Sam Altman said the release of the GPT-5 was “completely ruined” at first. Altman then followed by a B-word at dinner with a reporter. “When bubbles occur, clever people get too excited about the core of the truth.” Barge We have reported on comments by Openai CEO. Then it was a massive MIT survey that put the numbers on what so many people seem to be feeling.
Tech sales continued as rattling investors reduced the S&P 500's value by $1 trillion. Given the increased advantage of that index with high-tech inventory, which has been converted to AI stocks, it was a nerve sign that the AI boom had changed to Dotcom Bubble 2.0. Certainly, the fear of AI trade has been proven as a hint of tear-repeated openness, even as a hint of openness from the Fed's chair to the September interest rate reduction set market, after Jerome Powell's semi-Dovish comment at WYOMING's Jackson Hole, as the S&P 500 snapped a five-day losing streak on Friday.
Gary Marcus has warned of the limitations of the largest language model (LLMS) since 2019, and warned of potential bubbles and problematic economics since 2023. His words have a particularly distinctive weight. Cognitive scientists have become AI researchers and have been active in the machine learning space since 2015 when they founded Geometric Intelligence. The company was acquired by Uber in 2016, and Marcus left shortly afterwards, providing voice criticism about what is considered a dead end in the AI space while working for other AI startups.
Still, Marcus doesn't see himself as “Cassandra.” luck In an interview. Cassandra, a Greek tragedy figure, was a character who uttered accurate prophecies, but was not believed until it was too late. “I consider myself a realist, foresee the problems and the right person about them.”
Marcus believes that the market wobble is more than anything else due to GPT-5. It wasn't a failure, he said, but it was “overwhelming” and “disappointing,” which “had really awakened a lot of people. The GPT-5 was basically sold like an AGI,” he added. “It's not a terrible model. It's not a bad model,” he said, but “it's not a quantum leap that led a lot of people to expect.”
Marcus said this is not news for those paying attention, as he claimed in 2022 that “deep learning is hitting a wall.” Certainly, Marcus openly wonders his subsack about when the generative AI bubble will fall. He said luck That “crowd psychology” is definitely happening, and he thinks about John Maynard Keynes' quote every day. “The market can stay irrational for longer than you can take the solvent,” or the Weil E. coyotes from Rooney Tunes hang out from the edge of the cliff, hanging in the air before falling to Earth.
“That's what I feel,” says Marcus. “We're off the cliff. This makes no sense. And there are some signs from the past few days that people are finally aware.”
Bubble Talk began to heat up in July when Torsten Sloak, the chief economist of Apollo Global Management, a widely read and influential Wall Street, issued an impressive calculation while not declaring the bubble. “The difference between the IT bubble of the 1990s and the AI bubble today is that today's top 10 S&P 500 companies are overrated than in the 1990s,” he writes, with companies like NVIDIA, Microsoft, Apple and Meta saying that the forward P/E ratio and incredible market capitalizations such as “bewildered by their turn.”
In the next few weeks, the disappointment of the GPT-5 was a significant development, but not the only development. Another warning sign is the enormous amount of spending on data centers to support all theoretical future demands for AI use. Slok also tackles this theme, finding that data centre investment contributions to GDP growth are the same as consumer spending in the first half of 2025. This is noteworthy as consumers make up 70% of GDP. (Wall Street JournalChristopher Mims provided the calculation a few weeks ago. ) Finally, on August 19th, former Google CEO Eric Schmidt co-authored what is widely discussed. New York Times On August 19, he argued that “it's uncertain how quickly artificial general information will be achieved.”
According to political scientist Henry Farrell, this is important in a critical aspect. Financial Times In January, Schmidt was a key voice forming the “new Washington consensus,” partially saying that the AGI is “around the corner.” In his material, Farrell says Schmidt's manipulation indicates that his previous set of assumptions are “visibly crumbling,” and warns that he relied on informal conversations with people he knew at the intersection of DC's foreign and technological policy. The title of the post is “Twilight of Technology.” He concluded: “If Aggie's bets are bad, much of the rationale for this consensus collapses, and that's the conclusion that Eric Schmidt appears to be coming.”
Finally, the atmosphere has shifted to AI rebound in the summer of 2025. Darrell West warned Brookings In May, both public and scientific opinion trends would soon violate the AI universe masters. immediately, First Company We predicted that summer would be full of “AI slops.” By early August, axios They identified Slang's “Clunker” as widely applied to AI accidents, especially when customer service became fraudulent.
John Thornhill's Financial Times It provided some perspective on the bubble question, advises readers to decorate themselves for crashes, but still advises them to prepare for the future “golden age” of AI. He highlights data center build-outs. This is part of a $750 billion investment from Big Tech over 2024 and 2025, and a global rollout projected to reach $3 trillion by 2029. Thornhill becomes a financial historian for comfort and perspective. Again and again, this type of enthusiastic investment usually causes bubbles, dramatic crashes and creative destruction, but ultimately results in durable value.
He points out that Carlota Perez has documented this pattern. Technological revolution and financial capital: bubbles and the dynamics of the golden age. She identified AI as the fifth technological revolution following a pattern that began in the late 18th century. As a result, the modern economy now includes railway infrastructure and personal computers. At some point, there was a bubble and a crash. Thornehill didn't quote him in this particular column, but Prime Minister Edward documented a similar pattern in his classics The devil takes the most afterA book worth noting that not only discussing bubbles, but also for predicting bubbles on dot com before it happens.
Owen Lamont of Acadian Asset Management cited the Prime Minister in November 2024. He claimed that the important bubble moment had passed. An extraordinarily large number of market participants claiming the prices are too high, claiming they are likely to rise even further.
Wall Street banks are barely seeking bubbles. Morgan Stanley recently released a memo, seeing great efficiency for businesses as a result of AI. $920 billion a year for the S&P 500. UBS agreed to the flagged warnings in the News Manufacturing MIT study. It warned investors to expect a period of “CAPEX indigestion” along with data center build-outs, but argued that adoption of AI is expanding beyond expectations, citing increased monetization from Openai's ChatGPT, Alphabet's Gemini and AI-powered CRM systems.
Bank of America Research wrote a note in early August before the GPT-5 was released. AI saw it as part of the “Sea Change” worker productivity that promotes the ongoing “innovation premium” of S&P 500 companies. Head of US equity strategy Savita Subramanian, insisted essentially that the inflation wave of the 2020s taught companies to do more in less, turn people into processes, and AI to turbocharge this. “I don't think it's necessarily a S&P 500 bubble,” she said. luck In the interview, “I think there are other areas that are a bit like a bubble.”
Subramanian refers to small businesses and potentially private lending as areas that may be “actively revalued.” She also notes that there is a huge risk of companies jumping into data centers, representing a shift towards an asset seafarer approach, instead of an asset lighting approach that increasingly distinguishes top performance in the US economy.
“I mean, this is new,” she said. “Technology used to be very light assets and spent money on R&D and innovation, but now we are spending money building these data centers,” she added, potentially marking the end of the light of assets, the end of the existence of high margins, essentially turning them into something “more property and manufacturing than before.” From her perspective, it guarantees a low multiple in the stock market. When asked if it equals a bubble, she says, if not a revision, “it's starting to happen in places,” and she agrees to a comparison with the railway boom.
Gary Marcus also cites the Mathematics Basics as a reason for his concern, with nearly 500 AI unicorns being valued at $2.7 trillion. “That doesn't make sense compared to how much revenue you're coming. [in]He said. Marcus cited Openai, reporting $1 billion in revenue in July, but has yet to make a profit. He speculated that Openai has about half the AI market, providing rough calculations that meant about $25 billion a year for the sector. [invested]. ”
So, if Marcus is right, why have people not listened to him for years? He said he was warning people this For years, I have called the “Gap of Deceitability” in my 2019 book Restarting AI And discuss New Yorker In 2012, that deep learning was a ladder that didn't reach the moon. During the first 25 years of his career, Marcus trained and practiced as a cognitive scientist, learning about “anthropomorphized people do.” [they] Looking at these machines, they make the mistake of ascribe to them intelligence that is not actually there, humanity that is not actually there, and they use them as companions, and they think they are closer to solving these problems than they actually are. He said he believes that the bubble, which is expanding to its present level, is largely inflated because of what cognitive scientists are trained not to do because of the human impulse that human impulses project onto things.
These machines may look like humans, but “it doesn't actually work like you,” Marcus says, “this whole market imagines solving all of this because they don't really understand the problem, imagining that scaling will solve everything. That is almost tragic.”
Subramanian said she “feels like magic, so people love this AI technology. It feels a bit magical and mystical… The truth is, it hasn't changed the world much yet, but I don't think it's something to be dismissed.” She was also really taken by herself. “I'm already using ChatGpt over the kids. So it's kind of funny to see this. Right now I'm using ChatGpt for everything.”
This story was originally featured on Fortune.com.
