on stepping up their spend, which they will only do if they can prove a return on investment for this spend. Then this this engine, this circular financing engine, will carry on firing like super hot. But if at any point there's not an ROI sort of evidence, then this is where your kind of wheels come off this circular financing machine. >> In order to sort of hedge yourself then in this scenario, is this where Stephen and I were talking about this a few episodes ago? So this is where you what? Diversify by moving into the other parts of the supply chain, namely as you were just talking about energy. >> Yeah. >> Is the ingredient energy and utilities and data centers requiring power, driving demand for nuclear, nat gas, grid infrastructure, then you've got the industrial material orientated category, so electrical transformers, cooling systems, construction, so forth. Interestingly I I was with my friend from uni at the weekend and he's a surveyor and he was saying it's it's crazy where Google has these like secret entities that it goes around just buying up swabs of like real estate so they don't get basically charged a ton of money and they just trying to find any location in proximity to natural water >> Right. >> to cool down some of these centers and it's going crazy. And then there's the value and defensive, so high dividend sectors providing that stability while the mega cap tech digests its capex investments. So maybe we could to kind of finish this section off then, um one other thing that we did see was Anthropic. >> Mhm. >> And the FT just broke a few hours before we were recording this. The valuation of two trillion. I thought we were at one. What have I missed? Have I I've blinked, I've missed it. Revenue growth. That's what you've missed. We're a little bit in the dark, right? If you go back to the start of the year, we know that at the end of last year they were on a 10 um a 10 billion dollar revenue run rate. So in that month of December, if you just took what they did in that month, multiplied it by 12, then that's 10 billion. They were at 1 billion at the end of 2024. So that's a 10x growth, right? Now at the start of this year, everyone was going, they can 10x again. So 1 billion to 10 billion to 100 billion, they can 10x again here, right? And everyone's going, "No way, not possible, not possible." In May, which is the last official revenue kind of news we got, they're at they're already at 80 billion. Now they're in a blackout phase because they've filed for their IPO, they go dark. So we actually haven't heard anything much from them. But you know, people who were in the know, they're talking about they're talking about by the end of this year, forget 100, they're 110, 120 billion, which would be a 12x. So the the the growth rate is accelerating. And then you're talking about people saying, "Well, it's going to 10x again the year after." So they'll do that people are saying they'll do a trillion dollars of revenue, or the run rate at least will be a trillion dollars by the end of 2027. Bare in mind the biggest companies on the planet do what? 400 billion? Like the hyperscalers, they do 400, 500 billion revenue a year. We're talking a company that didn't exist 5 years ago. Didn't exist. Doing it double that trillion dollars by the end of 2027. Look, that's that's the that's the bull case, right? Um So, the point about valuation then, well, cuz that is hard, right? How do you compare it to you know, what's what's the kind of market comp here? And you could look to people like Palantir, for example, or Nebius, right? These other they're obviously much smaller, but they're trading at 55 times revenue in the open market now, right? So, if Anthropic's growing basically was like 1,000% a year, um then you would think, you know, even at the low end, they'd be getting 30 times revenue. Like low end, 30 times revenue would be 3 trillion, not two. Some are saying 2 trillion is an absolute steal. So, >> Slight caveat though, when I was reading this report in the FT, the person quoted in saying these numbers, you're right, let's let's say a blackout period, not allowed to speak, just so happens to be an investor in the company. They're you know, talking their book up. I mean, I don't think they're wrong. Don't get me wrong, but I think it's a Yeah, it's so funny. This is like the marketing uh the marketing person just juicing just you know, we haven't heard from Anthropic in a while, it feels like. It's been a couple of weeks. Hang about. Back front and center, please. All this CoreWeave CoreWeave, go away. Come on, this is Anthropic story. This is October 3 trillion. Here we go. So, yeah, interesting. Um so, one thing then to wrap up this bit, and we'll talk to close on US CPI and tie this to the macro. Cuz as you said, the two major narratives for markets definitely are this AI infrastructure and where it might go and the spend around it. Given the magnitude of the companies involved in it lifting the stock market to these record highs, but also the macro climate in regards to inflation and interest rates. Before we talk US CPI, is Wall Street's $500 billion private credit pool a masterstroke then in scaling this global compute infrastructure, or is it too much leverage before the RRI has even begun to be proven? I don't want your answer, Piers. I want anyone listening. Cuz this is quite divisive. What do you think? Yeah, I'm I'm I've got a feeling everyone's going to be slightly bearish here, but I'd love to see the thesis behind the bullish or bearishness that people have. Um all right, let's let's say Let Let's talk about the US CPI report then cuz lo and behold, stocks did rise on the back of this, and it was uh a bit of a surprise. There were the one that Bloomberg and the rest of the financial media were latching onto, rightly or wrongly, was the core CPI. That came in at 0.2% month-over-month, 2.5% year-on-year, the slowest annual pace since March of 20 21. So, how do you How does that fit then to where the market is at with its thinking with the new Fed chair, Walsh, uh and interest rate expectations going out for 2026? >> Yeah, so it's a good report. Um 2.5% You're right, it's the lowest since March 2021. However, it matches January and February 2026 numbers. So, we have been here. >> We don't say that. We don't say that. Doesn't sound as sexy to me. You don't Why did you have to go and mention that? So, look, we had I I I I think obviously it's tied all to Straits of Hormuz and what goes on and energy prices and how that filters down through the system. But look, we're back to pre Straits of Hormuz. Right? That's the point here. January and February before things kicked off in the Gulf, we're at 2.5 and we were expecting the downward trend of the previous couple years to continue and that back then, remember, we were expecting couple of rate cuts this year. If that trend were to continue, it obviously didn't. We boxed higher. So, we went 2.6 March, we went 2.8 April, we went 2.9 May. But now it's toppled back down. June 2.6, July 2.5. So, it's like, all right. That inflation concern maybe's over. So, just means Wash doesn't have to hike. I think you forget, oh, we've got one more So, this was inflation for July, right? We will actually have the August inflation data announced before the next Fed meeting. But based on this, >> Yeah, the the Fed Watch tool, which allows us to look at short-term interest rate futures. So, it basically gives us an implied probability of how the markets are expecting what for when. And so, that now sees a September rate hike odds down to 38%. >> I think that's too high. I'd be selling that. Myself. I'm a seller. Um there are there there is the one thing in the mix in the basket that's causing concern. Computer software and accessories. That part of the basket's up 21% year-on-year. Why? Because of everything we just spoken about, right? The demand or the lack the demand supply imbalance for everything AI has meant the cost of these semiconductors, chips for example, has just gone through the roof. So obviously that's feeding through into inflation, but it's not enough. That component's currently not big enough to or I mean, you could argue the basket isn't set up appropriately for the modern days expenditure. So I don't know, there's two sides to that argument, right? But for now, the way the inflation basket is measured at the moment that computer software and accessories classification isn't enough to kind of take the whole inflation number back higher, right? Um so there is one thing in the basket, just as a quick aside biggest down biggest faller in the basket lettuce the price of lettuce dropped 16.4%. There you go. >> I can I can stack my burger now with tomato and lettuce. >> There's a disease there's a parasite or a disease in the Midwest in the US in lettuce. People aren't buying it cuz they don't want to get stomach bug. So you've got now over supply cuz the demand has dropped. So you've got lettuce down 16.4%. So there's a supply and demand case study if you want to go and grab hold of that. >> And then one thing on the timeline also to be aware of next month is Jackson Hole. >> Mhm. >> The Jackson Hole symposium is one of those platforms where the Fed chair gets to give a keynote speech. That's happening at the end of August, I think 27th through to 29th. He's normally like the main headline act, like the Glastonbury headline. And it's definitely outside of the fixed set schedule of Fed events that happen eight times a year. That's the other one where if there is a little signal to be issued to the market about what's going to happen in September, the likelihood is he might give some some some things or not given what he's like. I don't know. Yeah, don't hold your breath. He'll come up and go, "Hi, nice mountain scenery. Thanks very much." >> [laughter] >> He's going to say nothing. I think that's Yeah, I I know historically that I'm not I'm not I'm not not a thing anymore. I don't [clears throat] think Well, I think it's definitely going to become less and less a thing whether it's he's only just got into the seat, so maybe it's still a thing, but I doubt he would use that platform to signal forward guidance given that everything he's about is not giving forward guidance, right? So, >> It's good night life though out in Wyoming. He might have a few beers the night before and then the genuine Kevin comes out to play. And then we can have some trading activity there. We can trade him. >> [laughter] >> Uh okay. So, that concludes the episode. As I said, love to hear your thoughts on what you think about a lot of the particularly the AI investing side of things. Uh let us know, but thanks very much, Piers, and thanks very much for listening. Thanks a lot.