Between 2025 and 2026, the global stock market experienced an unprecedented concentration of demand. According to calculations by several asset management firms, Micron Technology and SK Hynix, two memory chip companies, contributed approximately 17% of the returns to the MSCI All Country World Index in a single month. These two companies, with a combined weighting of only about 1% of the index, drove a significant portion of the combined returns across thousands of stocks and dozens of national markets worldwide. This means that for every $1 increase in the global stock market, approximately 17 cents came from these two stocks highly correlated with AI high-bandwidth memory (HBM). This data is not an isolated event. In the first half of 2026, the South Korean stock market significantly outperformed major global markets, driven by HBM demand. SK Hynix's year-to-date gains exceeded 250%, and Micron Technology also performed strongly during the same period. The weighting of technology and semiconductors in the MSCI ACWI index continues to climb, with the "Big Seven" and related AI supply chains in the US market accounting for nearly or more than 40%. Market concentration has expanded from the early "Big Seven" to the entire AI infrastructure chain—chips, power, real estate, construction, and related equipment. Superficial diversification strategies are failing. The Russell 2000 Small Cap Index recorded one of its best half-year performances since 1991 in the first half of 2026, with a gain of approximately 22%. However, of the 50 best-performing stocks in this index, about 16 belong to semiconductor or chip equipment companies, with some individual stocks seeing year-to-date gains of 250% to 380%. Investors attempted to avoid the AI ​​bubble by investing in small-cap stocks, but actually ended up buying into companies supplying cables, testing equipment, and related hardware to AI data centers. Value investing was not spared either. The Russell 1000 Value Index outperformed the growth index for a period in 2026, seemingly indicating a return to value investing. However, deeper analysis reveals that before the mid-year rebalancing, the value index had already heavily allocated to high-flying semiconductor stocks (Micron, AMD, Western Digital, etc.). After the rebalancing removed chip stocks from the value index and added Amazon, Apple, and Microsoft, the "defensive" nature of the value portfolio effectively transformed into exposure to large technology and AI-related companies. The mechanical rebalancing accidentally captured a selling opportunity at the momentum peak, but instead caused the so-called value portfolio to fall back into the AI ​​chain. The impact of AI has far exceeded the technology sector. The enormous electricity demand of data centers has directly linked the valuations of some US utility companies to their expected AI capital expenditures; the boom in warehousing and server real estate and data center construction has driven demand for building, electrical installation, and related equipment. The wealth effect is further spreading: liquidity obtained by employees of AI-related companies through IPOs or share buybacks has flowed into the high-end housing, private jet, and luxury goods markets, indirectly impacting a wider range of services and consumption. Ordinary investors' portfolios, whether actively or passively managed, are unlikely to completely escape this chain. II. Potential Wealth Evaporation: A Consensus of Trillions of Dollars If valuations revert to their long-term averages or experience a correction similar to that of 2000, the scale of wealth loss will far exceed historical precedents. Economist Dean Baker's AI bubble monitoring estimates that the total market capitalization of the US stock market is approximately $80 trillion. If the price-to-earnings ratio only falls back to its long-term average (without a severe crash), it could wipe out about $40 trillion in market value, equivalent to about $300,000 per American household. It's important to note that this figure is an arithmetic average: the wealthiest 10% of American households hold about 90% of stocks, so the actual loss for a typical household is far lower. However, the impact is significantly greater on the middle class, as stocks have surpassed real estate as the primary component of their wealth (for the first time since World War II). Former IMF chief economist Gita Gopinath estimates that a correction similar to the dot-com bubble could destroy about $20 trillion in US wealth, plus about $15 trillion in related assets held by foreign investors. The US portion accounts for about 70% of GDP. A scenario analysis by consulting firm Oliver Wyman points to a market capitalization loss of about $33 trillion, exceeding a year's US economic output. In contrast, the bursting of the dot-com bubble in 2000 wiped out only about $6 trillion in equity value. Current mainstream estimates are five to six times that amount. These figures are not predictions of an inevitable collapse, but rather scenario calculations based on valuation deviations and historical corrections. The key point is that the wealth effect of stocks has amplified its impact on consumption. Research shows that every $100 of paper stock wealth corresponds to approximately $3 of actual consumer spending. The evaporation of trillions of dollars in wealth will directly impact the real economy through canceled renovations, postponed car purchases, and reduced vacations, generating chain reactions of unemployment and demand contraction without a systemic banking crisis. III. AI Capital Expenditure: Growth Engine and Hidden Promises An analysis of the latest financial statements of major tech companies by the Wall Street Journal reveals that nine major tech companies, including Alphabet, Amazon, Meta, and Microsoft, reported approximately $600 billion in capital expenditures over the past 12 months. However, their off-balance-sheet AI-related commitments (long-term data center leases, chip and computing power procurement commitments, energy lock-in agreements, etc.) total approximately $3 trillion, about five times the reported capital expenditures. Alphabet alone has over $800 billion in procurement and contract commitments. These commitments are legally disclosed in the footnotes but are not all included in the balance sheet, meaning that the actual scale of future spending locked in by these companies far exceeds the figures the market typically focuses on. Each commitment is based on the assumption that AI will ultimately generate sufficient revenue to cover costs. Meanwhile, the private lending market is showing signs of stress. The US private lending default rate is expected to reach record levels in 2025, with some listed private lending funds experiencing rising problem loan ratios, and default rates recorded by institutions such as Fitch remaining high in mid-2026. Software and AI-related borrowers are a significant exposure. Private lending, as a lightly regulated financing channel outside the banking system, has reached trillions of dollars in scale, and its lack of transparency increases the possibility of simultaneous risk exposure. Unlike 2008, large banks' capital and liquidity buffers have significantly improved, making a direct recurrence of the systemic banking crisis less likely, but some risk has shifted to off-balance-sheet and private markets. IV. Historical Lessons: Real Technology Does Not Protect Overpaying Investors The Bank for International Settlements (BIS) compares the current AI development boom to the British railway boom of the 1840s, electrification in the 1920s, and the dot-com bubble of 2000. Railways transformed transportation and housing, but over-construction on lines with low passenger traffic caused stock prices to fall by about two-thirds around 1850. Electrification and the internet are both real and enduring technological revolutions, yet early investors suffered significant losses due to overvaluation and overcapacity. BIS research shows that, relative to their starting points, AI development has already surpassed these historical examples, teetering on the steepest trajectory. Historical frenzy typically reverses around the fifth year, and we are currently in a relatively early stage. The dot-com bubble provides a more recent reference. Telecom companies borrowed heavily to lay fiber optic cables in the late 1990s, creating a vast amount of "dark fiber," with most companies going bankrupt before traffic arrived. The Nasdaq fell by nearly 80%, and the S&P 500 by about half. Even Amazon, the eventual winner, saw its stock price fall by about 90% when the bubble burst, and investors who bought at the 1999 peak had to wait until around 2009 to break even. Early leaders in search (Infoseek, Lycos, AltaVista, Excite, etc.) were ultimately marginalized, while Google only truly rose to prominence after the bubble burst. Technological success and stock investment success are two entirely different things: the former concerns long-term productivity, while the latter depends on the purchase price and holding path. Repeated prediction failures are also cause for concern. Between 2011 and 2021, many well-known investors and observers repeatedly warned of a tech bubble, only to be proven premature or wrong over a long period. When a New York Times article summarizing "The Tech Bubble That Never Burst" was published in April 2022, it coincided with a significant correction in the Nasdaq. Bear market predictions are often premature and frequently wrong, but when they are occasionally correct, market participants have become accustomed to ignoring them. This does not constitute evidence that "all is well," but only illustrates how extremely difficult precise market timing is. Long-term holding and systematic rebalancing, rather than emotionally driven trading, can still yield positive returns after multiple historical tops. V. True Diversification: Reducing Exposure to a Single Narrative In highly concentrated environments, nominal global or style diversification may still be highly correlated. Truly effective diversification requires holding assets that perform significantly differently under the same shock. The European market is often described as slow-growing and aging, but its technology weighting is significantly lower than the US (approximately 10% compared to nearly half), valuations are lower, and it has a higher proportion of cash-generating sectors such as banking, industrials, and healthcare, with a dividend yield of around 3%. Some fund managers view European exposure as a "ballast" in case the AI ​​narrative fails to materialize: lower expectations mean greater upside potential and less speculative premium. This is not a recommendation to completely withdraw from US or AI-related assets, but rather to marginally increase exposure to less correlated assets. The core objective of long-term investing is to accumulate wealth that can support retirement, not to chase the hottest narratives of each era. The Nifty Fifty of the early 1970s, the Japanese stock market of the late 1980s, and the Nasdaq of 2000 were all seemingly obvious choices at the time, but subsequently experienced deep drawdowns or prolonged stagnation. Investors don't need to attend every party. AI technology itself has the potential to genuinely change productivity, and most market participants invest capital and resources based on this belief. However, the reality of the technology and the rationality of current valuations are two separate issues. Confusing the two is a costly mistake that has repeatedly occurred throughout financial history. The current market structure shows that the AI ​​chain is deeply embedded in global equity returns, economic growth accounting, and household wealth composition. Assessing downside risks is not about denying the technological outlook, but rather about acknowledging the vulnerabilities revealed by concentration, off-balance-sheet commitments, and historical precedent. For institutional and individual investors, the key is to quantify their portfolios' actual exposure to any single narrative and consciously allocate assets that can function independently under different macroeconomic scenarios. True protection comes from structural decoupling, not from faith in any single story.