The World’s Largest Wealth Fund Is Preparing for an AI Crash

When the person responsible for managing nearly $2 trillion of a nation's savings starts discussing the possibility of losing all of it, you should probably listen. Though deep down, you already know what I'm talking about. You think everyone is long in buying the stock market, but the truth is, people are already checking out, especially the institutional types. Nikolaj Tangen runs Norway's government pension fund Global, the largest sovereign wealth fund in the world. He isn't predicting that the fund is going to disappear tomorrow. He is explaining the extreme risks an institution of that size has to prepare for. One of the stress tests modeled an AI bubble bursting. The estimated loss was a shocking 35%, which would erase about $700 billion in value. Norway's fund can't simply avoid the stock market. It owns a piece of thousands of companies. In many ways, it is the market. That's the defining contradiction here. You're told everyone has to keep buying because stocks are rising. AI is transforming everything, and missing the boom could be the greatest financial mistake of your life. But behind closed doors, many professional investors are no longer asking how much more they can buy. They're asking how they might eventually get out. I spent the past week at Hugh Hendry's Asset Capital summer camp, and I also interviewed money managers Ed Dowd and Ben Bray on Eurodollar Talks channel. These people don't all own the same assets or make the same trades, but I kept hearing versions of the same conclusion. This bubble is different from the dot-com bubble, sure, but at the end of the day, it's still a bubble with all the same characteristics, the bubbly characteristics, and this one more susceptible than that one to the debt cycle, which is really about the macroeconomic cycle that almost everyone has basically written off, everyone except maybe Norway's sovereign wealth fund. What they were talking about in their stress tests and saying out loud what a lot of people believe is more than simply just a warning about stocks. And the easiest response is that today is nothing like the dot-com era, and in many important ways, that's correct. In 1999, companies could add dot-com to their names, create a website, and raise enormous amounts of equity just by having little revenue and no plausible route to profitability. Today's largest AI spenders are companies like Amazon, Microsoft, Alphabet Google, Meta, Oracle. These are real companies with global businesses, enormous revenues, and valuable assets. AI is also real. It can write software, analyze documents, create images, and automate parts of complicated workflows. But then again, the internet back then was real too. The mistake during the dot-com bubble wasn't believing the internet would change the world because it did. It was assuming every price, every projection, and every single investment made in the internet's The same distinction applies now. AI can become one of history's most important technologies, while AI investments made in 24, 25, 2026 could still produce terrible returns. The bubble isn't necessarily inside the technology, it is inside the assumptions being used to finance it, including how much money it's actually going to take. Before we go on with our video today, let me ask you a question. What if your gold could actually pay you every single month in more gold? That's what today's video sponsor, Monetary Metals, lets you do. And here's how it works. You still own the gold. It's still in your name. It's fully insured. But instead of just sitting there doing nothing, it actually starts working for you. Monetary Metals has built a system that turns idle gold into productive gold, earning real yield, paid in physical gold, not dollars. No selling, no trading, no converting. You keep your gold, and every month, you get more of it. It's a simple idea, but it solves a big problem. If gold is money, it should act like money, and now it finally can. If you're a long-term gold holder looking for a smarter way to store value, check them out at monetary-metals.com slash Snyder. Again, that's monetary-metals.com slash Snyder. I highly recommend booking a call with them today to see if they're the right fit for you. We thank Monetary Metals for their sponsorship, but now let's get back to the rest of our video. So this is a common theme that I heard at Hugh Hendry's place, Acid Capital, this past week in St. Barts. All the professional money managers, and there are some good ones, they keep running into the same thing over and over again. We all know it. We all see it. It's not like they're saying we can time the market and tell you exactly where the top is. It's just that we know this pattern. They've seen this pattern before. We've all seen this pattern before. It's just that nobody wants to say it on the way up because if you're too early, you're going to be made fun of. All these institutional money managers are doing something more important than just trying to time the market, which they know they can't do. They're questioning the assumptions that now have to be true simultaneously. AI revenue has got to grow incredibly quickly, which it is. Computing costs, they have to follow, which they're doing. Electricity has to become available. It is, to a certain extent, but becoming hard. Data center construction has to remain on schedule. The customers have to keep paying. Credit has to remain accessible, and that's a big one. Plus, stock valuations have to stay elevated long enough to finance and grease the entire transition. And if just one of those assumptions is out of place, sure, a project might survive. But if several of them are wrong together, the math changes completely. That's how institutions think about exits. They don't wait for a television anchor to officially declare a bubble or somebody famous to tweet about it. They hedge. They reduce concentration. They do stress tests and think forward. They demand better terms and build liquidity and roadmaps to liquidity before everyone else recognizes the problem. Norway's fund isn't stress testing a 35% decline because it knows one is coming. It is doing so because everyone knows it is. There isn't a person out there who doesn't see it the same way, including a wager. You, dear viewer. We all see it. Hardly anyone wants to break from the herd and say it out loud, except maybe that's the part that's also changing. During the first stage of the AI boom, money was everywhere, including from the companies themselves. They were highly profitable. Therefore, the idea was that they could internally finance much of this build out capex and everything that was going to be needed to get from the starting point, point A to the golden future, point B. However, very quickly, the numbers started to change and so too have the conditions. But the scale of the build out has really changed the equation. Like anything, you budget X and it seems a reach, but maybe doable. But over time, the budget swells, then balloons and halfway through you're stuck, but you have no other choice than press forward and hope everything just goes perfectly. Data centers require a whole hell of a lot. They've got chips, memory, networking systems, cooling systems. You got to have land, specialized constructions and enormous, gigantic amounts of electricity. Every major company is chasing those resources, those finite resources at the same time. It's leading to a supply shock, a mini supply squeeze. Now, Alpha recently reported its first quarter of negative free cash flow since it became a public company. It spent a gargantuan $44.9 billion on capital expenditures in one single quarter and increases projected 2026 spending to between $195 billion and $205 billion. And management says spending is going to rise significantly again next year. And it's not just Google, Alphabet, Amazon, Microsoft, Meta, Oracle. They all say the same things. They're all spending, therefore borrowing aggressively. And now NVIDIA has joined BlackRock, Goldman Sachs, KKR and a couple of other major financial institutions in an effort intended to mobilize as much as half a trillion in financing just for NVIDIA's customers. Now, those customers include companies that want NVIDIA's chips and data center capacity, but they're struggling to secure financing through ordinary channels. That's the key here. Think about what that means. When customers can't afford the product and they can't get financing for it, financing the customer becomes part of selling the product. That's the circular nature of the AI bubble that everybody's talking about. The AI race is no longer merely about who develops the best model. It's about who can obtain enough capital to continue building. And credit providers, they're already neck deep in this boom. So they have every incentive to keep it all going. Not to mention, convince the world to keep it going by keep shoveling money at it. But while all this is taking place and the ask gets even harder and more difficult, everybody has set aside economic concerns because they are told the economy, the labor market, not just here in the US, but also around the rest of the world, it's resilient, it's holding up reasonably well. But what if that isn't true? What happens then if the macro cycle starts to turn? What does that do to the credit cycle? And therefore everything else beside. One wild card out there, that's the macro climate that everyone has written off because Jay Powell and now Kevin Warsh said so. Maybe it's not great, but it's at least supportive of the bubble framework. More and more, however, we're seeing that's not the case either. July payrolls just came out in the US last week. They declined. Yes, negative number 23,000. We're repeating the pattern that we saw last summer. In the previous two months, revised down by a combined 103,000. Again, similarities all over the place. The supposed labor market strength from earlier this year was basically revised away after it already supported briefly the resilient economy narrative. The narrative is taking as much of a hit. And the household survey, way, way worse. Employment over there fell again. Full-time employment declined by more than 100,000 and roughly another quarter million people left the labor force after three quarters of a million people left the labor force the month before. Since January, employment is down by more than 900,000, while 1.4 million former workers dropped out entirely. These are huge numbers. They're bright red flashing alarms. The payroll numbers, all they're really doing is just coming back down into flat beverage where the rest of reality has been this entire time. And that last part matters about employment and participation, because people who stop actively looking for work are no longer counted as unemployed. So it leads to this misidentification, the official unemployment rate drops, while unofficially unemployment is a major problem already. That's why the official unemployment rate can look stable, while actual employment and participation both deteriorate substantially. The labor force participation rate has fallen to 61.4 percent. It was 62.5 percent back in November. So when we do some adjusting and account for these missing workers, you get the effective unemployment rate that's closer to 6.2 percent, not the official 4.1 percent. And in this context, that is everything. That weakness is already showing up in income data. It's showing up in consumer behavior. I talked about how real personal income, excluding government transfers, suggests the major downturn began around October and November, right around the time the labor force started to experience hundreds of thousands and now almost a million and a half dropouts. So while the stock market prices decades of AI-led productivity, households right now in 2026 are confronting fewer job opportunities, weaker income growth, and declining full-time employment right now. And that matters because as the economy gets even flatter beverage, the credit cycle becomes even harder, the funds dry up ever more, the pathway to reach the goal line grows murkier and less likely. And in many ways, the macroeconomy becomes the credit cycle. But from the perspective of all these hyperscalers, not to mention the credit providers that are already neck deep in all of this, they have no choice. They have to keep pressing ahead because they've already sunk everything in this. Not only do they have to keep pressing ahead, they've got to try to convince you to go along with it because there's no way to do so otherwise. So just this week, six of the biggest credit investment names on Wall Street announced a half a trillion dollar effort to raise money for customers of Nvidia to pay for computing power. They got to. They have no choice. As the New York Times writes, quote, the firms BlackRock, Goldman Sachs, and KKR among them said they were working together to come up with that huge sum to lend to Nvidia's customers, including the startups that use the company's chips and data centers to develop and operate AI software. These customers, Nvidia said, have been struggling to secure financing for chips and data centers. Huge red flag here. Nvidia will connect its customers with one of these six lenders, which will provide financing that could range from loans to credit. And the financing, Nvidia said in a blog post, will be at attractive rates. Again, a huge, huge red flag if there ever was one. It admits the money is getting harder, the credit cycle is getting tighter, and the pathway forward is more uncertain and difficult. Now for BlackRock and Goldman and the entire boys club over there, what else can they do? They need this to happen to keep the AI bubble alive and progressing. The firms also hope that this is going to send some kind of message and boost confidence at a time when confidence is fading fast. And in fact, again, that's what this is responding to. But it just raises further questions. What else can they do? The only way forward for them is more and more money. The more they do, the more you know it's not coming in the way that it was before. People are asking questions and taking action, even if stock prices aren't seeing the consequences just yet. Larry Fink at BlackRock sure does, which is why he and the others are trying to dig up another half a trillion just to keep the dream alive. And after investing hundreds of billions, stopping becomes extraordinarily difficult if not entirely possible. And admitting that this is getting difficult and admitting it's getting more and more unlikely is impossible too. Slowing down raises questions about previous spending. It threatens supplier. It damages confidence in the entire AI narrative. So instead of slowing down, they are doubling down. Then they're going to triple down. And they make a very public show of it because when a boom requires increasingly elaborate funding merely to preserve momentum, credit is no longer standing behind the story. It has become the story. And the problem with that is this accelerating borrowing need from the AI buildout is coming during a credit cycle that by almost every account has already shifted. And private credit, which is at the center, but not the sole story anymore. They keep reporting lower asset values, rising non-accruals, however, slowly and dividend reductions. FSKKR, they've marked down the net asset value over there for consecutive quarters after previously receiving a $300 million support package. We see stress appearing inside mortgage finance, as I talked about last week. United Wholesale Mortgage, which is America's largest private mortgage originator. They reported a quarterly loss of roughly $452 million. They suspended the dividend. And they more importantly arranged a $2 billion capital infusion. Talked about that in the video this past weekend. Now, it doesn't mean American homeowners are suddenly repeating 2008 because there's not going to be a repeated 2008. But most existing borrowers, they have fixed rate mortgages and they still possess considerable equity. The weakness is appearing first in the machinery around the loans, originators, warehouse financers, servicers, the hedges that go along with it. And of course, the leverage, highly leveraged intermediaries and financial structures. And in every credit cycle, it's the plumbing that tends to leak before the building starts to flood. So these are the mistakes accumulated when lenders assume transaction volumes are going to remain high. Financing would stay available forever. And assets would be sold or refinanced easily. Now you add hundreds of billions of dollars in AI-related obligations to the same system. And that's the difference between the dot-com era and today. Many dot-com failures destroyed equity because speculative companies had no profits. This cycle is constructing physical infrastructure with debt, with private credit, with leases, and of course, these complicated financing partnerships. The assets are real. But so are the payment obligations. The second weren't there back in the 90s. The data center can't tell its creditors that AI will eventually change the world, even if it will and even if it's likely to. It's got to produce enough cash to service its financing and increasingly do so on schedule. And that's why institutional investors anyway are looking to the credit market for signals about the credit cycle because that's where everything is this time around. The credit cycle, therefore the macroeconomic cycle, they're not merely risks. A stock investor asks how large a company could become. A lender asks whether it's going to be able to make the payment. Credit default swap prices for several major AI-linked businesses have been rising. It doesn't mean defaults are imminent. But what it does mean is that investors are paying more to protect themselves against the possibility that their projections fail. That behavioral change, the need for hedging, is more important than any single spread or any single price. A year ago, the primary concern surrounding an AI bond offering might have been, can you get a large enough allocation? Now the question is becoming, what's the actual compensation I'm going to receive? And how do I hedge this exposure? Even the strongest issuers are encountering greater resistance. Recent giant bond offerings from companies such as Amazon and Alphabet, they've required more generous terms than investors once expected from these elite borrowers. They can still raise the money. That's not the warning. The warning is that lenders are becoming less willing to treat every dollar of AI spending is automatically productive and automatically worthwhile. Because adoption isn't the same as profitability. The industry keeps assuming that costs are going to fall while pricing power rises. And that could happen. In many ways, it's still happening. But when trillions of dollars depend on everything going just right, an assumption, several assumptions become systemic risk. Now, Michael Burry is basically one of the few who has been willing to say the quiet part out loud, what we're all actually thinking. Now, he said the stock market is setting up for a 1987 style crash. Maybe, maybe not. But the reason he's saying it and the reason we're all nodding our heads in agreement is because we see that as a legitimate risk. It's more than a legitimate risk. It is a recurring pattern. And paradoxically, Burry pointed out, again, something we're all thinking, the mechanism that can make this final stage look the strongest as markets rise in price and volatility falls. Volatility targeted funds are permitted to take on more risk and momentum strategies to add more exposure. New highs attract new money. The market becomes increasingly leveraged precisely because it appears to some people to be safer. That cycle worked beautifully on the way up when nobody's really questioning anything. When volatility returns and when rationality returns, those strategies can be forced to reduce exposure and do so together. This is why institutional preparation doesn't necessarily look like immediate selling. Large funds can't liquidate everything without moving prices against them. So instead, they purchase protection. They shorten their duration. They demand larger concessions. They preserve as much cash as they can. And they start doing stress tests and identify what it is they can sell quickly if they need to. Publicly, they remain fully invested. And in many ways, they say they remain fully committed. But privately, they're over there by the exit door measuring its proportions. The Norway Fund's warning, it's not that they intend to dump every stock. It's that passive confidence is not a risk management strategy when market returns depend on a small number of companies making historically large investments based on an unknowable future demand. The most dangerous idea in this market is that because everyone has to participate, everyone must also believe in it. They don't. The professionals I spoke to in St. Bart's, as well as here at URUDAL University in our URUDAL University talk channel, they recognize that AI is real. In fact, they're advocates for it. But they also recognize that a bubble built on wild assumptions and tons of leverage, not to mention mistakes accumulated across credit and economic cycles, is not as stable. It is a bubble, whatever name you want to apply to it. The hyperscalers, they cannot stop building. Their suppliers can't afford them to stop. The lenders financing the expansion need confidence to remain high to get it all done, to have a prayer of getting to where they want to go and making their debts pay off. So the announcements get bigger. The commitments become more public. And the industry doubles, then triples down. Meanwhile, employment, income, credit data, the real economy, everyone's now being forced to think credit cycle as well as macroeconomic cycle. Norway's sovereign wealth fund is preparing because it understands the fundamental rule of risk. You don't build the exit after the fire starts. You think that everyone has to be in this market long and loving it. The truth is, increasingly, for institutional managers anyway, they're not thinking about the next thing they can buy. They're already thinking ahead to what they might sell or what the exits are going to look like if they really have to rush over there. For Norway's sovereign wealth fund, they know they can't reach the exit. They know they can't get out. So they have to do the next best thing, which is to prepare people for the proportions of what it might look like when everyone does try to get there. And if you're interested, we just finished up our last webinar. We talked about portfolio management. Construction is more importantly maintenance. During circumstances like these, using your dollar signals, the curves, the data series that either no one else uses or if they do, they get it wrong and backward and why they do all those things. A couple of webinars in the series, the last two, there's a link in the description for both of them, as well as to our webinar subscription offer. Highly recommend you check all of those out. Like I said, link in the description for all of it.