Steve Eisman warns of a potential AI collapse: the lesson from the 2001 bubble

Steve Eisman, the investor who accurately predicted the 2008 mortgage crisis and made millions from its collapse, has once again sounded the alarm. This time, his forecast isn’t about credit markets but the rapid spending by major tech companies channeling funds into artificial intelligence. Through his YouTube channel, Eisman draws a historical parallel that worries Wall Street: the excessive investments that characterized the dot-com bubble could repeat now in the race to dominate AI.

The recurring story of overinvestment: from 1999 to 2026

Eisman’s warning is based on an uncomfortable precedent. In the late 1990s, analysts worldwide claimed that the internet would revolutionize the economy. They were right at a fundamental level, but the timing was disastrous. The industry was flooded with speculative capital fueling an unprecedented investment frenzy. The result was predictable: too much money, too quickly, in still immature technologies.

When the dot-com bubble burst in 2001, it not only triggered an economic recession but also caused tech stocks to take years to recover. Overinvestment was largely the root cause of the crisis. Today, Eisman sees similar movements around AI, though he acknowledges that the context is different and his analysis warrants caution.

Over $300 billion in CapEx: Is AI spending sustainable?

Eisman’s argument rests on a concrete figure: Meta, Google, Amazon, and other tech giants are collectively spending over $300 billion on capital expenditures (CapEx) related to AI projects. All are competing in the same direction, pursuing the promise of increasingly advanced AI systems. But here’s the uncomfortable question: is this massive level of investment justified by the actual results it’s producing?

Signs of fatigue: the limits of ChatGPT as an indicator

One of the first cracks in the narrative of unstoppable progress appears when examining recent developments in language models. According to critics cited by Eisman, the current approach of scaling models—increasing computational resources to train larger systems—may be approaching its limits. The newly released ChatGPT 5.0 has not shown revolutionary improvements over its predecessor, ChatGPT 4.0, suggesting a slowdown in the innovation curve.

This deceleration, though still early, points to a fundamental problem: no one knows for sure what the actual return on investment (ROI) of this colossal AI infrastructure spending will be.

The scenario Eisman fears: a “painful digestion” period

If the returns on this monumental investment turn out to be disappointing in the short term—which could happen soon—the AI spending race will experience a sharp slowdown. Companies will cut budgets, projects will decelerate, and the industry will enter what Eisman describes as a “painful digestion” period. A scenario similar to what the tech sector experienced after 2001, when years of investment euphoria had to be slowly processed by the market.

Eisman warns that history isn’t always the master of the future. But its parallels are clear enough to keep the radar on.