Transcription
Would the .com bubble even be that bad by today's standards? That sounds like a dumb question to ask about what is widely recognized as the single largest speculative market crash in recent history. But there is also a lot of misunderstanding about what the dot bubble actually was, the lasting impact it had, who it affected, and perhaps most importantly, who it didn't.
Of course, there are the usual surface-level talking points about historically stretched valuations, circular financing deals, overly optimistic projections about unfounded technologies, a massive and ultimately unsustainable amount of infrastructure spending to support everything. All held together by a level of irrational exuberance that looks almost comical in hindsight. And yes, these historic anecdotes have once again come into focus for, um, some reason. But there are also a lot of details about the dot crash that tend to get ignored in favor of just pointing and laughing at everybody for being so stupid, even if it was partially deserved.
For starters, a lot of people recognized it was a bubble. It wasn't like the entire world was fully on board, and there were plenty of naysayers in the media, finance, and even within the tech sector itself. The .com crash was also really just two separate crashes, which both impacted very different corners of the market. And well, yeah, if 2001 really is going to be a cautionary tale about speculative manias, it's also worth understanding how it could have been a lot worse.
For how bad it was, the dot-com bubble was largely isolated to a select group of companies. It didn't involve that much debt. It was relatively easy for regular investors to avoid completely. The infrastructure that developed alongside it remained relevant and valuable even after the hype died down. And more broadly, the rest of the real economy was still relatively strong. Employment was high, debt was low, inflation was stable, and most people were still doing very well. The bursting bubble certainly wasn't a walk in the park. But if, you know, hypothetically, the entire economy was dependent on this line going up with nothing else to support it, then things very easily could have been a lot worse. Which also raises perhaps the most important unanswered question of all: If we all knew this couldn't go on forever, what did it take to finally actually cause the pop?
>> Tonight, the information superhighway and one of its main thoroughfares, an online network called internet. Every business, no matter how large and no matter how small, will be on the internet.
>> Any medium like this creates an opportunity for new ways to sell and distribute products. And I think the web will be as important to retail distribution as television is.
>> I mean, you've got millions of square feet now of real estate. You've got, uh, a growing, huge and growing inventory of items that you keep in stock, and you've got thousands and thousands of employees now.
>> I brought something for you from pets.com. I'm here to play with the tabby cat. I'm not going to tell you what it is, but it makes you crazy.
>> Well, it's very hip to be on the internet right now.
>> Other economic news tonight. There is a new victim of the falling fortunes of the new economy. Pets.com is closing down.
Before a very brief summary of the run-up to the dot-com bubble, it helps to remember that the internet itself had been around for a long time before anybody had heard of pets.com. For about three decades, it was a tool for academics, military researchers, and a small community of hobbyists with the patience and very specific expertise to wrangle command-line interfaces and Unix manuals just to read a paragraph of text on somebody else's computer. The web that you actually recognize today is a much more recent invention.
In 1993, a small team at the National Center for Supercomputing Applications released a browser called Mosaic, which for the first time put images and text together on the same page in a way that did not require a computer science degree to navigate. The same year, Congress passed the legislation that officially opened the internet to commercial use. And a small group of very excited people from both tech and finance backgrounds very quickly realized that this might be a slightly bigger deal than free electronic mail and bulletin boards for university students.
The first big jolt came on August 9th, 1995, when Netscape went public. The shares had been priced at $28 on the morning of the offering, double what the bankers had originally penciled in, and they closed the day at $58.25, 25% more than doubling again on the way. Wall Street had its first proper internet stock, and it had decided that the internet was going to be very, very profitable.
Less than six months later, in February of 1996, Congress passed the Telecommunications Act, which deregulated huge chunks of the industry and basically encouraged anybody with a lot of cash on their balance sheet to start laying fiber optic cable, even if telecom didn't have anything to do with their core business whatsoever. Companies like WorldCom, Global Crossing, and Lucent spent billions trenching cable across the country and across the ocean floor on the assumption that the internet was about to use all of it.
To get that money moving, a lot of these companies started doing something that should have been a very loud warning sign. They would lend money to their customers, who would then turn around and use that money to buy equipment from them. The same dollar would get counted as a sale on one side of the ledger and a loan on the other, which made revenue look like it was exploding when, in reality, it was just running in circles.
By 1998, the Federal Reserve cut interest rates 75 basis points to keep the spillover from the Asian financial market and the Long-Term Capital Management blow-up out of domestic markets, which was probably good emergency medicine for the global financial system. But it also poured a lot of cheap money on top of a stock market that was already running extremely hot. By 1999 and 2000, IPOs were routinely doubling or tripling on day one. Companies were buying Super Bowl ad slots before they had a viable business model. And a sock puppet was the most recognized face of online retail. At one point, a company could just add a dot to its name and watch its share price jump by double digits the next morning. Something that we obviously would never do again today.
Now, to be fair, it wasn't like the whole world was completely oblivious to what was going on. Plenty of serious people noticed the makings of a bubble forming and said it out loud. Federal Reserve Chairman Alan Greenspan used the phrase "irrational exuberance" in a speech all the way back in December of 1996. More than three years before the eventual peak, fund managers were warning their clients. Journalists were warning their readers. Even some of the executives running these companies were quietly cashing out and admitting in private that they did not believe their own valuations. The market just did not care.
And well, yeah, of course, there is no real point in being coy about it. Here we sit 25 years later, watching a new technology pull in eye-watering amounts of capital with a familiar cast of warnings, pundits, and very confident sock puppets. Learning from history is easy until there is a profit to be made by ignoring it. But the parallels are actually the least interesting part of this comparison. And if you actually pull apart what made the dot-com bubble survivable and containable, three of those things have either been weakened or completely reversed in the version we are living through right now. So, it's time to learn how history works and to find out if the .com crash would even be that bad by today's standards.
If you've ever tried to teach yourself something difficult, you know the hardest part is not having anyone there to walk through it with you. That is exactly the gap the new Brilliant just closed for me. The new Brilliant is a personal tutor for math and coding named Koji that sits right there on your screen with you. It can see what you are working on, draw diagrams, ask the right question at the right moment, and adapt in real time to how you think. I tried it on a problem I would normally have skipped. And instead of just handing me the answer, Koji walked me through it until the idea actually clicked. It does not make the hard stuff easy. It makes it make sense. Behind it is a world-class curriculum that runs from grade 5 through college and beyond. Built with experts from MIT, Harvard, and Stanford. You can go from fractions to algebra to calculus or learn to think like a real software engineer with a focus on logic and debugging. One-on-one tutoring has always been the most effective way to learn, but it costs hundreds of dollars an hour, and most families can never make it work. Brilliant puts that kind of tutor in front of you every single day with no scheduling and no waiting. Click the link below or scan the QR code on screen to get started with Brilliant Tutor for free. You can upgrade to premium to unlock every course. And right now, How History Works viewers can save 20% off an annual subscription at brilliant.org/howhistoryworks.
Okay, so the first concerning comparison is just the scale of spending involved. Across the peak years of 1997 to 2000, the entire venture capital industry raised somewhere in the neighborhood of $344 billion in today's money to fund .com businesses. It was a generational amount of risk capital chasing one corner of the economy. But in 2025, AI startups alone hoovered up around $222 billion of US venture capital in a single year. Roughly 65% of every venture dollar invested in the country. And that is before you count the infrastructure spending sitting on top of it. The big four hyperscalers, Amazon, Microsoft, Google, and Meta, are all expected to spend somewhere north of $600 billion on capital expenditure in 2026 alone. The overwhelming majority of which is going to be AI infrastructure.
To put that in context, the entire fiber build-out of the late 1990s, while expensive in its time, looks like a rounding error next to a single year of modern AI capex. And once that fiber was in the ground, it was relatively cheap to keep running. A data center full of GPUs is the opposite. It needs constant power, constant cooling, and a constant replacement budget because the chips inside it become obsolete every other year, if the claims of the company selling them are to be believed. So the running cost of AI is almost an order of magnitude greater than even after adjusting for inflation.
Now, circular financing has also returned, just with extra steps. For example, Nvidia invests in OpenAI. OpenAI agrees to buy Nvidia chips. Nvidia books the revenue. The revenue justifies the valuation. The valuation funds more deals. It is the same trick the telecom companies were running in 1999, just dressed up in nicer fonts.
But the most important difference is in how all of this is being funded. In the dot-com era, most of the mania was financed by selling shares to people who at least understood that they were buying an asset with some risk. They might not have appreciated quite how thin the odds were, but the risk lived in the equity with people who knew, or at least should have known, they could lose every cent of it. Today, the funding stack is very different. OpenAI alone has committed to spending around $1.4 trillion on data centers and compute over the next eight years, the bulk of it underwritten by debt. Morgan Stanley estimates that the entire industry will spend somewhere around $3 trillion on data centers between 2025 and 2028, of which roughly $1.5 trillion has to come from external financing rather than the cash flow of the companies doing the buying. The single biggest contributor to filling that gap is private credit, on the hook for around $800 billion of it, which, if you have been paying attention to that corner of the financial system over the last 12 months, already has its own problems.
And stack on top of all of that the fact that every dollar a bank lends to a hyperscaler to build a data center is a dollar it does not lend to a small business, a family taking out a mortgage, or a power utility trying to upgrade the grid. So the bill for this bubble can quite plausibly land on people who never so much as opened a brokerage account.
The second problem is that this time around, the bubble is genuinely unavoidable. The dot-com bubble was, in retrospect, surprisingly easy to opt out of. If you thought it was insane, you could just, you know, not buy any of it. Most of the mania was attached to brand-new companies that had only just listed. And back then, index investing was still a niche product, not the default way Americans saved for retirement. If you stuck with broad index funds in 1999, you still got hit when the NASDAQ rolled over. But the damage was much smaller than if you had been picking individual internet stocks.
Today, the structure of the market is fundamentally different. The biggest companies in the United States, the ones that make up the largest weights in the largest index funds and most pension portfolios, are also the ones with the most direct exposure to the AI market. Meta, Microsoft, Google, Amazon, Nvidia, Apple, and Tesla have all bet some material chunk of their business on the AI trade. In plain English, back in 1999, you sort of had to opt into investments exposed to the dot-com bubble. A lot of people did, and there was genuine FOMO, but it was still a decision you had to make. Today, if you are investing in any capacity, you have to take very deliberate steps to opt out.
Now, to be fair, the size of these companies does offer some margin of safety in redundancy. The mega-caps holding up the AI market are also real businesses, unlike pets.com. With the exception of Tesla, they generate enormous amounts of cash from boring, established products. And even if every AI bet they have ever made gets written off, the underlying companies do not disappear. So a future correction would not necessarily be the kind of total wipeout that knocked out half the listed .com companies. But for the average passive investor, the line going down would still hurt a lot more than it did in 2000. The same is not true for OpenAI or Anthropic, which still need a steady drip of fresh investor checks just to keep the lights on. But at least for the next year, they will still be largely optional for private investors.
But well, yeah, the other half of the unavoidable problem is private markets. In the dot-com era, if you wanted to chase internet money, you basically had to go public. Private equity and venture markets were a fraction of their current size, and most companies that wanted real money had to ring the bell at the New York Stock Exchange, or more commonly, the NASDAQ. Today, the leaders in AI development have stayed private until they were close to a trillion dollars each. If and when they eventually go public, reportedly at some point this year, they are going to land in many indexes as enormous components on day one.
Now, investors losing their money is obviously not great, but at this point, it could be argued that they knew the risks, and if they are willing to take the upside, they should be prepared for the downside as well. This should also be remembered in the context that stock ownership today is even more concentrated than it was in the 1990s, with the overwhelming majority of equities being held by the top 10% of households who can generally speaking afford the loss.
But the third big difference is that this time, the bubble is most of what we have left. When the dot-com bubble was inflating, the rest of the American economy was actually in pretty good shape. The federal government was running a budget surplus. Household debt was sitting at manageable levels. The unemployment rate had drifted down to multi-decade lows, and labor force participation was within touching distance of its all-time peak. Inflation was well-behaved. Manufacturing was still a real chunk of the economy. And even outside of the very obvious mania in tech stocks, there was a lot of genuine productive economic activity going on. So when the bubble eventually popped, the rest of the country still had legs to stand on. The wealth that vaporized was, broadly speaking, paper wealth held by people who could afford to lose it. The 2001 recession, when it came, was actually one of the shortest and shallowest in modern American history.
Today, the picture is almost the inverse. The federal government is running deficits north of 6% of GDP in what is supposedly peacetime, with no real plan to bring them down. The richest 10% of households now account for roughly half of all consumer spending. And that spending depends almost entirely on how the stock market is feeling on any given Tuesday. Outside of the AI trade, productivity growth has been disappointing. Real wages have been stagnant for most workers. And a frankly silly share of recent corporate earnings has been concentrated in a handful of tech names. Strip those names out, and the rest of the market looks a lot more like a slow grind than a roaring boom.
So, if AI does correct, the damage is going to be a lot harder to contain to the people that accepted the risks. A serious downturn in the Magnificent 7 is a serious downturn in the wealth effect, which is a serious downturn in consumption from the people doing most of the spending, which is a serious downturn in the jobs that exist to service that spending, from restaurants to renovation contractors to private school administrators. None of those workers ever saw the upside from Nvidia, but their paychecks are ultimately exposed to the downside.
The infrastructure question is the other piece of this. The fiber that got laid down in the dot-com era was eventually very useful. By the mid-2000s, broadband adoption was exploding. Streaming was becoming a real thing, and a lot of that wasted cable suddenly had a job. There was an oversupply for a while, sure, but the underlying asset kept its value over decades. A GPU is a different animal. The most expensive chips in 2025 data centers are likely to be at the back of a depreciation schedule by the end of the decade, replaced by something faster, cheaper, and much less power-hungry. If demand for AI companies does not show up in the size everybody is currently underwriting, those buildings end up full of equipment with a use life measured in months, not decades.
And then there is the, uh, positive outlook. The internet, for all of its disruption, was largely a tool that companies were excited to hire for. Online retail, web design, search marketing, e-commerce operations – these became real careers and they pulled labor into new sectors faster than the old ones could shed it. The case for AI is almost the opposite. The selling proposition baked into pretty much every CEO earnings call you have heard in the last two years is that the technology is going to let companies do the same amount of work with fewer people. Whether the technology actually delivers on that promise or not, it is already being used as a very convenient excuse to cut headcount. "We are leveraging AI to cut down on expenses" is a much easier sell to investors than "We overhired during the years of low interest rates and now we are doing massive layoffs." So even before any popping happens, the AI bubble is doing damage to the same labor market that is supposed to absorb the shock when it does finally burst. That is a meaningful difference from a bubble that was, on the surface, a job creator.
So if all of this sounds a little bleak, the obvious next question is: What does it actually take to finally pop? The honest, slightly anticlimactic answer is that the market can stay irrational for an embarrassingly long time. And what eventually killed the dot-com hype was not particularly cinematic.
The first crack came right after the NASDAQ peaked in March of 2000. A wave of insider stock lockups – the rules that prevent founders and early employees from selling their shares – were usually six months after a listing started expiring all at the same time. The people who knew their companies best very calmly started cashing out in the public markets, and the supply of new shares started to overwhelm demand.
Around the same time, the Fed had begun raising interest rates again to lean against the bubble, which forced some investors to sell stocks to cover other commitments and shifted the flow of money into safer, higher-yielding assets. None of that was a death blow on its own, but it was a hiccup in an industry that was very dependent on a steady stream of fresh capital. After those first few drops, the early-stage venture markets got nervous. Funding rounds got harder to close. Terms got tighter. And a lot of companies that had been losing money on every transaction in the name of growth suddenly had no way to plug the hole. The companies started folding. And then more companies followed. And when everything dried up, the market discovered it had been dealing with several large pieces of accounting fraud the entire time.
WorldCom, the same telecom giant that had been laying all that overpriced fiber, turned out to have been booking ordinary operating expenses as capital investments to the tune of around $11 billion. Enron, which was technically an energy company but had gotten very enthusiastic about anything involving bandwidth, collapsed under the weight of off-balance-sheet partnerships designed to hide losses. Neither of those was a pure .com story, but they did dive into that space. And when the tide went out, it became a lot easier for the market to see who was swimming naked.
So, if you want to know what could plausibly do the same job for AI, you do not really need a dramatic catalyst. You just need a few quarters of disappointing capex spend, a few high-profile data center projects that quietly slow down or get restructured, and a private credit market that suddenly starts asking more questions about what exactly its loans are backed by. Whether that happens in six months or six years is anybody's guess. And frankly, anybody who tells you they know the answer is selling something.
What is much easier to predict is what happens after the .com bubble. Small enough, contained enough, and sat on top of an economy strong enough that we walked away from it with a few good jokes about sock puppets and a national fiber network we eventually figured out how to use. The version we are living through is bigger, more leveraged, more concentrated, and built on top of an economy that has very little spare capacity left for absorbing shocks. And when it does, maybe we will see who else was swimming naked in this market. But until then, go and watch this video next to find out how Enron got away with it for so long. And don't forget to like and subscribe to keep on learning how history works.