Summary
1. The AI Infrastructure Investment Boom
At the beginning of the article, Wu describes how AI has reached a new inflection point. The largest US technology companies – the so-called Magnificent 7 (Apple, Microsoft, Amazon, Meta, Google, Nvidia, and Tesla) – have initiated a massive wave of investment in AI infrastructure. By 2025, their combined capital expenditure (capex) is projected to reach nearly $400 billion, and according to McKinsey, total AI-related investments could exceed $5.2 trillion over the next five years.
Exhibit 1 (“AI Investment Boom”) illustrates this development: AI investments have grown sharply since 2020, with growth accelerating particularly after the launch of ChatGPT.
The markets have so far reacted positively to this investment wave. For example, Oracle’s stock rose 36% after the company announced it would build data centers for OpenAI, and CoreWeave’s stock tripled after its IPO. Valuations of AI-related companies reflect great optimism for future growth.
However, Wu points out that return expectations may be unrealistic. According to Bain’s calculations, AI data centers would need to generate $2 trillion in annual revenue by 2030 for the investments to be economically justified. Currently, however, the total revenue from AI is only about $20 billion – a hundredfold increase would therefore be necessary. Companies are struggling with the practical utilization of AI, and even ChatGPT has not yet managed to significantly monetize its user base.
Wu draws a direct parallel to the 1990s telecommunications boom, when companies like Global Crossing and AT&T invested over $500 billion in fiber optic networks. As was the case then, there is now also a risk of overbuilding and oversupply of capacity, which could lead to price collapses and weak returns for years to come.
2. Market Concentration and Vulnerability
In the next section (“AI-Driven Stock Market Fragility”), Wu examines how the AI theme has begun to dominate the entire stock market. According to JPMorgan, AI stocks have accounted for 75% of S&P 500 returns, 80% of earnings growth, and 90% of capital expenditure growth since the launch of ChatGPT.
Exhibit 2 (“Top-Heavy Stock Market”) shows that the Magnificent 7 now accounts for over 30% of the S&P 500’s total weighting, a higher concentration level than at the peak of the dot-com bubble in 2000. This makes the markets vulnerable: if AI investment return expectations disappoint, the impact on the entire index and the economy could be significant.
Wu notes that AI investments are currently so large that they practically support US economic growth. Estimates suggest that AI-related capital expenditure has accounted for up to half of the country’s GDP growth over the past year. This makes the economy and investors increasingly dependent on the success of AI projects.
3. Historical Comparisons: Capital Cycles
Wu expands his analysis by comparing the current AI boom to previous technological investment cycles, such as the railroad boom in the 1800s and the internet infrastructure boom in the 1990s.
Exhibit 3 (“Tech-Led Investment Booms”) shows that, relative to GDP, AI investments have already surpassed the internet’s peak and are approaching the level of the railroad era. When considering the faster obsolescence of AI hardware (e.g., GPU upgrade cycles), the current boom is even more intense than historical benchmarks.
Exhibit 4 (“Railroad and Internet Bubbles”) illustrates how stock prices during these eras first rose sharply but then collapsed when oversupply and poor profitability became apparent. Wu refers to the so-called “capital cycle” theory, according to which investment booms often lead to oversupply and weak returns when demand fails to keep pace with supply growth.
4. Empirical Evidence: Capital-Intensive Companies Underperform
Wu supports his claims with extensive empirical data. He demonstrates that companies that grow their balance sheets or invest heavily in physical infrastructure yield weaker stock returns on average.
Exhibits 6–10 show that companies with high investment growth have historically underperformed across all sectors and geographical regions. This holds true in the United States, Europe, and Asia.
5. Risks and Transformation of the Magnificent 7
Wu proceeds to examine the Magnificent 7 companies in detail. Over the past ten years, these companies have generated an average of 27.5% annual returns and created over $23 trillion in shareholder value.
Exhibit 13 (“Magnificent Dominance”) shows how they have far outpaced the rest of the S&P 500. Their success has been underpinned by an “asset-light” model – a business based on intangible assets such as software, brands, and network effects.
Now, however, the situation is changing. Exhibit 15 (“Magnificent 7’s Asset-Heavy Transition”) shows that their capital intensity has risen from 4% to 15% of revenue since 2012. Meta, Microsoft, and Alphabet are already spending 21–35% of their revenue on investments – more than the average global energy company or even AT&T at the peak of the telecom bubble.
Wu warns that this transformation is turning them into “new-era utilities” – companies that tie up vast amounts of capital and whose rates of return decline over time.
6. Weakening Fundamentals and Financial Risks
Exhibit 19 (“Magnificent 7 Free Cash Flow”) shows that free cash flow has already begun to decline due to AI investments. Furthermore, “circular financing” phenomena are emerging, where companies artificially fund each other: for example, Nvidia invests in OpenAI, which in turn buys Nvidia’s chips.
Wu notes that while the Magnificent 7 remain highly profitable, their increasing depreciation and potentially overly optimistic assumptions about the lifespan of data centers could weigh on earnings for several years.
7. “AI Prisoner’s Dilemma”
At the end of the article, Wu describes the AI competition from the perspective of classical game theory. While it would be rational for companies to curb their investments and maintain their current oligopoly, no one dares to fall behind. Everyone fears that a competitor will win “all the markets” through AI. This leads to a “prisoner’s dilemma”, where everyone invests too much – even if it destroys collective profitability.
8. Conclusion
Wu concludes the article by stating that while AI is technologically transformative and productive in the long term, investors should prepare for the risks of overheating and overbuilding. Historically, similar capital cycles have led to weak returns and market corrections.
His recommendation is to diversify investments and seek beneficiaries of the AI ecosystem that do not require massive capital investments – for example, software, service, and data companies that can leverage AI without heavy infrastructure costs.
[/details]In summary: the AI boom in many ways resembles previous technological bubbles. While AI will revolutionize the economy in the long term, in the short term, investors may face overbuilding, weakening returns, and market instability – especially if the investments of the Magnificent 7 do not yield the expected results.
