Take shelter! The wave of AI mania is about to break

The AI investment surge is reminiscent of the classic indicators of a late-stage mania: soaring borrowing, stagnant earnings, inflated valuations, and surging interest rates. Natale Labia, an economist and partner at a global investment firm, writes these observations in his personal capacity. He refers to Nikolai Kondratieff, a Soviet economist who, in 1925, analyzed economic cycles in Britain, France, and the United States.

Kondratieff identified long-term cycles of roughly 50 years, characterized by boom periods, followed by manias, crises, and depressions. Despite facing repercussions from Marxist critics and Stalin, Kondratieff's work has proven influential.

Joseph Schumpeter later incorporated Kondratieff's concept, referring to these cycles as "waves" and attributing the term "creative destruction" to them. Each wave, Schumpeter argued, was propelled by a cluster of innovations, such as steam, cotton mills in the 1780s, railways in the 1830s, oil, the motor car, and mass production after World War II, and more recently, the microprocessor and the internet in the 1970s. These technological advancements not only transformed society but also led to financial collapses.

Carlota Perez, building on Schumpeter's work, identified two critical phases of technological revolutions. The first phase, known as the "installation" phase, involves speculative finance pouring capital into new infrastructure faster than the real economy can absorb it. This phase culminates in a mania and a subsequent bust. Only after this does the "deployment" phase commence, during which the new technology is put to productive use.

The current AI mania is evident in the significant increase in capital expenditure by major AI hyperscalers. Combined AI-related capital expenditure has roughly doubled, from $450-$500 billion last year to nearly $1 trillion in 2026. However, this spending is not being funded by profits and cash flow. In fact, the combined free cash flow of the four biggest hyperscalers is expected to turn negative in 2027.

These companies have more than doubled their bond issuance in a single year, indicating record borrowing to build capacity for revenue and demand that does not yet exist.

Valuations for AI companies are at their highest since the dot-com bubble, but there is a peculiar circularity in their earnings growth. Much of this growth is simply another company's capital expenditure. Chipmakers' sales are actually the hyperscalers' capital expenditure. If the anticipated revenue does not materialize, the earnings bubble will inevitably burst in both sets of income statements.

The final signal of a late-stage mania is the increase in insider sales to outsiders. Currently, this is occurring in the context of the Federal Reserve's hawkish stance, with a 25 basis point rate hike on September 16, 2026. This decision was made unanimously, citing a strong labor market, persistent inflation, and the energy shock from the Iran War.

The market is now anticipating another rate hike this year, with the European Central Bank having already begun raising rates and the US 10-year bond yield surpassing 5%. These macroeconomic conditions align with textbook late cycle dynamics, suggesting that the AI mania may be nearing its end.