Learning from history, when will the capital expenditure boom turn into a bubble burst?

From the 19th-century railroads to 21st-century artificial intelligence, every major technological innovation in history has sparked a capital expenditure boom, but the frenzy often ends in a bubble burst.

BCA Research's special report released in November this year, "When Capex Booms Turn Into Busts: Lessons From History," reviews four typical capital expenditure booms, revealing the core logic of the transition from prosperity to collapse, and issues a warning regarding the current AI boom.

The report summarizes five common patterns: investors overlook the S-curve of technology adoption, revenue forecasts underestimate the extent of price declines, debt becomes the core reliance for financing, asset price peaks occur before investment declines, and capital expenditure collapses exacerbate economic recessions. These patterns are already emerging in the current AI field—technology adoption rates stagnate, token prices have plummeted over 99%, corporate debt has surged, and GPU leasing costs have decreased.

Based on historical comparative analysis, BCA Research concludes: the AI boom is following the historical bubble path and is expected to end within the next 6 to 12 months. The report advises investors to maintain a neutral allocation to stocks in the short term, moderately underweight stocks in the medium term, and closely monitor forward-looking indicators such as analyst expectation revisions, GPU leasing costs, and corporate free cash flow.

The report particularly points out that the current economic environment adds to concerns, as U.S. job vacancies have fallen to a five-year low. If the AI boom fades without a new bubble to offset the impact, the upcoming economic recession could be more severe than during the 2001 internet bubble burst.

Historical Reflection: The Collapse Trajectories of Four Capital Frenzies

BCA states that the essence of capital expenditure booms is the collective optimistic expectation of capital regarding the commercialization prospects of new technologies, but history has repeatedly proven that such optimism often deviates from the objective laws of technology implementation, ultimately leading to collapse due to supply-demand imbalances, debt accumulation, and inflated valuations.

The 19th-century British and American railroad boom demonstrated the destructive power of overcapacity.

The report notes that the success of the Liverpool-Manchester railway in 1830 ignited an investment frenzy in Britain, with railroad stock prices nearly doubling between 1843 and 1845.

By 1847, the proportion of railway construction expenditure to the UK GDP soared to a record 7%. A tightening of liquidity ultimately triggered the financial crisis in October 1847, with the railroad index plummeting 65% from its peak.

The report states that the U.S. railroad boom peaked during the panic of 1873, forcing the New York Stock Exchange to close for ten days, with corporate bond default losses reaching 36% of face value between 1873 and 1875.

After the U.S. railroad mileage reached a peak of over 13,000 miles in 1887, overcapacity led to a collapse in transportation prices, and by 1894, about 20% of U.S. railroad mileage was in bankruptcy management.

The electrification boom of the 1920s exposed the fragility of pyramid-shaped capital structures.

The report points out that the proportion of households with electricity rose from 8% in 1907 to 68% in 1930, but this process was mainly concentrated in urban areas Wall Street is deeply involved in this craze, with utility company stocks and bonds being marketed as "safe assets that even widows and orphans can invest in." By 1929, holding companies controlled over 80% of the electricity generation in the United States.

The report states that after the stock market crash in 1929, the largest utility group, Insull, went bankrupt in 1932, reportedly causing the life savings of 600,000 small investors to vanish. U.S. electric utility construction spending peaked at approximately $919 million in 1930, then plummeted to $129 million in 1933.

The internet boom of the late 1990s confirmed that innovation does not equate to profitability.

BCA noted that from 1995 to 2004, the annualized growth rate of non-farm productivity in the U.S. reached 3.1%, far exceeding subsequent periods.

However, the proportion of technology-related capital expenditures to GDP soared from 2.9% in 1992 to 4.5% in 2000, with over-investment putting immense pressure on corporate balance sheets.

The report pointed out that free cash flow in the telecommunications industry peaked at the end of 1997 and then declined continuously, crashing significantly in 2000. The NASDAQ Composite Index rose sixfold between 1995 and 2000, then plummeted 78% over the next two and a half years.

Multiple oil booms perfectly illustrate the cyclical nature of supply and demand imbalances.

BCA stated that after the discovery of massive oil reserves in East Texas in 1930, daily production exceeded 300,000 barrels within 12 months, but the worsening Great Depression caused oil prices to plummet to 10 cents per barrel.

In 1985, Saudi Arabia abandoned production limits, leading to oil prices dropping to $10 per barrel at one point.

Between 2008 and 2015, the U.S. shale oil boom pushed crude oil production from 5 million barrels per day to 9.4 million barrels, while OPEC's refusal to cut production in 2014 caused oil prices to fall from $115 per barrel in mid-year to $57 by year-end.

Five Common Patterns: The Inevitable Path from Prosperity to Collapse

Reviewing the rise and fall of four typical booms, BCA Research summarized five common patterns that provide key metrics for judging the current trajectory of the AI boom. Specifically:

The first pattern is that investors overlook the S-curve of technology adoption.

Technology adoption is never linear; it follows an S-curve of "early adopters accepting—mass adoption—laggards following." Stock prices typically rise in the first phase, peaking in the mid-second phase when the growth rate of adoption turns negative.

The current AI field is exhibiting this characteristic: most companies express intentions to increase AI usage, but actual adoption rates have shown signs of stagnation, with some indicators even declining in recent months. This divergence between "willingness and action" is a typical signal of technology adoption entering the late second phase.

The second pattern is that revenue forecasts underestimate the magnitude of price declines.

In the early stages of new technology, scarcity grants pricing power, but as technology becomes widespread and competition intensifies, prices inevitably drop significantly. From 1998 to 2015, the annualized growth rate of internet traffic reached 67%, but the unit price of information transmission simultaneously fell sharply. The price of solar panels has continuously declined since their inception, dropping 95% from 2007 to the present The AI industry is repeating past mistakes: Since 2023, the launch of faster chips and better algorithms has led to a decline in Token prices of over 99%. Although new applications such as video generation have emerged, users' willingness to pay for these applications remains unclear.

Rule three is that debt has become the core reliance for financing.

In the early stages of a boom, companies can usually meet capital expenditure needs through retained earnings, but as investment scales expand, debt gradually becomes the main source of financing.

In October 2025, Meta announced a $27 billion data center financing agreement through an off-balance-sheet special purpose entity; Oracle, after obtaining a $38 billion loan, also financed $18 billion in the bond market, with total debt nearing $96 billion.

More concerning are "new cloud vendors" like CoreWeave, which, as of October 2025, saw its credit default swap rates rise from 359 basis points at the beginning of the month to 532 basis points.

Rule four is that asset price peaks occur before investment declines.

Historically, during capital expenditure booms, asset prices such as stocks often peak before actual investment expenditures begin to decline. Even if investment expenditures fall from their highs, their absolute values may remain elevated, continuing to exacerbate overcapacity. This means that if investors wait for clear signals of "investment decline" before taking action, they often miss the best opportunity.

Rule five is that the collapse of capital expenditure and economic recession mutually exacerbate each other.

The bursting of a technology bubble typically occurs in two stages:

The first stage is the retreat of technology speculation, revealing overcapacity; the second stage is the collapse of capital expenditure dragging down the overall economy, leading to deteriorating corporate profits and creating a vicious cycle.

The report points out that the 2001 U.S. economic recession was not triggered by a deterioration in economic fundamentals but rather stemmed from the collapse of capital expenditure following the burst of the internet bubble. The rise of the real estate bubble in 2002 temporarily alleviated the impact of the internet bubble's collapse, but it remains uncertain whether a new bubble will emerge to offset the impact of the AI boom's collapse.

Risk Signals of the AI Boom: Turning Points in 6 to 12 Months

Based on a comparative analysis of historical patterns, BCA Research believes that the AI boom is following the path of historical bubbles and is expected to end within the next 6 to 12 months. This judgment is based on multiple risk signals that have already emerged in the AI field.

From the perspective of technology adoption, the actual implementation speed of AI has not kept pace with the fervent expectations of capital, with stagnation in enterprise adoption rates and consumers' willingness to pay for AI applications not yet fully validated.

From the price trend perspective, the significant decline in Token prices has shown deflationary pressure, while the commercial value of new applications such as video generation remains in doubt.

From the perspective of debt risk, the financing structure of AI-related companies is increasingly reliant on debt, with some companies' credit risks beginning to surface.

The report suggests focusing on four leading indicators:

First, the revision of analysts' expectations for future capital expenditures; if the continuously rising expectations begin to level off, it may be a dangerous signal;

Second, GPU leasing costs, which have begun to decline since May 2025; Third, the free cash flow situation of super-large enterprises, although still at an absolute high level recently, has shown a deteriorating trend;

Fourth, the emergence of the "metaverse moment," where the stock price of an AI company declines after announcing a major project, will be a clear sign of a shift in market sentiment.

For investors, BCA Research recommends adopting a "moderate defensive" strategy at present. In the short term, that is, over the next three months, maintain a neutral allocation to stocks, while in the medium term, that is, over the next twelve months, moderately underweight stocks, and further enhance defensiveness in the coming months.

Specifically, it is important to closely monitor the aforementioned four leading indicators and avoid passive adjustments only when investment expenditures clearly decline; at the same time, attention can be paid to defensive sectors and high-quality bonds to hedge against potential significant fluctuations in AI-related assets