Are We In An AI Bubble?

It's Friday, July thirty first. On Wednesday, the Dow dropped over 1,100 points, tech stocks got crushed, and South Korea's stock exchange triggered circuit breakers twice. The word I kept seeing everywhere was "bubble." AI bubble. Tech bubble. Market bubble. Then on Thursday, the market bounced back and everyone relaxed, which is actually the more interesting thing to pay attention to, because that cycle of panic followed by relief followed by "see, everything's fine" is exactly what happens inside a bubble before it pops. The drops feel scary for a day, the recovery makes everyone feel smart again, and the pattern repeats until the one time it doesn't recover. But I realized that even though I keep hearing the word "bubble," I wasn't sure I could actually define what one IS versus just a market that went up a lot and then came down. Because markets go up and down all the time, and not every drop is a bubble popping. So what makes something a bubble specifically, how do they form, and is there any way to know you're inside one before it's too late? The answers go back centuries and they follow a pattern that's so consistent it's almost eerie, because the same psychology that inflated tulip prices in 1637 Holland is the same psychology driving AI stock prices in 2026 America, and the people inside each bubble were equally convinced that their situation was different. A bubble isn't just when prices go up a lot. Prices can go up a lot for perfectly good reasons: a company invents something transformative, demand increases, revenue grows, and the stock price reflects that real value. That's not a bubble. That's a market working the way it's supposed to.

A bubble is when the reason people are buying shifts from "I think this is worth the price" to "I think someone else will pay more than I did." That shift is everything. Before it, you're investing. After it, you're speculating, and the only thing supporting the price is the belief that there's always someone willing to pay more. Economists call this the greater fool theory: you knowingly buy something at a price you suspect is too high because you're confident you can sell it to a "greater fool" at an even higher price. The game works until it doesn't, and when it stops working, everyone who bought late is holding an asset that's worth a fraction of what they paid for it.

The tricky part is that the shift from investing to speculating happens gradually and it's almost impossible to see clearly while you're inside it. Early buyers have legitimate reasons to be excited, and the gains in the early phase are often justified by actual growth. But as prices rise and the stories get bigger, new buyers enter who aren't evaluating the fundamentals at all. They're buying because prices went up last month and they assume prices will go up next month. The buying itself becomes the reason for more buying, which creates a feedback loop that detaches the price from any reasonable estimate of what the thing is actually worth. That feedback loop is the bubble. An economist named Hyman Minsky mapped out the stages of a bubble in the 1980s, and the pattern has repeated so consistently across centuries and asset classes that it's become one of the most referenced frameworks in finance.

It starts with displacement, some real innovation or shift that creates a real opportunity. For tulips in 1630s Holland, it was the introduction of exotic varieties that wealthy collectors valued. For the dot-com era, it was the internet, which really was transformative. For housing in the 2000s, it was financial products that made mortgages accessible to millions of new buyers. For AI right now, it's a technology that is reshaping how businesses operate in measurable ways. The displacement is what makes every bubble convincing, because the opportunity that starts it is never imaginary.

Then comes the boom. Prices rise, early investors make money, and the success stories start spreading. The media covers it, and more people start paying attention. This phase can last years and the gains during it are often legitimate, which makes it even harder to separate the real growth from the speculation building on top of it.

The boom tips into euphoria when new investors start entering not because they've evaluated what they're buying but because they're afraid of missing out. This is where you start hearing phrases like "this time is different" or "the old rules don't apply" or "we're in a new paradigm." William Bernstein, who wrote The Delusions of Crowds, identified a few signs that you've entered the euphoria phase: people quit stable jobs to pursue the opportunity full-time, people get angry or dismissive when you express skepticism, extreme predictions become normal ("this stock will 10x by next year"), and the thing being traded becomes a common topic of conversation among people who have never previously shown interest in investing.

After euphoria comes profit-taking, where the institutional investors and early movers start quietly selling while the narrative is still positive. This is what Goldman Sachs reported hedge funds doing over the past two months: selling US tech stocks at the fastest pace in over a decade while the broader public was still buying. And then comes panic, where the price drops fast enough that the feedback loop reverses. Falling prices cause fear, fear causes selling, selling causes more falling prices, and the people who bought during the euphoria phase discover that there's no greater fool left to sell to. So is AI a bubble? The honest answer is that nobody knows, and anyone who tells you they know for certain is either lying or selling something.

Here's what makes the AI question so difficult: the dot-com bubble popped in 2000 and wiped out hundreds of companies, but Amazon, Google, and eBay survived it and became some of the most valuable companies in history. The bubble was in the pricing, not the product. The people who said "the internet will change everything" in 1999 were right about the technology and wrong about which companies would survive to deliver on the promise. Most of the companies that people were pouring money into had no revenue, no business model, and no path to profitability, but the few that did became the backbone of the modern economy.

AI looks similar in structure. The question isn't whether AI matters but whether the prices being paid for AI-related stocks reflect the actual revenue and profits those companies will generate, or whether the prices reflect a greater fool assumption that someone will always pay more. Nvidia trading at enormous multiples of its earnings, companies adding "AI" to their name and watching their stock jump, startups raising billions without proven revenue models: all of these are patterns that appeared in the dot-com era too.

The uncomfortable truth about bubbles is that you usually can't confirm you were in one until after it pops. During the euphoria phase, the people warning about a bubble look like pessimists who are missing the opportunity of a lifetime. After the crash, those same people look like geniuses. And the best illustration of how this works in real time is happening right now, this week, to a 25-year-old former OpenAI researcher named Leopold Aschenbrenner (ASH-en-bren-er).

Aschenbrenner left OpenAI in 2024 after being fired for raising internal security concerns and published a widely read essay called "Situational Awareness" arguing that AI scaling laws were so predictable that AGI could arrive within a few years and that massive investment in AI infrastructure, chips, memory, data centers, energy, was the obvious trade. He launched a hedge fund with the same name, started with roughly $225 million, and used about four times leverage to bet heavily on AI infrastructure stocks while shorting software companies. Through June 2026, the fund had grown to as much as $45 billion with returns over 400 percent. He looked like a genius.

Then July happened. His AI infrastructure longs dropped 40 to 50 percent in a few weeks while his software shorts rallied against him. The leverage that amplified his gains on the way up amplified his losses on the way down, and his prime brokers, Goldman Sachs, JPMorgan, and Bank of America, started issuing margin calls the fund couldn't meet. Yesterday, Ken Griffin's Citadel bought Aschenbrenner's entire public portfolio in a single block trade before the market opened. The fund lost roughly $35 billion in value in less than a month. And here's the detail that captures the cruelty of leverage in a bubble: his positions bottomed the day before the forced sale and bounced sharply the day Citadel absorbed the block. He wasn't wrong about AI. He was leveraged and early, and the market punishes that identically to being wrong.

Buffett, meanwhile, is sitting on $397 billion in cash, the largest position in Berkshire Hathaway's history, watching from the sideline. Aschenbrenner bet his career and his fund that AI infrastructure was the trade of the decade. Buffett is betting that whatever is happening right now isn't worth buying yet. One of them used four times leverage and lost $35 billion in a month. The other used zero leverage and is waiting. Both believe they're right about the future. The market only cares about who survives long enough to find out. The thing I keep coming back to is Bernstein's observation that bubbles last long enough to make skeptics feel foolish for not participating, which is exactly what pulls the last wave of buyers in right before the top. The pattern has played out with tulips, railroads, dot-com stocks, housing, crypto, and now potentially AI, and in every single case the people inside it were convinced their situation was different. The bubble was never in the technology. It was in the pricing, in the distance between what something is worth and what people are willing to pay for it. A 25-year-old turned $225 million into $45 billion betting on that distance and then lost $35 billion in six days when it snapped back. Whether that distance exists in AI right now is the $397 billion question that Buffett is quietly answering by doing nothing.

Stay informed, stay curious, and we'll see you tomorrow.