The intense excitement around artificial intelligence feels familiar. For anyone who remembers the start of the new millennium, it’s a clear echo of the dot-com bubble—a time when speculation about emerging technology was far removed from actual business fundamentals.
Back then, optimism about an internet-driven economy drove stock prices for companies like AOL to unbelievable heights. The problem was straightforward: revenue was speculative, profits nonexistent, and business models only half-developed. When reality hit, it hit hard. AOL Time Warner reported a staggering $98.7 billion annual loss in 2002, the largest in U.S. corporate history at the time, mainly due to writing down the value of its internet division.
There are plenty of other sad stories. The nation briefly mourned Pets.com, the online pet-supply store whose sock-puppet mascot charmed America even as the company burned through over $100 million, including a Super Bowl ad blitz. Nine months after its IPO, the lovable puppet was out of a job.
Fast-forward to today’s AI boom, and the sock puppet seems to be whispering from beyond the grave: “Be careful out there, kids.”
The parallels are hard to miss. The market is flooded with AI hype, yet very few companies—outside of key infrastructure players, especially Nvidia—are turning a profit. Many data-center operators, GPU manufacturers, and hyperscale cloud providers are in the black. But makers of generative AI applications, the companies training and deploying the large language models behind AI chatbots, are a different story. Credible profit-and-loss disclosures for these mostly private companies are rare, but based on the available evidence, few are profitable. As the Wall Street Journal explains, they are “sinkholes for AI losses that are the flip side of chunks of the public-company profits.” Most are losing money as fast as they can raise it, and plan to keep on doing so for years. Even OpenAI, one of the field’s most prominent companies (and one with powerful corporate patrons), estimates that it won’t become cash-flow positive until around 2029.
What is keeping these companies alive? Two factors: hope and speculative investment. More troubling is where much of this speculative money comes from—AI infrastructure companies themselves, including Microsoft and Nvidia, which have every reason to keep demand high. This circular capital cycle is not built for durability.
Justifications That Seldom Justify
When pressed to explain why these giant losses might work, company executives invariably attempt to justify them by arguing that it’s worth losing substantial sums now in the hope of locking in customers and gaining market share later. To be fair, this strategy did work fantastically well for a few companies, including Amazon, eBay, and a handful of others.
Those were the lucky few. Thousands of others suffered the fate of Pets.com, eToys.com, Webvan.com, Buy.com, Free-PC, Kosmo.com, Outpost.com, and MP3.com, and too many others to name. A September 22, 1999 New York Times article gives an idea of the zeitgeist of that era:
And the logic goes across every conceivable category of merchandise. You name it and someone has started a Web site to sell it: Autoweb, Furniture.com, Ebags, Garden.com, Netgrocer, Pets.com, even Partyjunk.com, scheduled to be introduced this fall, for children’s party favors. There are thousands more, and nearly all of them are planning to lose money for years.
Picking eventual losers then was a challenge few investors could meet. Most investors lost lots of money. Investing in AI app developers today feels more like speculation rather than investment. How many are willing to gamble against those odds?
The same AI developers often claim that AI’s rapid adoption (100 million users in months) is proof that “this time is different.” But there’s another way to interpret early mass adoption: if the bubble bursts, it could do so faster and deeper because expectations inflated so quickly.
Failing company founders can become quite creative in justifying bad business decisions. There are loads of other excuses for the losses, too many to include in this article. Some of the more popular are explained here: Debunking Dubious Justifications for the AI Bubble, but one of the newest is so dangerous that it must addressed here:
AI companies have recently suggested, both directly and indirectly, that their substantial losses might require public support or socialization, especially as the financial risks and capital expenditures in the sector have ballooned in late 2025. Notably, OpenAI’s chief financial officer mentioned the idea of needing “federal guarantees” to enable the scale of investment required for U.S. leadership in AI, later attempting to clarify that any “backstop” would mean significant government involvement—implicitly suggesting a safety net funded by taxpayers if things go wrong.
Socializing losses while privatizing profits creates giant moral hazards, but the current situation in Washington we should expectnd be ready to fight such efforts.
None of this means the AI revolution is fake. Far from it. Like the early internet, the underlying technology is transformative. But the first wave of companies rarely ends up being the ones that change history. A few pre-crash survivors—Amazon, eBay—went on to succeed. Still, most of the internet’s true, world-changing value was created after the bubble burst, by firms like Facebook (launched in 2004) and Google (whose now-dominant search engine was still in beta in 1999).
Cautionary Voices
In a recent interview, Bill Gates provided a clear reminder of how technological revolutions develop. Comparing today’s AI boom to the dot-com era, he
“In the end, something very profound happened… Some companies succeeded, but a lot of the companies were kind of [copy-cats]], fell behind, burning-capital companies. Absolutely, there are a ton of these investments that will be dead ends.”
Yet in the same breath, he called AI “the biggest technical thing ever in my lifetime.” The message: the technology is real, even if many of today’s investments are not.
Gates is not the only one with reservations about the short-term risks of AI investment. Other skeptics who share his view include insiders like OpenAI CEO Sam Altman, former Intel CEO Pat Gelsinger, OpenAI chairman Bret Taylor, C3.ai CEO Tom Siebel and Amazon founder Jeff Bezos.
One notable outlier is Nvidia CEO Jensen Huang, who told Bloomberg TV: “I don’t believe we’re in an AI bubble.” This isn’t exactly shocking, coming from the leader of one of the few companies consistently making money from AI.
It says something that investor legend Michael Burry–famous for predicting the 2008 subprime mortgage crash, as dramatized in The Big Short–has reportedly bet a billion dollars that Huang is wrong.
The signal is real, but the noise is deafening. And we are now in the phase where noise dominates.
1999 vs. Today
As Mark Twain reminded us, “History doesn’t repeat itself, but it does rhyme.” Consider the rhymes:
Foundational technology meets a speculative gold rush. The internet in the late 1990s; AI today. Both promised sweeping change—and attracted staggering capital before business models solidified.
Infrastructure built before monetization. Then: routers, switches, fiber, and PCs. Now: data centers, GPUs, and model-training clusters.
Copycat entrants are everywhere. In 1999, countless startups mainly set themselves apart by adding “.com” to their names. Today, many AI startups exist primarily to make pitch decks sound modern.
Valuations outrunning the math. In both eras, unprofitable firms reached multi-billion-dollar valuations on the strength of promises rather than profits.
Survivors are few; winners dominate. The dot-com crash didn’t kill the internet—it purified it. AI is likely to undergo the same winnowing.
Few informed observers today are suggesting this is pure mania for mania’s sake. This is not the equivalent of 17th-century tulip mania, when valuations had no connection at all to fundamentals.
The Long View
The genuine, lasting AI revolution will most likely happen after the shakeout—when the companies that survive (or those founded after the crash) develop sustainable models instead of depending on speculative capital injections. Exactly one hundred years ago, F. Scott Fitzgerald wrote: “The test of a first-rate intelligence is the ability to hold two opposed ideas in the mind at the same time and still retain the ability to function.” Today’s challenge is precisely that. We must accept two truths:
- There is an AI bubble.
- In the long term, AI will transform society.
Hedge fund icon Ray Dalio puts it succinctly: “There’s a major new technology that certainly will change the world and be successful, but some people are confusing that with the investments being successful.”
The question isn’t whether AI will reshape law, business, and society—it’s which firms, frameworks, and governance models will endure after the inevitable correction.
This is merely the opening act.
Credit: Thanks to Ahsan Nasar for his significant help in refining these ideas.