Thanks to Robert McKay for making me aware of this via the dreaded Linked In
Fully concur with what’s said. It’s all swings and roundabouts They are all wrapped up in their own cult and we all suffer when the fuck it up. What’s new?
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:
LLRX
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:
The Imminent AI Bubble Crash (and Why It Won’t Matter in the Long Run)
