Transcription
Before the 2008 financial crisis, the most powerful financial institution that you've likely never heard of issued a warning that the system was building toward collapse five full years before it actually happened. And now that same institution is issuing a warning about AI. All those years ago, they warned that indicators of risk perception tend to decline during the upswing. Those warnings were ignored, and we all know how that ended in 2008.
And now that same institution is showing that markets are once again ignoring risks. They are accepting less and less payment for holding it. Exactly the pattern they flagged before 2008. And they warn of four major pressure points converging at this moment, with AI at the center of all of them. The report warns that "a major equity market correction could have larger macroeconomic consequences today than in the past." That it would have more pronounced wealth effects, a sharper consumption pullback, and with US stocks at 64% of global equity markets, a US-led repricing could "propagate globally." Financial stability, they write, could also be at risk in the event of an AI burst. In plain terms, AI coupled with the oil shock of the ongoing war in Iran are creating a flash point that we must understand.
Finally, we also want to stress developments in the public finances. Near record-high public debt and higher interest rates are straining fiscal positions globally, leaving governments with limited room to respond to crisis as the cost of servicing debt escalates and deficits remain stubbornly high.
What's worse about this particular confluence of risks is that while the short-term gains of this boom flow to the 10% of our population who own 87% of the stock market, aka the rich, the bust will be distributed to everyone. So I don't think it has a problem with productivity. I do think that, uh, but productivity, it has a big wealth gap implication. A very small percentage of the population is going to do unbelievably, and a lot of people won't. So what do we do? Can we work together politically to deal with those issues? And how do you optimist? I do not believe I'm not optimistic on us working together.
And how this ends may come down to one chart, which is buried deep in the report on page 23. And what it shows is that using the AI industry's own projections, the math puts the industry $2 trillion in the hole unless AI delivers everything that's been promised. And we all know that so far, it has not come close. Goldman Sachs finds "no meaningful relationship between productivity and AI adoption at the economy-wide level." And their own chief economist calls AI's contribution to GDP growth basically zero. And if this happens, if the bust comes, governments have a fraction of the room that they had last time to help counter the impact. When 2008 hit, US debt was 35% of GDP. Today, it's 100%. And interest payments already eat a fifth of federal revenue, double the prior crisis. So let me show you what this report shows, what it means for you, and what you can do to prepare. I'm Brendan Dell. This is the Leverage Class. Let's see through it.
There's a tower in Basel, Switzerland that houses one of the most powerful institutions that you've likely never heard of. It's called the Bank for International Settlements, but it's a bank in the technical sense only. You can't open an account there. Neither can I. Neither can Apple or Goldman Sachs or any company on Earth. Its clients are only central banks. Things like the Federal Reserve or the European Central Bank or the Bank of Japan, which is why it's often called the Central Bank of Central Banks. When the people who print the world's money need somewhere to meet, they go to Basel. They meet every two months behind closed doors, and then once a year, they go to receive a report card on the system that they run. In fact, the rules that govern how much capital every major bank on the planet must hold are named after this city because they're written here.
And in August of 2003, two economists from this institution walked into the Federal Reserve's own annual symposium in Jackson Hole, Wyoming, and told the assembled central bankers of the world that their victory was making them blind. The paper was by Claudio Borio and William White, two of the BIS's most senior economists, and their argument, compressed, was that central banks had spent 20 years winning the war on inflation. And the prize was a new kind of danger. With inflation conquered, nothing forced the brakes anymore. Credit could expand, asset prices could climb, and every instrument on the dashboard would read normal, right up until it didn't. In their words, the system's very stability was raising its "elasticity," which was making it more vulnerable to boom and bust cycles. And the central bank, they wrote, can be a victim of its own success. In layman's terms, they told them, "You guys are getting cocky and you're not appropriately managing risk." And they even dedicated that paper to the memory of Charles Kindleberger, the man who wrote *Manias, Panics, and Crashes*, and who himself was a former BIS staffer.
That symposium was opened by Alan Greenspan himself, who was chairman of the Federal Reserve at the time, aka the man in charge of America's money. And he told that room that uncertainty was "the defining characteristic of monetary policy." And then two of BIS's most senior economists presented a paper showing that room exactly where the uncertainty was hiding. But the Fed was not persuaded. They kept rates low. The housing market continued its tear, and by 2006, William White, the official we just met, wrote that "one hopes that it will not require a disorderly unwinding of current excesses to prove convincingly that we have indeed been on a dangerous path." But unfortunately, we as human beings seem to like learning things the hard way. And two years later, that disorderly unwinding arrived in the form of the Great Recession, which is one of the worst economic events in modern history. The BIS had been correct. They had been early, and they were ignored. And being ignored in 2008 cost millions of people their jobs and their homes and their savings.
Which brings us to two weeks ago, when this same institution published its annual report card on the world economy. And their warning lights are flashing again. But this time, they're pointing at AI. And using the AI industry's own numbers, their analysis shows that if things don't go perfectly to plan, the fallout would hit harder [snorts] than it did in 2008. Because this time, the exposure runs through your retirement account, whether you've ever bought a share of Nvidia or not. So to understand why we have to do the thing that always gets us closer to truth, we have to follow the money. And the money is going one place: AI.
So the single biggest thing that we must understand to properly weigh the risk of the AI boom and its impact on our lives is that unlike the last 15 years where companies were borrowing money because they were making so much of it, what this report shows is that the most profitable companies in the history of the world have completely flip-flopped and are now borrowing money because they can't keep up with their costs. And the only rational thing that they can do, even if it seems crazy, is to keep spending. And spending is spending. Or major players like Meta and Microsoft and Google all run the risk of ruin. The report names four pressure points converging on the world economy all at once: persistent inflation, AI investment, growing financial vulnerabilities, and weakening fiscal positions. And this video walks through all four. We start with the engine: the AI spending spree.
30 seconds of boring finance because three dull terms are about to matter a lot. Term one: corporate bonds. So, when a company wants money without selling ownership, it borrows from investors and promises to pay it back with interest. This is like if you ask your buddies to loan you, you know, like a cool billion, and then you'll pay them back with 3% interest. But the bigger you look from the outside, the more you can raise, and the cheaper it gets. Term two: free cash flow. The money a company has left after paying to both run and grow itself. It's the finance world's answer to, "Yeah, but how much do you actually make?" And then term three: financial engineering. So if you're a person or a small business, you basically have two options, which is earn money or borrow it. And how much you can borrow is capped by what your monthly income supports. But if you're a very big business, those rules loosen, and you can structure your borrowing and your taxes and accounting so that spendable cash shows up where you need it. Yes, accountants, I know that is not a precise definition. It's the sentiment. Apple ran a very famous version of this. So in 2013, it borrowed $17 billion while sitting on the biggest cash pile in corporate history because that cash was overseas, and borrowing it was cheaper than paying the taxes for bringing it home. S&P even gave this trick a name, which they called synthetic repatriation. Most people borrow money if they don't have enough money, but big companies will often borrow when they have too much of it.
The BIS data shows that AI commitments at the five biggest spenders have passed earnings and passed free cash flow. And Morgan Stanley estimates that $3 trillion of data centers through 2028. Only half are covered by the cash that these companies can generate. And the rest has to come from debts and from cuts. For example, Microsoft just laid off 4,800 people while carrying a $190 billion spending plan. What makes this scary is that despite these huge risks, it's basically a forced play. AI threatens the base business of all these major players: ads, search, enterprise software, everything that these companies own. All great technology changes, um, produce bubbles. And the reason they produce bubbles is because nobody can get, get it exactly right. Okay, there, um, you have to either spend a ton of money to capture your market share and so on, or, um, and don't worry about whether it's too much or not, or you don't spend enough money and you lose your market share, and it's very imprecise with a lot of competition. Okay. And the BIS sees this precise risk growing. They say it's a contest. Bond issuance by these AI-related firms is now a very significant share of total issuance by corporates. And what they're finding is that even if things go completely to plan, if AI delivers absolutely everything that it's promised, we are still entering a very risky scenario.
Which brings us to the chart from the beginning of this video. So first, the roughly $2 trillion already committed. This is the red line. AI delivers everything promised, and then the sector's payoff that we see here still falls with every trillion spent because they're all chasing the same prize. Now, the blue line is AI disappoints, which in their model means AI delivers half. It doesn't mean it fails outright. It doesn't mean it produces no value. It just means it does only half of what they're saying. So at marker B, the $3 to $4 trillion that Nvidia's own CEO projects by 2030, and this is one of the largest bulls in the entire AI economy, by the way, puts the entire sector $2 trillion underwater. But the most important point is this: the money is borrowed. The shortfall lands on whoever lent it. And increasingly, that is not the banks that you'd expect. It's private credit funds, it's bond portfolios, it's pensions, and as we're about to see, no one can even fully understand where all this is going to land.
So, the big question then becomes, how much is AI actually delivering? Well, this is being measured, and so far, things are not looking good. The biggest challenge of large language models is that the technology itself is being anthropomorphized and used as a synonym for all of AI, all technologies of artificial intelligence. And as a result, the companies building the models are being priced as a replacement for all of thinking when they are actually normal technology, when specific applications whose limits are already being found. And we see this happening in three places in real time.
The first is the frontier models are being commoditized by their own customers. Yesterday, Bloomberg reported that Microsoft has started replacing OpenAI and Anthropic with its own cheaper models inside Excel and Outlook. Microsoft's AI chief on the record: "We pay a lot of money to Anthropic. So our goal is to reduce these costs and ultimately eliminate them." Chinese open models charge a 20th of frontier prices for all of the everyday work, which is most of what the technology is used for. If AI were a thinking replacement, its biggest customers would not be swapping them out to save money on spreadsheet formulas.
Second, the returns from scaling's flattening, which said plainly means the models are unlikely to keep just getting better and better and better. Ilya Sutskever, who co-founded OpenAI and built the scaling era, in his own words said, "is the belief that if you just 100x the scale, everything would be transformed. I don't think that's true. The age of scaling is over." He says, "We are back in the age of research." Two computational scientists published the math of why the scaling laws' own exponents make reliability gains brutally expensive. The paper is literally titled "The Wall Confronting Large Language Models."
Third, the economy-wide reports. Goldman Sachs reports that there is no meaningful relationship between productivity and AI adoption. Their chief economist puts AI's GDP contribution at basically zero. They do show real gains of about 30% in exactly two jobs: coding and customer support. 30% in two places, zero everywhere else. That's what a tool looks like. That is not a workforce replacement, and it cannot justify these valuations.
Now, in fairness, many technologies have a lag between when they deploy and when we can actually measure productivity improvements. The normal technology researchers who I referenced earlier explain in that document that electricity took nearly 40 years to show up in the productivity statistics. The internet took a long time also. The diffusion of technology is slow, and it is very likely that AI is in that gap. But even if it is, this won't save the industry. Let's go back to that chart. The industry spending plan only works if AI delivers absolutely everything. That's the best case, per the biggest AI bull alive. And even his line barely pays. Missed by half. And every measurement that we just looked at shows them missing. And the sector is $2 trillion in the hole with borrowed money that has to be paid back.
Which brings us to the single largest insight of this report. The plumbing that moved all this money into AI creates risks that reach far past the industry and into the financial system, into your finances. And it's revealed through one phrase hidden on page 25 of the report. And that phrase is "pledged multiple times." That's the risk that the report calls "growing financial vulnerabilities," and it's pointing at a financing structure that means no one, not the BIS, not anyone, can fully see where all this will land. But what we do see is the people best positioned to understand where those losses will land are already heading for the exits.
So, you've likely seen these spaghetti diagrams running around the internet showing how the AI boom is being financed. What it shows is that big companies like Nvidia agree to invest in startups like OpenAI in exchange for purchase orders on their chips. And the whole thing then spins round and round and round, fueling this economy. But the BIS flags this risk in the same understated but severe way that they flagged the mortgage risk. They explain the opacity of AI sector financing compounds these vulnerabilities. Hyperscalers, chip makers, and AI labs are linked through a complex web of private arrangements. The most prominent is circular financing. Chip makers and hyperscalers take equity stakes in AI labs or cloud providers who in turn commit to multi-year purchases of chips or computing power. They continue by saying signs of stress are already visible and that the real economy implications could be substantial.
So then, what specifically are those signs of stress? Insurance against these companies defaulting, called credit default swaps, have been growing steadily more expensive since January of last year, even as the stock prices keep climbing. What this means is that people in the know see risk. The bond market and the stock market are pricing very different futures for the same companies. And the retail credit funds that lent to this sector are already facing redemption requests and for sales. And what we see as all this is happening is the people with the most knowledge of what's going on inside starting to run for the exits.
The companies that go public are those whose current investors are saying, "We don't believe this company will go up in value, so we'll sell it to public market investors." Last month, the biggest IPO in history happened, and it was priced at $135 a share. Morningstar's analyst said it was worth only $63 a share. It spiked and then has since declined. SpaceX set aside roughly 30% of the shares for retail investors, which is to say ordinary people, through things like Robin Hood or Schwab. Normal [snorts] IPOs allocate single-digit allocations to retail investors. What we are seeing happen is after 20 years of private appreciation, they stacked companies together, then open the doors to the public at twice what the professionals said it was worth. And it's what OpenAI and Anthropic have confidentially filed to do next. And yes, I know that OpenAI offering is delayed. We can't go into all of it in this video. And they're planning to IPO at sales-to-price ratios that far exceed what Jay Ritter, who is a man who has studied every major IPO of the last century, calls the danger zone. We'll link that video in the description if you want to learn more.
To be fair, none of this is illegal, and none of it is new. Telecom vendors ran money in circles in 1999. As one example, Lucent alone extended more than $8 billion in loans to customers so they could buy their stuff. But when the funding stopped in 2001, most of those loans were never repaid, and Nortel went from a $398 billion valuation to bankruptcy. What's new about this particular situation is the scale, and the fact that this version runs through all these private deals that no one can fully audit, which is what creates so much risk for the average person and for the economy at large. The professionals are buying insurance, credit default swaps. The lenders are all stretched. And the founders are getting out, which leaves one group still fully committed, mostly without knowing it. If you have a retirement account at all, or a job, it is you.
And the risk the report flags is that unlike 2008, when the government was able to rescue our economy through bailouts, this time it would be much harder for it to do so. This is the pressure point that the report calls weakening fiscal positions. So let's say that the BIS is right. The returns disappoint, financing unwinds, then those losses land on everyone. When Lehman collapsed, US government debt stood at 35% of GDP. Today, it's roughly 100%, nearly three times. And interest payments alone now eat about a fifth of all federal revenue, which is double the burden of any prior crisis. The United States now spends more servicing its debt than it spends on national defense. The BIS says in their calm and understated way, "Near record-high public debt and higher interest rates are straining fiscal positions globally, leaving governments with limited room to respond to crisis." They are calmly warning that the whole system may go up in flames with no fire extinguisher.
Which brings us to the last pressure point of the report, which is inflation. Specifically, the oil shock out of the Strait of Hormuz. This week, the Iran ceasefire collapsed, and oil started climbing again. Last time that shock hit, the report notes that AI spending is what held up the economy. But this time, the shock and the doubts about AI are arriving together. So at the press briefing for the release of the report, a Reuters journalist asked the question directly, "How big could this get? Could this be financial crisis big?" And the response was, "Individually, each pressure point might not be particularly worrisome, but it's the combination of the four. We cannot exclude that they might materialize and damage the global economy in a more significant manner." The general manager of the central bank of central banks was asked on record whether this could be 2008 scale, and he didn't say no.
And the distribution of the losses of all of this is what makes it worse. The boom's gains have already been distributed. The top 10% of our economy own roughly 87% of the stock market, which means that the benefit of the AI boom has flowed mainly to people in that sector who hold those stocks. The IPOs move the risk to retail, the debt moved to bond funds and pensions, and if the bust comes, it arrives as a recession, which will impact consumption and jobs, all with the government will be too stretched to cushion the blow. So, the gains went to the people who own the boom, but the costs will be borne by everyone.
Last week, the Financial Times reported that OpenAI has proposed the US government take a 5% stake in their company. And in every leading American AI lab, through a sovereign wealth fund. In 2008, the government took stakes in companies as the price of rescue after they had failed. This is a company offering equity prematurely while insisting everything is fine, and offering that equity to the entity that writes its rules and would run its bailout. But regardless of the framing, however, they couch the offer, what it would unquestionably do is give the government a multi-billion dollar interest in these companies staying afloat, as well as regulatory capture. Talk for another video. Now, these talks are very early. It might require an act of Congress. It may never happen. But it's the implication of the proposition itself that we need to understand.
The BIS closed its briefing with one sentence of advice: "Policymakers must act now to label only make the necessary adjustments more costly." In 2006, the BIS warned, "One hopes that it will not require a disorderly unwinding of current excesses to prove convincingly that we have indeed been on a dangerous path." But unfortunately, it seems that it may again require that disorderly unwinding to show that again, though we very much hope not.
So, I've been asked in the comments to conclude these videos with my short perspective on what I'm doing. Here it is. Years ago, I got to meet one of my friends' godfathers, which this was a guy who had made close to a billion dollars in real estate. And this guy knew everything about real estate. He knew rental rates in every market. He knew everything about his buildings. He could tell you what the copper wire was worth in his walls. And me being young and trying to be smart, I asked him, "So, what's the difference between a good investment and a bad investment?" And I thought he was gonna have a fancy answer. But without blinking, what he said was, "10 years." So I always remember this quote because what I find is that all too often in life, rather than seeking compounding, looking for long-term returns, we try to run to find an exit. We run away from what we don't want instead of toward what we do.
So I would suggest asking yourself this question: If I could spend my days doing any kind of work for the next 30 years, can't be lying on a beach. It has to be doing something productive for others, but it could be anything. What would that be? And then I would make a direct plan to pursue leverage in that direction. Gain specific skills that you can sell. Specialize. Don't generalize. Build an audience of people who know who you help and how you help them. And you don't have to dance on TikTok. If you have a list, an email list of a thousand CFOs, right? You're in finance. You will always have opportunity if those people look at you as a voice to listen to in your space. And this is buildable by anyone. And you will always have opportunity. Be an expert in how modern technologies can give you leverage in your specific area of expertise. Build processes so that you can detach money from time, and then, and only then, use money as an amplifier to compound. Don't make bets on things you can't control. I built my life as a sovereign professional where I control my time and I do work I enjoy and I earn money far in excess of what I need. And it was using that exact process. And it's exactly what I would do again if I had to start from zero. It works in any economic environment, of course, save, you know, total world collapse. And if that happens, then I'm all bets are off. And it compounds in a way that will fuel your life into perpetuity. Start by building those skills and then learn how to sell them in a way that gives you income far in excess of your needs through consulting, through productized services, through content. There's a variety of ways to do this. And do it in a way where you control your hours so that you work in a way that suits you and not in excess, and that will provide far more life returns than some windfall bet that you keep waiting for.
So then I would ask you, what work would you do if you knew that you couldn't fail? What's like the one part of your job that you wish was the whole part of your job? What are the skills you have that people talk to you about, that friends ask you about, that are your unique moat? That is where both your long-term protection and your big upside lies. If you want to check out additional resources to help with the above, I'll leave links in the description. To understand more about these IPOs, I'll link that video next. If you want to learn more about the leverage stack, see the "I'm 43" video. See you in the next one.