Many fear, while others embrace, the stock market because they believe that it is essentially a national gambling casino. Instead of betting on whether red or black will come up at roulette, investors wager on whether a stock’s price is about to go up or down.

Nearly a century ago, John Burr Williams wrote a PhD thesis at Harvard titled The Theory of Investment Value — not “a theory,” but “the theory.” Despite the pretentious title, his thesis was based on common sense. When dairy farmers buy cows, they compare the price of cows to the value of the milk the cows produce. Egg farmers compare the price of hens to the value of the eggs they produce. Beekeepers compare the price of bees to the value of the honey they produce. The intrinsic value of cows, hens, and bees is the value of the milk, eggs, and honey they produce.

In the same way, the intrinsic value of bonds, apartment buildings, and businesses is the value of the interest, rents, and profits they generate. It’s the same for stocks. The intrinsic value of a stock is the value of the income it yields. Value investors buy stocks not because they are wagering on the price going up but because the income is worth the price.

Sometimes, market prices stray far from intrinsic values. Indeed, the very first sentence of Williams’s treatise is, “Separate and distinct things not to be confused, as every thoughtful investor knows, are real worth and market price.”

The bicycle stock bubble

During speculative bubbles, market prices balloon far above intrinsic values. For example, a series of technological innovations in the late 1800s led to “safety bicycles” that had a diamond frame, a chain drive, two identical wheels, and inflatable tires. These were far safer, more comfortable, and less expensive than previous dandy horse, penny farthing, and boneshaker models.

Safety bicycles were wildly popular and generated substantial profits for bicycle manufacturers, whose stock prices rose along with their profits. But, as J.P. Morgan observed, “Nothing so undermines your financial judgement as the sight of your neighbor getting rich.” Rising bicycle-stock prices attracted the attention of speculators who hoped to profit from the soaring prices. This is called the Greater Fool Theory: pay a foolish price, hoping to sell to an even bigger fool who will pay a price that is even more foolish.

The Greater Fool Theory can be a self-fulfilling prophecy that works for a while — until the supply of greater fools is exhausted. Once the price stops rising, speculators no longer have any reason to buy and the bubble pops. During the bicycle-stock bubble, stock prices tripled in three months and then deflated like an inflatable tire that had run over a nail.

Lessons from past bubbles

There have been many such bubbles, usually starting with a great invention, discovery, or other compelling story that generates profits that lure in speculators, pumping up prices far beyond intrinsic values, and ending with a crash: Dutch tulip bulbs, South Sea Stock, railroads, radios, and more. More recently, we had the dot-com bubble in the late 1990s sparked by the creation of the Internet and the current AI bubble sparked by the introduction of ChatGPT on November 30, 2022.

During the dot-com bubble, new companies raised millions of dollars despite scant hopes of ever achieving profitability. One of my sons joined with four other new college graduates to start an Internet company. They had no business plan beyond a belief that they were smart and nimble enough to seize business opportunities that might arise. They were acquired a few months after the business’s formation, without their ever demonstrating their nimbleness.

Established companies doubled the price of their stocks, on average, simply by adding “.com” to the company’s name. At its peak, Yahoo sold for 2240 times earnings in 2000, a price that could only be justified if it was as profitable as Walmart in 2001, twice as profitable in 2002, three times as profitable in 2003, and so on indefinitely. When the bubble popped, the Internet-heavy NASDAQ fell by 75 percent from its March 2000 high over the next three years. Yahoo dropped by 95 percent.

Now we have the AI bubble

OpenAI, the creator of ChatGPT, was valued at $157 billion in a funding round last October and is now in a funding round that could see the company valued at up to $340 billion. How do these valuations compare to the profits OpenAI generates? Well, there aren’t any profits.

Image Credit: Thapana_Studio - Adobe Stock

OpenAI earned about $2.7 billion in revenue from 10 million paid subscribers and modest revenue from other sources in 2024. But, taking costs into account, it lost $5 billion. OpenAI is planning to double its subscription fees over the next five years but that won’t generate much revenue and it is also planning to increase its spending by even more.

The fundamental problem is that LLMs are not useful enough to yield much revenue. Dot-com prices were too high in the late 1990s but the Internet is actually useful for commerce, advertising, search, email, social media, video streaming, and more. Indeed, it is hard to imagine life today without the Internet.

By contrast, I can certainly imagine life without ChatGPT. The only thing I use it for is to demonstrate that it is not intelligent in any meaningful sense of the word. ChatGPT and other large language models (LLMs) are statistical text predictors — a marvelous feat that they accomplish without knowing the meaning of any of the texts they train on or predict. Even with extensive post-training to correct their incessant hallucinations, they are inherently unreliable when confronted with tasks that are novel or require critical thinking and subjective probabilities. (See here, here, and here, for example.)

The Apple executive in charge of Siri has described the delayed release of promised Apple Intelligence features as “ugly” and “embarrassing.” A likely explanation for the delay is that Apple has a well-earned reputation for delivering reliable products and they have been no more successful than OpenAI, Google, Microsoft and other companies in figuring out how to make LLMs dependably reliable. It’s not just Apple. There is understandably limited demand for unreliable products, especially if mistakes are expensive.

Unreliability isn’t the only problem. Some ill-intentioned people find LLMs useful for generating firehoses of disinformation. Others may find them useful for generating dodgy papers, computer code, and customer service when the perceived cost of mistakes is relatively small. But, even for those who find LLMs useful, there are many alternatives to ChatGPT that are just as good (or arguably better) and either free or very inexpensive. I cannot envision LLMs becoming as important as the Internet. I doubt that customers will pay the hundreds of billions of dollars needed to justify the resources being spent to create, develop, and maintain ChatGPT and other LLMS.

Yes, it is a bubble.