A London-based research analyst is making one of the most sweeping bear cases yet on artificial intelligence. Julien Garran of MacroStrategy Partnership argues in a report circulated in October 2025 that the AI investment boom is 17 times the size of the dot-com bubble and four times the size of the housing bubble that preceded the 2008 financial crisis, according to coverage by CNN and MarketWatch.
The claim arrives at a moment of apparent contradiction. Talk of an “AI bubble” has become routine among analysts, yet markets have continued to climb, and the largest AI startups have added hundreds of billions of dollars in private valuation over the past year while most still report losses. Investors, in other words, are still paying for a future that has not yet arrived.
The case for a misallocation of capital
Garran’s central argument, as reported by CNN, is that the United States is experiencing a historic misallocation of capital. He measures the bubble not by stock prices alone but by the gap between the money flowing into AI infrastructure and the economic return it is producing, and on that measure he concludes the current frenzy dwarfs the late 1990s.
His skepticism starts with the technology itself. Garran describes large language models as statistical systems that predict the next word rather than reason or create. In his view that makes them impressive text-prediction tools but a weak foundation for durable commercial applications, and he doubts they will deliver the transformational economic gains their backers promise.
He also questions the scaling strategy the industry has relied on. Each new generation of models requires more data, more electricity and more servers, Garran argues, while the improvements in capability have grown smaller. He contends that no model released since OpenAI’s GPT-4 in 2023 has represented a comparable leap forward, a view that many AI researchers dispute.
Profits concentrated in one company
Garran points to the uneven economics of the AI ecosystem. Nvidia, whose chips power most AI training and inference, is highly profitable because nearly every participant needs its hardware. Data center operators, software developers and model builders, by contrast, are largely running at a loss and depend on continuous fundraising to keep operating, which he likens to an unending funding tour.
That dependence is the vulnerability. Garran argues venture capital firms are showing signs of fatigue as valuations climb, leaving a shrinking pool of deep-pocketed backers such as SoftBank and sovereign investors including Saudi Arabia. If those sources pull back, he says, the structure could unwind quickly.
Timing is the open question
Garran acknowledges that he may be early. Markets remain near record highs, and he has said the boom could persist for several more years. His concern is that a prolonged mania would divert time and resources into projects with limited social or economic value, slowing rather than accelerating progress.
The counterargument is the one his critics raise most often: the dot-com bubble also destroyed many companies, but it produced Amazon and Google. Even if Garran is right that spending is far ahead of returns, the technology could still mature into something valuable. And if AI developers achieve the kind of general intelligence they say they are pursuing, the economic calculus would change entirely.
For now, the debate turns on whether AI is as powerful as its valuations imply, or whether the market is pricing in a revolution whose foundations are still being laid.
