In the space of a single month, SpaceX lost $1.2 trillion in market capitalization. Tesla's operating margin collapsed to 1.4%. Nvidia lost roughly $1 trillion from its May peak. Meta's free cash flow dropped 91%. Reddit crashed 23% despite beating every earnings estimate. Amazon discovered its own engineers cannot track what AI costs. and Croup declared the Magnificent 7, the seven companies that accounted for roughly a third of the entire S&P 500, dead as a construct in part because they lost $2 trillion in a single month. When Wall Street declares a core financial construct dead, they don't really buy flowers and hold a funeral. They just quietly roll it up in a carpet, drop it into the river at 3:00 a.m., and just pretend they never really wrote a glowing 30page analyst report about it 6 weeks ago. Something is happening across the AI industry simultaneously, and it is not a coincidence. So today, I'm going to walk through the numbers, and there are a lot of them, so bear with me on this one. the financial structure connecting every single one of these companies to each other, the internal admission from the world's largest cloud provider that it cannot figure out what AI even costs, and whether or not this correction might actually suggest the bubble is starting to pop. Let's go through these one at a time because the scale of what's happening only becomes visible when you see it all together. First, Nvidia down approximately 18% from its May 14th record of $235.47, lost roughly $1 trillion in market value in under two months, dropped nearly 5% in a single session on July 28th amid what Yahoo Finance described as renewed worries of circular financing. Insiders sold $410.6 million in stock over three months. CEO Yansen Huang has publicly called the stocks decline a mystery he cannot reconcile with the fundamentals which is basically corporate speak for I built the most valuable semiconductor engine in human history and Wall Street is currently treating it like a volatile memecoin. Well, I wonder why Jansen the company that built the AI era is underperforming the market it created when the founder of a near 5 trillion company says the market and his business have stopped agreeing with each other. The warning could not be any clearer. Something is definitely going wrong. Next up, our favorite, Meta. Q2 profit fell 14%. Expenses up 55%. But are you ready for the big one? Here it comes. $2.4 billion in legal charges. Legal charges. $1.18 billion in severance from the big MetaMay layoffs. And free cash flow, which is the money left over after paying operating costs and capital expenditure. also legal fees dropped from $8.55 billion to $784 million. That is a 91% decline. While this was happening, Mark Zuckerberg published an op-ed in the Wall Street Journal, arguing that AI may soon deliver personal super intelligence to everyone, which is a very confident thing to announce. The same week your free cash flow drops 91%. It's kind of like standing on a sinking speedboat soaking wet, telling the passengers that you've nearly perfected telepathic waterskiing, right? E-arketer analyst Minda Smiley put it very directly, saying, "The optimistic positive tone he's striking stands in stark contrast to the negative sentiment that's building." On SpaceX, the largest IPO in human history, launched on June 12th at $135 per share. It peaked at $225.64. Now it's trading at roughly $113.50, well below its IPO price. The total market cap lost since the peak is approximately $1.2 trillion, which is almost exactly equal to the entire current market value of Tesla. SpaceX has lost a Tesla. That sentence shouldn't even be physically possible. And yet, here we are. I mean, losing an entire Fortune 500 company out of your back pocket in under 60 days is definitely some form of macroeconomic magic. I mean, there should be awards for this, or at the very least, some loss on Wall Street bets on Reddit. On the Tesla side, revenue was a record 28.24 billion, deliveries were a record 480,000 vehicles around that number, and operating income fell 57%. So, operating margin was 1.4%, free cash flow negative 1.09 09 billion capital expenditure up 142% and roughly 68.5% of Tesla's reported net income came from a paper gain on his SpaceX shares. So no cash, no actual products sold, just the stock price of one of Musk company temporarily inflating the earnings of another. Next quarter with SpaceX crashing, that paper gain of course reverses into a paper loss. record revenue, record deliveries, and somehow the company is barely breaking even. Electric summary was as follows. Here's what today really is. Investors finding out that Elon Musk's word isn't worth what they've been paying for.

Now, here we arrive at the most ridiculous one. Reddit Q2 revenue grew 61% to 185 million, beat every Wall Street estimate. The stock collapsed 23% anyway, the worst day since its IPO. And the reason for this was CEO Steve Huffman admitted that Google search referrals were choppy and traffic was more volatile later in the quarter. Google's AI overviews, you know what I'm talking about here. We've discussed this at length. Those automated summaries that answer questions before users ever click through to the actual website are eating everyone's traffic. USA Today's Google traffic fell nearly 50% between June 2025 and June 2026. Business Insiders drop over 85%. And the most poetically self-defeating detail in this entire story, Reddit licensed its data to Google in 2024 for $60 million per year so that Google could train its AI models. Google used that data to build the AI that now answers questions using Reddit's content without sending users to Reddit. Reddit literally sold the ammunition that's now being used to shoot it for $60 million a year. I'm not really even sure that there is an analogy for this because the situation is its own analogy. It's a bit like paying somebody to learn your recipe and then being surprised when they open a restaurant across the street using your own dishes.

In the space of a single month, SpaceX lost $1.2 trillion in market capitalization. Tesla's operating margin collapsed to 1.4%. Nvidia lost roughly $1 trillion from its May peak. Meta's free cash flow dropped 91%. Reddit crashed 23% despite beating every earnings estimate. Amazon discovered its own engineers cannot track what AI costs. and Croup declared the Magnificent 7, the seven companies that accounted for roughly a third of the entire S&P 500, dead as a construct in part because they lost $2 trillion in a single month. When Wall Street declares a core financial construct dead, they don't really buy flowers and hold a funeral. They just quietly roll it up in a carpet, drop it into the river at 3:00 a.m., and just pretend they never really wrote a glowing 30page analyst report about it 6 weeks ago. Something is happening across the AI industry simultaneously, and it is not a coincidence. So today, I'm going to walk through the numbers, and there are a lot of them, so bear with me on this one. the financial structure connecting every single one of these companies to each other, the internal admission from the world's largest cloud provider that it cannot figure out what AI even costs, and whether or not this correction might actually suggest the bubble is starting to pop. Let's go through these one at a time because the scale of what's happening only becomes visible when you see it all together. First, Nvidia down approximately 18% from its May 14th record of $235.47, lost roughly $1 trillion in market value in under two months, dropped nearly 5% in a single session on July 28th amid what Yahoo Finance described as renewed worries of circular financing. Insiders sold $410.6 million in stock over three months. CEO Yansen Huang has publicly called the stocks decline a mystery he cannot reconcile with the fundamentals which is basically corporate speak for I built the most valuable semiconductor engine in human history and Wall Street is currently treating it like a volatile memecoin. Well, I wonder why Jansen the company that built the AI era is underperforming the market it created when the founder of a near 5 trillion company says the market and his business have stopped agreeing with each other. The warning could not be any clearer. Something is definitely going wrong. Next up, our favorite, Meta. Q2 profit fell 14%. Expenses up 55%. But are you ready for the big one? Here it comes. $2.4 billion in legal charges. Legal charges. $1.18 billion in severance from the big MetaMay layoffs. And free cash flow, which is the money left over after paying operating costs and capital expenditure. also legal fees dropped from $8.55 billion to $784 million. That is a 91% decline. While this was happening, Mark Zuckerberg published an op-ed in the Wall Street Journal, arguing that AI may soon deliver personal super intelligence to everyone, which is a very confident thing to announce. The same week your free cash flow drops 91%. It's kind of like standing on a sinking speedboat soaking wet, telling the passengers that you've nearly perfected telepathic waterskiing, right? E-arketer analyst Minda Smiley put it very directly, saying, "The optimistic positive tone he's striking stands in stark contrast to the negative sentiment that's building." On SpaceX, the largest IPO in human history, launched on June 12th at $135 per share. It peaked at $225.64. Now it's trading at roughly $113.50, well below its IPO price. The total market cap lost since the peak is approximately $1.2 trillion, which is almost exactly equal to the entire current market value of Tesla. SpaceX has lost a Tesla. That sentence shouldn't even be physically possible. And yet, here we are. I mean, losing an entire Fortune 500 company out of your back pocket in under 60 days is definitely some form of macroeconomic magic. I mean, there should be awards for this, or at the very least, some loss on Wall Street bets on Reddit. On the Tesla side, revenue was a record 28.24 billion, deliveries were a record 480,000 vehicles around that number, and operating income fell 57%. So, operating margin was 1.4%, free cash flow negative 1.09 09 billion capital expenditure up 142% and roughly 68.5% of Tesla's reported net income came from a paper gain on his SpaceX shares. So no cash, no actual products sold, just the stock price of one of Musk company temporarily inflating the earnings of another. Next quarter with SpaceX crashing, that paper gain of course reverses into a paper loss. record revenue, record deliveries, and somehow the company is barely breaking even. Electric summary was as follows. Here's what today really is. Investors finding out that Elon Musk's word isn't worth what they've been paying for.

Now, here we arrive at the most ridiculous one. Reddit Q2 revenue grew 61% to 185 million, beat every Wall Street estimate. The stock collapsed 23% anyway, the worst day since its IPO. And the reason for this was CEO Steve Huffman admitted that Google search referrals were choppy and traffic was more volatile later in the quarter. Google's AI overviews, you know what I'm talking about here. We've discussed this at length. Those automated summaries that answer questions before users ever click through to the actual website are eating everyone's traffic. USA Today's Google traffic fell nearly 50% between June 2025 and June 2026. Business Insiders drop over 85%. And the most poetically self-defeating detail in this entire story, Reddit licensed its data to Google in 2024 for $60 million per year so that Google could train its AI models. Google used that data to build the AI that now answers questions using Reddit's content without sending users to Reddit. Reddit literally sold the ammunition that's now being used to shoot it for $60 million a year. I'm not really even sure that there is an analogy for this because the situation is its own analogy. It's a bit like paying somebody to learn your recipe and then being surprised when they open a restaurant across the street using your own dishes.

So the next logical question is of course why is all of this happening at the same time? And the simple answer, AI isn't making enough money is correct but also incomplete. The more interesting and more accurate answer involves a financial structure that connects every single company that we just discussed to every other one and it has a name corporate musical chairs where everyone is sitting on each other's laps also known as circular financing. The mechanics simplified are as follows. Essentially Nvidia makes the chips that power AI. The hyperscalers Microsoft, Amazon, Google Meta buy these chips to build data centers that they need. They pay for the chips using capital raised against projected future AI revenue. That AI revenue depends on customers and end users like us adopting AI products at scale. But the adoption hasn't matched the projections. So the hyperscalers buy more chips to build more capacity hoping the demand will arrive before the money runs out. NVIDIA books revenue, the hyperscalers book capital expenditure. Everyone's numbers look big, but the actual enduser demand, I'm talking real people here, real businesses paying real money for AI products that generate real returns, hasn't really scaled to fill the infrastructure being built for it. The money goes in a circle. Nvidia sells chips to Microsoft. Microsoft invests in OpenAI. Open AAI signs contracts with Oracle and Amazon. Oracle borrows to build data centers. Soft Bank funds OpenAI. OpenAI buys more compute from Microsoft. Microsoft buys more Nvidia chips. On and on it goes. Everyone books revenue from each other. Everyone reports growth, but the growth is substantially funded by other participants in the same loop. Nvidia itself is reportedly working on AI infrastructure deals potentially worth more than $750 billion, including financing arrangements that skeptics warn are artificially inflating demand and valuations across the whole industry. It's essentially financial alchemy of a sense. You lend your neighbors 50 bucks so he can buy your lawn mower, then claim your lawnmower business is experiencing unprecedented organic hypergrowth. As Jim Kramer noted on CNBC just last week, companies are putting money into money losing firms so those firms can order their trips. This has actually happened before. During the telecommunications boom of the late 1990s, equipment manufacturers like Cisco, Lucen Technologies, and Nortal Networks extended billions in loans to their own customers, cashstrapped internet service providers usually, and telecom carriers as well, so those customers could buy their networking equipment. Revenue looked incredible back then. Growth charts pointed only upward. And then between the year 2000 and 2003, 47 competitive local exchange carriers declared bankruptcy because they'd never generated enough real revenue to repay the loans. The equipment vendors were forced to write off billions. Cisco stock dropped 89% from its peak and as of 2026, 26 years later, has never recovered to its 2000 high despite growing its earnings sevenfold. I mean, imagine quadrupling your business output for a quarter of a century and then market still looks at you like you're that disgraced former child star. That number still catches me every single time. So, I had to mention it. Now, I want to be careful here. The AI circular financing structure is not identical to the telecom version. The hyperscalers today have vastly more cash and stronger underlying business than Lucent and North ever did. Microsoft, Amazon, and Google are not fly by night operations. They literally have real revenue, real products and real customers beyond AI. The comparison is more to give an illustration, not a prediction. But the structural similarity is documented, named and being discussed by serious analysts across the industry. A Colombia Business School analysis of AI circular financing warns that trouble propagates quickly through the ecosystem and that if OpenAI's monetization disappoints, the impact cascades to its hardware suppliers, cloud service providers, and ultimately to Nvidia's customer base. The risk, as Kramer put it, if the market decides it doesn't want to fund any more data centers and the companies themselves don't have the money or they don't get paid, then we've got a problem, which in my view is a very gentle way of basically saying the economic gravitational pull collapses and the entire galaxy goes dark all at once. Joke aside, because contrary to popular belief, this is a serious channel, this is actually a more complicated picture than any single YouTube video can fully capture. Truth is, of course, always complex and the contours of a $2 trillion correction involve details that would take a semester of graduate finance to fully unpack. But the broad shape is visible. A lot of money is moving in a circle or at least a recursive graph that would make Oiler proud. And the question really is whether real demand fills the circle before the music stops.

Now, we've described the circular graph, but my next question was, okay, but what's happening inside it? What does it actually look like when companies try to use this AI that they're spending trillions to build? Well, Amazon's internal metrics provide the clearest view. I did tell you they're not having an easy time either. According to the Financial Times, Amazon employees deployed one of Anthropic's Claude Sonnet models to perform a pretty menial task, matching author details with product listings on Amazon's e-commerce platform. The project was supposed to cost a fraction of what it actually ended up costing. The actual bill was 1.8 8 million, which is 860% over the allocated budget. And here's the part that should alarm anyone managing AI deployment at scale. It took 5 months for anyone to notice. 5 months of an automated algorithm, essentially running up a luxury penthouse tab on the company's credit card while everyone in management assumed it was just ordering office supplies. Of course, that was not an isolated incident. There was a second project building, and I really wish that I were making this up. A financial auditing tool ran $541,000 over budget. A financial auditing tool that couldn't audit its own finances. I mean, you really have to appreciate the profound poetic irony of building a digital accountant that immediately embarks on a half million unapproved spending spree. Right. There was also a third project, a logistics system designed to speed up deliveries, which added 134,000 in unplanned costs. together roughly $2.5 million in unplanned AI spending across just three projects. In an internal meeting, Amazon's own engineers described the situation in a phrase that I think captures the entire industry's predicament. Mistakes that used to be trivially cheap in traditional software have become catastrophically expensive when AI agents are involved. A senior employee told the Financial Times, "It's difficult to figure out how much anything related to AI costs." And this is Amazon we're talking about. This is the company that runs Amazon Web Services, the largest cloud infrastructure platform on this planet, the backbone of a significant percentage of the internet. If Amazon cannot track its own AI spending, the question is only how much worse is this everywhere else? Because other companies sure have the same problems. So why is this happening? What is going on technically? The reason is structural and is worth understanding because it explains a lot of the numbers in this video. When a human developer writes buggy code, the code crashes. An error message appears, somebody fixes it. The cost is non zero, but it is trivial. When an AI agent has a misconfigured task, it doesn't just crash, it keeps on running. And because most AI providers now bill based on tokens, which are the units of text the model processes, a misconfigured job just quietly keeps billing you. We're talking a retry loop that resends the same prompt. A job pointed at the entire product catalog instead of a sample. There is no crash. There is no error. Just an invoice that arrives a month later. That's how a $200,000 project becomes a $1.8 million project over 5 months without anyone noticing. The system did not crash. It just aggressively burned through capital with the cold, methodical efficiency of a microprocessing engine. And Amazon is not the only one discovering this. Uber's chief technology officer said back in April that Uber had already consumed its entire 2026 budget for cloud code. Its COO said the company had not yet established a clear connection between raising AI token usage and the number of useful features it actually produced. And of course these two examples are far from unique. Let me just reoff a couple of more. Microsoft canled most of its internal cloud code licenses after 6 months because the bills became untenable. This is OpenAI's biggest investor, by the way, telling its own engineers to stop using AI tools because of cost. Walmart built an internal AI assistant called Code Puppy, gave employees unlimited tokens, and had to cap usage within months after demand spiked far beyond budget projections. Axios reported that one unnamed company spent $500 million in a single month on clawed licenses because nobody had turned on a usage cap. Half a billion dollars. We're talking a single month because somebody forgot to flip a switch. There's stories like this a dime a dozen. I really don't want to bore you with a whole list, but I will refer to the academic literature very quickly who do full surveys for a proper conclusion on this one. There is a KPMG survey of 2,45 senior leaders across 20 countries that found that only 26% of organizations have realtime visibility into their AI spending. Another 22% discover costs only after the bills arrive and 29% of senior leaders admit they struggle to understand their own AI operating costs as developments scale. There is also a separate survey of 500 finance leaders across the US and the UK that found that 79% of enterprises had experienced AI cost overruns in the past year. 79%. That is a huge number. And I want to be very clear here. I am not advocating against using AI. I couldn't in good conscious as a computer scientist who uses it every day and loves what it can do. But when Microsoft cannot control AI costs, when Walmart has to ration its own tool, when a company can burn half a billion dollars in 30 days because nobody checked a settings page, this is not a technology-based problem. This is a deployment discipline problem. And the lesson is not don't use AI. The lesson is don't be the next Icorus. The sun is real. The wings are definitely real. The wax holding them together melts if you fly too close. And right now, a lot of companies are discovering the altitude at which the wax gives out. Prudence is valuable. Let's all use it. Now, I know what some of you are thinking. I can hear it forming in the comments already, and I'll just say it. It's a bubble. I will admit that the sentiment whispers like a sirens call. The numbers are staggering. The correction is visible. The circular financing structure looks like a textbook case. Everything is falling at the same time. The nifty50 precedent is ominous. If it looks like a bubble, sounds like a bubble, and loses $2 trillion like a bubble, surely it's a bubble. It's very tempting to diagnose it this way. It lets us put a nice, neat, historical bow on a $2 trillion fire. I covered this in depth in a previous video, and my position has not fundamentally changed. I understand why the word bubble appears obvious. Parts of this market may very well be exhibiting bubble conditions. Valuations have outrun earnings. Capital is circulating between companies whose growth depends on one another. Infrastructure is being built faster than proven demand can absorb it. None of that is imaginary. What I resist is treating bubble as though it settles the argument because the word collapses several different questions into one and we need more nuance. Are some AI companies grotesqually overvalued? Almost certainly. Has investment moved ahead of monetization? Clearly. Are circular financing arrangements disguising the weakness of enduser demand in some parts of the industry? Absolutely yes. But does that mean the underlying technology is fraudulent, commercially useless, or destined to disappear? That conclusion does not follow. The dotcom era was a bubble. It was also a period in which investors finance much of the infrastructure that would build the modern internet. Both statements can be true at once. The technology was real. The valuations were absurd. The investment was productive in aggregate and ruminous for many of the people who supply the capital. So perhaps the real disagreement is not over whether anything here is bubbly. It is over what precisely the bubble contains. The technology, the valuations, the financing structure, the scale of investment or the assumption that every company spending billions on AI will eventually earn those billions back. Howard Marx, founder of Oak Tree Capital Management, honestly that guy writes the best letters ever, argues that bubbles based on technological progress are actually beneficial because they excite investors into pouring in money, a good bit of which is thrown away to jumpstart a new area of opportunity. The hysteria spreads up what would otherwise be a long, slow process of development. And I think he's probably right. The dotcom bubble gave us the infrastructure for the modern internet. The railway mania of the 1840s gave Britain its rail network. The capital gets deployed. Much of it gets wasted. And the technology that survives the correction changes the world. But the wasted capital is not an abstraction. It is the retail investor who bought SpaceX at $225 and is now underwater. It is the pension fund holding Oracle bonds that may be downgraded to junk next week. It is the 8,000 Meta employees who were fired to fund AI capeex that produced a 91% decline in free cash flow. It is the Reddit communities whose content was sold to train the AI that is now replacing them. It is the Amazon engineer who deployed a $200,000 AI project that ballooned to $1.8 million while nobody was watching. The costs flow down. They always do. The technology survives, but the people who funded the correction with their jobs, their investments, their pension contributions, their content, they absorb the loss. This is why I don't like rooting for any bubble-like construct to just pop because the human cost is real. And the executives who made the decisions sail away on super yachts and write opeds about super intelligence while sipping handcrafted dirty martinis on international waters, reflecting deeply on how the struggle built their character. They would only get vodka martinis, by the way. They do not deserve the gin. Instead of calling it a bubble, let's try calling it a reckoning. Reckonings, unlike bubbles, don't pop and disappear. They restructure. The weak get absorbed or dissolved. The strong survive and eventually profit from the infrastructure the weak paid for. The technology survives, but the carelessness doesn't. and the bill for the circular financing for the catastrophic AI spending for the capacity built ahead of demand for the traffic stolen and the jobs eliminated and the trust eroded. That bill is arriving now for everyone all at the same time and it will be paid as it always is sadly disproportionately by the people who had the least say in how it was incurred. [snorts] If you want to understand how the world's greatest salesmen build a narrative that's now colliding with reality and why his fortune was cut in half in six weeks, I cover that in this video that I'm linking here on your screen. Thanks so much for watching. I'll see you in the next one.