Stock market frenzy is a louder echo of the 1990s

Throughout the late 1990s Fred Hickey, the editor of The High-Tech Strategist, was relentlessly critical of the wild market excesses of the time. So-called "bubbleheads" responded by leaving vile messages on his voicemail. However, he developed a cult following among bearish investors. Although Hickey was eventually vindicated, he turned out to be wrong in one respect. In December 1999 he wrote: "The lunacy of tech stock valuations is beyond description ... I'm certain I'll never see it again in my lifetime." Hickey still publishes his monthly investment newsletter. He finds much to remind him about the earlier dotcom mania. Except this time, he believes the market is even crazier.

Hickey has been writing his monthly publication since 1987. It is a discreet affair. It has no website, the letter goes out by snail mail or PDF, and its masthead has not changed in a quarter of a century. Hickey works from his home in a leafy suburb of Nashua, New Hampshire, where he peruses industry journals, listens in on quarterly earnings calls (assisted by his son Ryan) and corresponds with his contacts in the tech world. He believes it's a huge advantage to work far from the pressures of Wall Street and the "wild optimism (delusion)" of Silicon Valley. He sees himself as the little boy in the Hans Christian Andersen tale who announces what no one else dares: that the emperor's new clothes do not exist.

Today, Hickey observes speculative excesses across the investment world: an options market dominated by zero-day-to-expiry options that provide leverage for day traders; investors conditioned by years of support from the Federal Reserve to buy into every dip, giving the market an aura of invincibility; spoof cryptocurrencies serving as gambling chips for nihilistic speculators; margin debt far above its 2021 peak; and mutual funds' cash levels at record lows. "I've seen it all before during the height of the 1999/00 dotcom mania," writes a weary Hickey.

The excitement surrounding artificial intelligence overshadows everything else. Hickey points to a recent JPMorgan research report showing that 41 AI-related stocks account for nearly half of the market capitalisation of the S&P 500 Index. Bulls claim that market exuberance is not as intense as in the late 1990s because there have been relatively few initial public offerings. Hickey counters that startups no longer need to rush to the market to raise capital. Companies like OpenAI have been able to attract vast amounts of cash while avoiding the scrutiny that comes with a public listing. If the $500 billion OpenAI and other private AI firms are added to JPMorgan's list the total capitalisation of AI-related businesses in the U.S. is far larger than the technology, media and telecoms sectors at the turn of the century.

Though Hickey questioned the crazy valuations during the dotcom era he never doubted that the internet was truly a revolutionary technology which was going to change the world. By contrast, today he's extremely sceptical about the future impact of AI. He distinguishes between artificial intelligence, a technology that has been around for decades, and generative AI, which is at the epicentre of the current frenzy. "GenAI is probably the most over-hyped technology I've ever witnessed in my 45 years of following tech stocks," he writes. We are told that GenAI will shortly be in a position to cure cancer, write Shakespeare, win Nobel prizes, and drive soaring productivity growth. In short, as Meta's Mark Zuckerberg says, it may be "the beginning of a new era for humanity."

Hickey does not buy into this narrative. He points to the profound limitations of large language models (LLMs): chatbots are unable to reason, have no ties to the real world and are incapable of adapting to change. MIT recently tested the performance of a variety of chatbots against humans, and humans crushed every model. Hickey has been waiting in vain for a killer app that would prove him wrong. Instead, attempts to integrate AI into smartphones have flopped. While the public's use of chatbots has taken off, relatively few users are prepared to pay for the service.

The trouble is that LLMs are prone to errors – so-called "hallucinations" – which vitiate their commercial application. Proponents of AI believe that bigger models and more computational power will overcome these problems and lead to the holy grail of superintelligence. Hickey dismisses such predictions. While successive generations of early ChatGPT models delivered great advances, recent product launches have disappointed. OpenAI's GPT-5 showed little improvement on its predecessor, despite huge investment.

At the outset of every technology revolution there have been naysayers. The economist Paul Krugman, a Nobel laureate, once argued that the internet was of no greater economic significance than the fax machine. Why should anyone today pay attention to the rantings of a veteran tech analyst from the back of beyond?

The best reason is that some very eminent computer scientists make the same criticisms about GenAI. Yann LeCun, chief AI scientist at Meta Platforms and a winner of the Turing Award for computing, observes that although LLMs contain the entire corpus of human knowledge they have yet to make a single discovery. That's because humans think in terms of mental models whereas GenAI is only capable of "regurgitating" information, LeCun says.

Another Turing Award winner, the Canadian computer scientist Richard Sutton, claims that LLMs are incapable of learning from experience. They lack a goal which, Sutton believes, is the essence of intelligence. Gary Marcus, a neuroscientist and founder of two AI startups, says that chatbots aren't intelligent but perform "neat party tricks." These eminent AI sceptics all agree that building bigger models with greater amounts of data and compute won't achieve superintelligence or even get rid of those pesky hallucinations.

Hickey believes that the inherent shortcomings of GenAI will eventually be recognised. When that day arrives an enormous overbuild of computer servers and storage will become evident and tech profits and valuations will collapse – just as they did 25 years ago.