Loading summary
Transcript1 lines
- [00:01]
A
Welcome to Money for the Rest of Us. This is a personal finance show on money. How it works, how to invest it and how to live without worrying about it. I'm your host, David Stein. Today is episode 561. It's titled how to Navigate the AI Debt Bubble. I recently got an email from a member of Money for the Rest of Us plus and he was worried to what extent is all this debt that is being used to finance AI infrastructure, new data centers, all the backgrounds, backbone for this AI bubble we're in as hyperscalers, build out all this infrastructure, all this compute for frontier models, for non frontier models as we are in this grand experiment to implement AI as individuals and businesses use AI, figure out how this new general technology can help us in our daily lives, can help businesses boost productivity to innovate more. Now I've talked a lot about AI and the AI bubble and the very high valuations of AI related companies. And as I record this few episodes ago, we did a episode on SpaceX, its IPO and the exorbitant valuation of that IPO. Now it's selling for 20% less than the IPO price. And other AI connected companies have sold off. We're getting near almost a 10% type correction. Now this is a really volatile area, but this member was concerned, well, where's all this debt going that is being used to finance these data centers? Because something has happened in the past 18 months. Previously these companies like Meta and Amazon and others generated enough free cash flow to essentially build out infrastructure, data center infrastructures. But now that the level of capital expenditures is so high that they're having to borrow and a lot of these companies are investment grade and they're out borrowing trillions of dollars essentially, or it will be trillions of dollars to fund this grand AI experiment. And there's some mismatches in terms of the longevity of the chips that are used to run the compute in the data centers versus the length of the maturity debt. So we're going to take a look at it and figure out should we be worried, should we be concerned about the sheer amount of borrowing? Is it like the great financial crisis with all the borrowing related to the housing bubble that ended up being in pockets of finance on the balance sheet of many institutions that we wouldn't have thought had exposure, Is that what we're seeing here? We'll take a look, but let's first get a sense of the scale of the borrowing that is occurring in this AI debt bubble. Now the overall CapEx capital expenditures estimates by JP Morgan, McKinsey and Goldman Sachs and they're just estimates anywhere from 6 trillion to 7 trillion dollars invested in data centers and infrastructure power grids to power these data centers. Now I mentioned previously it was funded out of free cash flow, but they estimate, and this is JP Morgan specifically, 75% of this CapEx will be funded with debt. That's about $4 trillion of new debt outstanding linked to the AI infrastructure buildout. And that's huge. When we think about the overall size of the corporate bond market, both investment grade and non investment grade, it's around $11.7 trillion. So roughly a third really of new debt issuance as a percent of the overall bond market. The US bond market will be AI debt. That's meaningful. And one of the big questions is well how could this impact interest rates and spreads the incremental yield that investor demands to hold corporate debt? And we'll take a look at what spreads are in a few minutes and the potential risk there. But that gives you the idea of the size of the level of borrowing that's going to occur. Now let's step back and see kind of where this fits. In other cycles that we have. There's something called the capital cycle and this is from financial historian Edward Chancellor's. Others have sort of discussed this, but there's a new technology and it, it sparks excitement and potentially high returns, but uncertain returns. We saw this in the Internet bubble in railroads. Now we have it in AI, the potential for very high returns and if the technology pays out. So there's massive investment and new infrastructure, there's competition who can be the first mover, who can build the best frontier model, who will capture more of the spoils. And so there's a huge amount of capital that's raised partly due to this competition. And often you get more capital raised, more investment than what's needed. And so then you get to a part of the cycle where well, the market realizes there's too much. We saw this during the Internet bubble where there was so much capital invested in Internet infrastructure, laying fibers, lines and other cables to carry the Internet and way more than was needed. And so we saw spreads, especially for non investment grade bonds, investment grade bonds. The spreads ballooned and those bonds sold off and there were a lot of losses and there were defaults and bankruptcies and a lot of companies went out of business but the infrastructure was still there and it eventually the demand grew into the supply of infrastructure. We're likely to see the same thing with, with this AI build out with this we're all benefiting from it. There'll be a technology spillover and it's going to benefit all companies, both small and large, us, non US households, businesses. We're all benefiting from this experiment. And yeah, it's been raised with equity capital and debt capital and there will likely be an oversupply of that and a retrenchment. A lot of this borrowing won't, won't pay off and we want to make sure that we have minimized our exposure to it and that's why we're doing this episode. In preparation for this episode, I read a report by BIS and they looked at sort of other prior cycles and the AI boom in terms of the multiple of capital being spent in this arms race. It, it's over four times what was invested in AI kind of prior to it. And so the boom, the capex has been larger than the canal mania in the 19th century, which is another boom and bust building out infrastructure. It's larger than the railway mania, larger than the.com and the Internet in the roaring twenties. So the scale of this is definitely larger. As a percent of the overall capital, the returns aren't likely to justify what's being invested. Particularly as we see China and other what are called open weight models where someone can take the model and customize it. It's not a closed model and there's more and more of them because the advances of these frontier models are slowing down a little bit. They're hitting some ceilings now, they're still getting better. But the non frontier models, the open weight models are starting to catch up and so there's more competition. And given the level of CapEx, the level of borrowing, that's partly one reason we're seeing some of the DEB of these AI related investments are selling off a little bit. In terms of credit default swaps, we're seeing spreads widen a little bit for AI related debt. And it's because some of these other models are catching up and there's a fear that again it's too much capital expenditures that justify the potential opportunity. But we have five years of this in terms of kind of continue to build out because many people have never even used AI. So let's look at kind of the six financing channels for this AI infrastructure build out. The largest by far is investment grade bonds. They are found in, in ETFs and this is kind of what this member was getting at. How much exposure do I have now? These are, these are public bonds issued by Hyperscalers and Nvidia and Amazon Meta and others, Alphabet. And they're investment grade and they're borrowing. And so you're seeing it's about 14% of the investment grade bond market right now. So that's one aspect that's the largest, but another is just bank, bank lending. Banks are lending to, in the infrastructure bill. I mean that's what banks do, they lend. Now they've started to try to offload some of that risk and we'll take a look at that. But they're, they're a major source, about $170 billion for example, in 2025. Now there's a lot of private credit lending, private funds, business development companies, insurers are also participating. And in fact the insurance regulators are starting to worry a little bit about how much exposure insurance companies have. Now there's also off balance sheet lending with special purpose vehicles. So you have institutional investors partnering with these data center builders and sometimes there's a hyperscaler in the background, but it's, it's off balance sheet debt. And this is the aspect, there's just not as much transparency. Now a smaller portion is going into bonds, asset backed securities, commercial mortgage backed securities. So traditional bond channels that trade in, in the public market, but they could have some exposure. But it's been very, very small in terms of the public market. Non investment grade bonds has also been, been very small because at least at this point in terms of traditional high yield bonds, most of it at this point is going to investment grade bonds. You do have banks financing it, but a lot of it's private and a lot more of it. We're going to see sort of this off balance sheet loans. And so that's something we're going to want to monitor. All these different structures. Now a year ago we really didn't even have any sense of the numbers. You just started to see the first off balance sheet spv. So there's still a lack of transparency, which is one of the reasons there's some concerns. But the biggest issue right now or the biggest area has been investment grade bonds. But you got private bonds and the banks involved. And then we're going to look at sort of how they're trying to sort of shift around the risk, the various tools that they use to do that. One of the criticisms has been circular finance. And so you have the chip manufacturers like Nvidia, they're lending to these, these AI labs are lending to Neo Clouds, which essentially are our data centers that are just built to, to hold Nvidia chips and the other equipment for AI models and then they're leased to AI providers. Because one of the things we're seeing right now, and we discussed this in our, our strategy report last month, was the fact that this demand for compute is significantly outpacing supply. So we have this capex build out. But there's true demand here. This is not speculative demand. The concern is will the ultimate demand meet the supply as more supply comes online. Right now demand is greater than supply. We have these deals where Nvidia is providing financing to others to buy the chips to build out the infrastructure. Now part of that was there wasn't as many other vehicles to do that, but it is. Well there was a report this week on Bloomberg like another potential $750 billion for Nvidia backing this debt. And so you're seeing the credit default swaps. So basically a measure of the spread or what investors are demanding to protect against default on Nvidia's debt and that that's widened to, to it's a five year high right now because of the kind of this circular finance deals. Now one of the things that makes investors a little nervous is how these risks are getting passed on. So a bank might make a loan to the data center or, or something related to AI infrastructure, which is what banks do, but then they, they feel like they have too much exposure and so then they start selling off sort of, there's risk transfer agreements so they, it sort of like reinsurance but for banks. So they feel like they have too much risk and so they start selling off that risk and this ends up being another deal. We also have credit default swaps which are essentially derivatives that protect investors against a downgrade of a credit, not just default. Insurance companies are stepping in and they're providing sort of additional protection against defaults and downgrades and charging a premium for that. You have the hyperscalers that are guaranteed guaranteeing debt in terms of some of these SPVs and other off balance sheet structures and then you also have the off balance sheet structure. So one of the biggest risk is all of this gets spread around and if one of these companies, hyperscalers get downgraded, how could that cascade through the system? Before we continue, let me pause and share some words from this week's sponsors. Have you ever thought I should really be doing something to protect myself from stalkers, scammers and hackers? But you're not sure what? Well here's what you do. Go to www.joindeleteme.com david20 and enter code david20. You'll get 20% off DeleteMe DeleteMe removes your personal information that's being sold online. In the age of AI, we're all especially vulnerable to scammers using our personal data that's floating around on the Internet. And that's why I used Deleteme over the past several years. Several years. If you ever Google yourself and found your home address, phone number or name of a family member, it could be unsettling. And so you don't want that. I appreciate that Deleteme is working for me. They're always scanning the Internet seeking for where my information shouldn't be, and then they send me a quarterly report and say what they requested to be removed. DeleteMe is a top pick for data removal services by Wirecutter. So take control of your data and keep your private life private by signing up for Deleteme now at a special discount for our listener. Get 20% off your delete me plan when you go to JoinDeleteMe.com David20 and use promo code David20 at checkout. The only way to get 20% off is to go to www.joindeleteme.com David20 and enter code David20 at checkout. That's www.joindeleteme dot com David20 code David20 do you ever feel like in this age of AI you're just not keeping up? That so many businesses are figuring out how to incorporate AI? What you' and you with your business? Perhaps you're struggling to figure out, well, how do we use it? What do we do? Well, you probably know netsuite. They've been a long term sponsor of our podcast, but they also have an AI powered business management suite that securely connects all of your data. It's a unified suite that brings your financials, inventory, commerce, HR and CRM into one single source of truth. And it's trusted by over 43,000 customers. NetSuite Next is the next huge leap in how business gets done. Because AI is built into everything you do, it automatically surfaces custom insights throughout your day. AI agents work alongside you to solve problems and handle routine work. And anytime you have a question about anything, you just ask, just like you're having a conversation with a colleague. NetSuite is customized for a wide variety of industries, so it supports the way your business truly works. Whether your company earns millions or even hundreds of millions, it's time for NetSuite Next, where your business meets AI for the first time ever. You can try NetSuite Next for free if your revenues are at least in the seven figures go to NetSuite AI David built for every industry, ready for every boardroom. NetSuite AI slash David. Now, one of the areas in the debt markets that we haven't talked about that doesn't really have exposure or very little exposure to AI related debt is the leveraged loan markets and collateralized loan obligations. Now this is an area that I've invested in for years. AAA CLOs are, they're almost cash like and because of the protection, there's never been a default. I mean there clearly is credit risk there they could sell off. But leveraged loans and Clos, they're variable rate debt. And that's not attractive if you're building a data center. Because if you build a data center and you have a long term lease in place, you have these cash flows over a long term, you don't want your debt to be variable rate, you want it to be fixed rate debt, longer term debt that matches the cash flows you're going to get from the leases. And so there's been less issuance and the market's not really been receptive to leverage loans. And then those are split up and packaged into CLOs. They haven't really been accepting of that for data centers. So you don't really see it's not really there. Now where you're seeing exposure in the leveraged loan market for AI related is about 15% of leveraged loans are software companies which are threatened by AI potentially. And so you've seen some sell off in the leveraged loan market from that concern. But it isn't, it's sort of the threat from AI and the threat from AI, it's upending a lot of business models. So that will impact overall, potentially credit spreads. We'll look again, we'll look and see where we're at today because right now a lot of this fear or potential things that could go wrong is not being priced in in terms of the yields on bonds. We're starting to see some movement for credit default swaps on Nvidia and Hyperscalers, but it's been pretty minute moves. It could be way worse and we'll see. This member of Money for the Restless plus had some questions about, well, what is the exposure of AI related debt in various investment vehicles and versus on large bond index funds or ETFs such as those that follow the Bloomberg Aggregate bond Index. Well, right now it's about 14% of just the investment grade bond index, but that's, that's a smaller percent of the overall bond market because you have government Bonds to make up a huge slice of the bond market. And so overall exposure to at least at this point is going to be single digits and again it is investment grade. What about Equity REIT? Like if you own VNQ which is the Vanguard Real Estate ETF? Well about 9% or so would be depending on how you calculate it, that's tied to either data center specific REITs or could be some other type of infrastructure REIT. So less than 10%. There is some exposure to AI and AI companies. That is something we'll monitor. That's not so much that we're concerned about equity REIT markets are very dynamic and if you allocate to equity REITs, these are, this is commercial properties and there's just a wide range there, there's apartments, there's office, there's definitely data centers, there's self storage, single family homes, retail. And it evolves over time and we're seeing more exposure to data centers. But overall equity REITs pretty attractively priced right now. So not a great concern but some again another area where this AI exposure can creep in. One of the questions is well how do I, how do I check what my exposure is? And this gets a little more complicated. But you can get, you can download all the holdings for a particular etf, you can look at their annual or semiannual financial report. You can also look at their fact sheet to kind of get a sense. It's better in my view just to kind of understand well, what type of fund is this? Clearly like something that tracks the Bloomberg aggregate, it's going to be pretty small exposure right now. On the other hand, if it's an active fund that focuses in securitization or if it's a business development company, closed end fund. So they're going to have a lot more exposure either to software companies or potentially to data centers. So that's awesome. It's going to be the active funds, it's going to be the opportunistic credit funds, potentially private credit funds. I mean private credit is a participant in that and that's why retail investors have been demanding their money back. And it isn't so much data exposure but data centers, but it's also the software. So pay particular attention to your more aggressive bond funds, your opportunistic bond funds to kind of understand what potential data center AI infrastructure exposure they might have. What could go wrong with this AI infrastructure build out? Well, a big concern is the short lived assets are being paired with long dated debt. And so an AI chip might have a Useful life of four to six years. Some of the companies are depreciating these chips over seven years or more. But the debt tied to data centers long term. And so again, that's a risk because go back to this capital cycle. If too many data centers are eventually built, too much infrastructure or the chips change too quickly or something along that we could have debt tied to a data center that just isn't used or that's out of date. So that's to me the biggest risk. A second risk is just the concentration in terms of there's a data center could only have one or two tenants and maybe that particular tenant goes bust. So there's some concentration there. You're seeing concentration in terms of. In the BIS mentioned this BDCs, they're all kind of lending to the same company. So you see more. You might think you have diversification, but it's a pretty small pool and it's one particular niche. And so you do have some of that concentration. There's some refinancing risk as some of this debt needs to be refinanced. And it could be at a time where the market's just less receptive with much wider spreads. We have guaranteed cast state gates, so the risk of that, so you have a lot of guarantors. And we saw this during the housing bust as there were downgrades and so guarantee somebody tried to exercise it and that spread to another part of the financial market. So you kind of had this cascade and that's a risk that it spreads. And then the fifth is just a lack of transparency with a lot of this kind of in private credit funds, like we don't actually know. And that was one of the risk. With the housing crash and the great financial crisis, we didn't know where everything was. And to be honest, we don't have a complete understanding at this point who has exposure, where it's exposure. And that's why regulators on the insurance side, it's why bank regulators, they're all looking at it. I mean, they're much more aware of what could happen. But that is a risk. Now we're not quite there yet. I mean, this is not as, by any means as widespread as what went on the housing bubble. But even there you could see it evolving and you could reduce your exposure. It's just that investors were chasing yield and we don't want to be in that position that we're chasing yield. We want to be able to understand and explain what we own, what exposure we have. And this is no different. But Those are some of the risk I mentioned that spreads are very, very narrow. So this is just intermediate credit. So this would be the spreads for intermediate investment grade corporate bonds. And right now they're about 60 basis points. So 0.6%. The average is 1.1%. I could share a chart for non investment grade spreads or the overall corporate bond market and they're all close to their all time lows and well below their average. And yet we saw like in this chart, the spread was under 1%. But when you start started to see the cascade and the downgrades and the fear default during 2008, late 2007, that spread went over 5.5% for investment grade. Non investment grade went close to over 15%, close to 20%. And so if we get to the point where it becomes clear that much of this investment wasn't justified, you'll see AI exposed stocks sell off. You will also see the debt sell off. And so I monitor the spreads very closely. We're definitely in, in our monthly strategy report, we're, we're monitoring both sides of this, the potential for AI, the impact on productivity. Much of our strategy report last month was on are we seeing productivity show up the numbers, the AI and how this AI bubble compares to the Internet bubble. In fact, you can get a free copy of that investment strategy report by going to Money for the recipe us.com report and you can read that and get a better understanding. But we're monitoring it. I'm not overly concerned yet. It's too early, the potential is too great. And so we'll monitor the latest frontier models. We'll continue to monitor the amount of debt lending, the whole debate regarding open weight models versus closed models and weight. We're monitoring the productivity statistics. Are we seeing the actual boost in productivity innovation from AI? Obviously we're monitoring the overall economy in terms of will employees be replaced by AI or will AI create new jobs to offset the jobs that are lost? Cause jobs will clearly be lost, but what new jobs will be created? We are so early in this cycle, but we wanted to take a look, I wanted to take a look at where are we on the debt side and on the debt side again, also early stages, a number of vehicles being used, a number of ways to offload risk. But right now we're still early. So let's keep monitoring it. Be especially wary of more opportunistic credit funds that you might own to see what their exposure is in terms of CLOs and leveraged loans. The exposure's more on software and potential threats to businesses, which again threat to businesses. If it becomes more real, it'll show up in widening spreads. Which is why right now it's not a great time to be taking a lot of credit risk because you're not being compensated for it. Think about that. In an era where massive debt loads being taken on to build out the AI infrastructure bill, all that capital demand, yet we have spreads, the incremental yield still very, very narrow. At some point it's going to widen out and that debt will sell off. And so we should be cautious right now when it comes to corporate credit, corporate investing, having exposure to AI data center debt because you're just not being compensated given the risk. That's our discussion today on the AI debt Bubble. One of the most popular pages on the Money for the Rest of Us website is the list of books that have influenced me as an investor. Now there are a lot of books on investing out there. I've written one and I've read hundreds. Now many just aren't worth your time. And so I share the ones that have influenced me. But we're trying something new. I created a quiz to help you go beyond bestseller list and get a recommendation from my favorite investing and money books based on your current interest, questions and goals. And you can take that quiz@moneyfortherestofus.com quiz. There are books on investing, the nature of money, central banks, risk management, economics and retirement planning. The quiz is only six questions long. It takes just a few minutes, but you'll get a recommendation, you'll get a follow up email and I'll tell you why. I think it's an excellent book. So go ahead and check it out@moneyfortherestofus.com quiz and see which investing book is best for you. Everything I've shared with you in this episode has been for general education. I'm not considered your specific risks situation. I'm not providing investment advice. This is simply general education on money investing in the economy. Have a great week. It.