[source image]
Dow Jones NewsOct 24, 4:28 PM UTC
MW Here's one question about the AI bubble that even ChatGPT can't answer
By Jeffrey Funk and Gary Smith
Where's the profit? OpenAI and Big Tech can't really say.
AI disappointments in business have created is a new term: "AI slop," or low-quality, error-prone content.
Plenty of money but no earnings will cause the 'AI-slop' bubble to end badly.
The investment value of a stock comes from the cash flow it generates. Speculators, in contrast, care not a whit for cash flow; they are focused instead on future stock prices. This is the "greater fool" theory: Pay a foolish price with the hope that a greater fool will pay an even more foolish price.
A bubble happens when speculators push prices foolishly higher than justified by the cash flow. The bubble pops when speculators no longer believe that prices will keep rising - because there is no reason for them to buy if prices are not rising.
There is no reliable way to tell when a bubble will end. Driven by FOMO - fear of missing out - speculators may know they are in a bubble but keep speculating nonetheless. After James Milner, a member of the British Parliament, was bankrupted by the South Sea Bubble of 1720, he explained: "I said, indeed, that ruin must soon come upon us but ... it came two months sooner than I expected."
Shortly before the stock-market crash in October 1987, the Wall Street Journal ran a front-page story headlined, "Stock Market's Surge Is Puzzling Investors; When Will It End?" The story began with the exclamation "Wheee!" and went on to say that "the market madness is as puzzling as it is exhilarating." For a typical view, they quoted Steven Leuthold, the head of a financial advisory service: "You have to realize we're in Looney Tunes land and you should stay fairly close to the exits. ... [But] it's a lot of fun, and you could make a lot of money here."
Now we have the artificial-intelligence bubble, which we have written about extensively (for example, here and here). As with most bubbles - for example, railroads, radio, airplanes and the internet - the AI bubble began with an innovation that promised to be wildly profitable. In this case, it was the astonishing ability of OpenAI's ChatGPT and other large language models to generate convincing text in response to almost any prompt.
Wharton professor Ethan Mollick asserted that the productivity gains from LLMs might be larger than the gains from steam power. Sundar Pichai, CEO of Alphabet (GOOG) (GOOGL), proclaimed that LLMs are "more profound than fire." Turing Prize winner Geoffrey Hinton declared, "I think it's comparable in scale with the Industrial Revolution or electricity - or maybe the wheel."
In June 2023, OpenAI CEO Sam Altman said that businesses would be 30 times more productive and that "this, unlike other technologies, is a strong case of a technology that on the one hand, is the most exciting, most promising, coolest thing I think that humanity will have yet built. We can cure all diseases, we can get everybody a great education, better healthcare, massively increased productivity, huge scientific discovery."
In October 2024, Tesla (TSLA)CEO Elon Musk declared: "I certainly feel comfortable saying that it's getting 10 times better per year ... I think it will be able to do anything that any human can do possibly within the next year or two."
Last August, Altman boasted that interacting with GPT-5 "really feels like talking to an expert in any topic, like a Ph.D.-level expert."
The rampant mistakes and hallucinations that were initially made by LLMs have diminished over time due to pretraining on larger databases and post-training by tens of thousands of humans. However, the fundamental problem remains and cannot be solved by scaling or post-training: LLMs do not know how the text they input and output relates to the real world and consequently have no way of distinguishing between statements that are true and those that are false.
GPT-5's performance turned out to be disappointing, and many users concluded that it was not much better than GPT-4. Instead of the superexponential improvements Altman predicted, there have been sharply diminishing returns.
Show us the money
AI slop has also polluted the internet by flooding it with content that blurs lines between real and fake.
Many companies believed the initial hype and tried using LLMs to boost productivity and profits. They soon recognized that the reality did not match the hype. In August, an MIT study reported that 95% of generative-AI pilot business projects were failing. In October, another survey found that 96% of businesses "have not seen dramatic improvements in organizational efficiency, innovation, or work quality." What these disappointments have created is a new term: "AI slop," or low-quality, error-prone content.
AI slop in business reports, emails and meeting summaries reduces business productivity: A recent survey found that 40% of respondents had received AI slop at work in the previous month; that it takes nearly two hours, on average, to fix each instance of slop; and that they "no longer trust their AI-enabled peers, find them less creative, and find them less intelligent or capable."
AI slop has polluted the internet by flooding it with content that blurs lines between real and fake. It is estimated that AI-generated articles on the web now outnumber human-generated articles - which creates a feedback loop of LLMs training on their own slop and makes it increasingly dangerous to trust LLMs with tasks where mistakes have substantial costs, including lost revenue, lawsuits and reputational damage.
It is not surprising that the business demand for AI slop does not come close to covering the costs of creating it. Last year OpenAI had losses of $5.3 billion on revenues of $3.5 billion, and 2025 is just as bad or worse. In the first half of 2025, OpenAI reportedly had operating losses of $7.8 billion on revenues of $4.3 billion, and some outside estimates of losses are even higher if R&D and advertising are included. OpenAI forecasts $115 billion in cumulative losses by 2029, and this is likely to be an underestimate.
Circular financing is a thin lifeline
There were similar shenanigans during the dot-com boom.
Luckily for Altman, several big tech companies continue to support OpenAI. The most recent sleight of hand is circular financing. Nvidia (NVDA), AMD (AMD), Oracle (ORCL), Microsoft (MSFT) and Broadcom (AVGO) have given OpenAI a cash lifeline, with OpenAI expected to use the funding it receives to buy equipment and services from these companies.
There were similar shenanigans during the dot-com boom. Cisco Systems (CSCO), Lucent and other telecom-equipment makers loaned money to internet service providers so that they could buy equipment from them. The manufacturers reported strong equipment sales, but they were essentially buying from themselves.
When the bubble burst, the bad debts that were exposed meant that these companies had essentially given away their equipment for free. In other cases, one company would invest in another, with the money used by the second company to buy products from the first. Another variation was for two companies to buy products or services from each other: higher sales for both, but no profits for either.
AI's truth-telling pivot
OpenAI is certainly in a highly fragile situation. When the AI bubble pops, it will be one of the first casualties.
No wonder OpenAI and other companies are pivoting from their delusional claims about super-exponential increases in business productivity.
Generative AI can be used to create images and videos, and OpenAI recently launched Sora, a text-to-video system that allows users to create brief AI-generated video clips. All fake, all the time. This might be considered an improvement over content of unknown accuracy, but it is just more AI slop that reduces productivity and is unlikely to generate profits. Sora videos cost OpenAI about $5 of compute for each video, and users pay $20 a month for the ability to create 100 videos every 24 hours. It is hard to imagine that compute costs will drop below the amount users are willing to pay for this slop.
This past spring, OpenAI offered students two months of free access to ChatGPT, evidently hoping to lure them into buying paid subscriptions that will help them cheat on their homework, papers and tests. This might generate revenue, but it will surely reduce the students' productivity when they enter the workforce.
Altman also says that GPT can be used as an AI buddy that offers advice (and companionship), and it is reported that OpenAI is working on a portable, screen-free "personal life adviser." GPT might well be popular as a life adviser, since accuracy is of little importance. Lawsuits arising from bad advice will just be part of the cost of doing business.
These side hustles are unlikely to generate the increases in productivity and profits promised by LLM hypesters. When the bubble pops, the fallout will be a bit different from the popping of the dot-com bubble, when a large number of shaky companies went bankrupt. OpenAI is certainly in a highly fragile situation. When the AI bubble pops, it will be one of the first casualties.
Many of the other players - Microsoft, Alphabet, Meta Platforms (META) - have ample revenue from other sources. However, to the extent their current lofty valuations are based on outsize projections of LLM revenue, their stock prices will tumble and so will consumer and investment spending. The AI-slop bubble will end badly.
Jeffrey Funk is a retired professor, tech consultant and the author of "Unicorns, Hype and Bubbles: A Guide To Spotting, Avoiding, and Exploiting Investment Bubbles in Tech."
Gary Smith is the author of more than 100 academic papers and 20 books, including "Standard Deviations: The Truth About Flawed Statistics, AI and Big Data" (Duckworth, 2024) and (co-authored with Margaret Smith) "The Power of Modern Value Investing: Beyond Indexing, Algos and Alpha" (PalgraveMacmillan, 2024).
24 Oct 2025, 4:28 MW Here's one question about the AI bubble that -2-
More: These stocks are the real deal for investors in AI - Wall Street is just chasing bubbles
Also read: AI stocks are in a bubble. Why are so many investors refusing to believe it?
-Jeffrey Funk -Gary Smith
This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
10-24-25 1228ET
Copyright © 2025 Dow Jones & Company, Inc.
