Bank of America Warns AI Investment Cycle May Repeat Historical Disappointment
Bank of America strategists noted in a report that the current AI investment boom is following a classic capital cycle path, with intensified competition and declining prices potentially leading to a disappointing outcome similar to past cycles.
The 12-month forward consensus for capital expenditure among mega-cap firms has risen from $330 billion in early 2025 to $1 trillion, equivalent to 3% of U.S. GDP, surpassing the 2.2% from the 1870s railroad boom and 1.1% from the 1990s tech boom.
Strategist Sebastian Raedler pointed out that most major capital cycles in history ultimately end in disappointment, even though the underlying technologies often meet expectations; high early profit margins drive excessive investment and overcapacity, which erodes pricing power.
AI is facing intensified competition at the model layer (including Chinese open-source weight models), with average token prices dropping about 50% since summer; cloud businesses avoided this fate in the 2010s due to high entry barriers, but the current monetization disappointments reflect the failure patterns of past capital cycles.
AI-related assets have come under pressure, with the South Korean KOSPI down 20% since early October, and the Philadelphia Semiconductor Index down 13% from recent highs.
Bank of America previously noted that most clients believe the AI spending boom may end in a crash, but many still plan to maintain exposure while preparing to switch positions when signs of fuel exhaustion appear.
Source: Public Information
ABAB AI Insight
Bank of America’s strategy team continues to track capital expenditures among mega-cap firms, comparing the current scale of AI investments with historical capital cycles such as railroads, electrification, and the internet; recent notes clearly indicate that the expenditure expectation rising to $1 trillion (3% of GDP) has exceeded past peaks, and competition and price declines are eroding pricing power.
In the capital path, early high profit margins attract large-scale investments, leading to overcapacity; competition at the model layer (open-source and low-cost alternatives) has caused token prices to drop by about 50%, with monetization paths obstructed, resulting in corrections in related assets (semiconductors and the Korean index).
This is similar to the outcomes of the 1990s tech capital expenditure boom and the 1870s railroad construction: the technology itself may deliver, but excessive investment and competition lead to returns below expectations; currently, AI is transitioning from early high barriers to model layer price competition, with monetization pressure becoming a signal of cyclical turning points.
Essentially, this represents a cyclical adjustment following capital concentration: capital expenditure scales far exceed historical cycle peaks, and pricing power shifts from early monopolies to competitive erosion; when monetization fails to match investment, overcapacity and price declines reduce returns, triggering a shift from investment boom to disappointment or contraction, putting pressure on related assets and financing chains.
ABAB News · Law of Cognition
- After the peak of capital cycles, competition erodes pricing power
- When expenditure as a percentage of GDP exceeds historical levels, the probability of disappointing outcomes increases
- Technology delivering does not equate to investment returns delivering.