How Much Would an AI Crash Destroy?

Episode Summary

Executive Summary: The episode argues that the AI boom has become so concentrated that it now drives markets, household wealth, and parts of the real economy, creating meaningful downside risk if expectations cool. It stresses that AI may still be transformative, but real technology does not guarantee good investments, and diversification is increasingly harder as AI exposure spreads into many asset classes and sectors.

Main Topics: AI concentration in markets and portfolios: The transcript opens with the striking fact that two memory-chip companies accounted for 17% of global equity market returns in May, illustrating how concentrated the AI trade has become and how even ordinary portfolios may be more exposed than they appear. Why diversification is harder than it looks: The speaker explains that AI exposure is not limited to obvious tech stocks; it has spread into small caps, utilities, real estate, construction, and even value indexes through rebalancing, meaning many investors cannot easily escape the trade. Potential scale of an AI bubble unwind: Multiple estimates are cited for how much wealth could be erased if AI valuations normalize or crash, ranging from tens of trillions of dollars in U.S. wealth losses to larger losses when foreign holdings are included. Spillovers into the real economy: The episode emphasizes that rising AI-related capital spending boosts construction, energy, labor, and consumer spending, while a slowdown could reduce incomes and spending even for people who never bought AI stocks. Hidden leverage and private credit risk: Beyond public markets, the transcript warns that AI commitments extend into off-balance-sheet obligations and private credit, where opaque lending and rising defaults could amplify stress if the boom weakens. History of technology bubbles and long recovery times: The transcript uses railways, the dot-com bubble, and Japan’s lost decades to show that transformative technologies can coexist with disastrous investor outcomes when prices become disconnected from fundamentals. Practical takeaway: diversify into boring assets: Rather than predicting a crash, the speaker argues for modest global diversification and exposure to cheaper, less crowded markets such as Europe, along with boring defensive holdings that can cushion a reversal.

Key Arguments: AI is not just a tech-sector story; its gains and risks have spread into chips, utilities, construction, real estate, and consumer spending. Even investors trying to avoid AI often end up exposed through index composition, sector rebalancing, and adjacent suppliers. A correction in AI valuations could erase tens of trillions of dollars in stock-market wealth, with meaningful effects on household spending. Because stocks now represent a larger share of median household wealth than in the past, a market drawdown would hit more than just the rich. AI infrastructure spending has become a major driver of U.S. growth, so a slowdown could weaken the broader economy even without a banking crisis. The true risk is not that AI fails as a technology, but that investors overpay for its winners and get hurt even if the technology succeeds. Hidden commitments in footnotes and private credit exposures make the financial system more fragile than headline capex numbers suggest. History shows that being right about a transformative technology does not protect investors from buying too expensively at the peak. Diversification should mean owning assets that do not all fall together, not just owning many correlated growth stocks. Europe is presented as a lower-expectations, lower-tech-weight alternative that may provide ballast if the AI trade disappoints.

Data Points: Micron and SKHynex contribution to global market returns: 17% of the global stock market's return in May - According to an Acadian Asset Management report on the MSCI All-Country World Index Weight of Micron and SKHynex in global market: about 1% - They are tiny constituents of the global index yet drove a large share of returns U.S. stock market value: around $80 trillion - Used in Dean Baker's estimate of potential wealth loss if valuations normalize Potential wealth loss from mean reversion: about $40 trillion - Dean Baker estimate if U.S. P/E ratios returned to long-run averages Average loss per U.S. household: nearly $300,000 - Derived from the $40 trillion estimate, though highly unequal in practice Household stock ownership concentration: richest 10% own about 90% of shares - Explains why average losses overstate the impact on typical households IMF-style correction estimate: $20 trillion U.S. wealth + $15 trillion foreign wealth - Gita Gopinath estimate of a dot-com-style correction today Oliver Wyman estimate: about $33 trillion wiped out - Alternative estimate of value destruction from an AI bubble unwind Dot-com bubble destruction: about $6 trillion - Historical benchmark for comparison Stocks vs. real estate in household wealth: stocks became the largest component of American household wealth for the first time since WWII - Cited from Goldman Sachs and the Federal Reserve Household spending sensitivity to wealth: $3 spent for every $100 of paper stock wealth - Rule of thumb for the wealth effect AI infrastructure share of U.S. growth: about 90% in the first half of last year, or more conservatively about a quarter - Harvard economist Jason Furman and alternative estimates Big Tech reported AI capex: about $600 billion over the last year - Headline capital spending disclosed by major tech firms Off-balance-sheet AI commitments: roughly $3 trillion - Wall Street Journal analysis of footnotes and purchase commitments Alphabet purchase commitments: over $800 billion - Example of a single company’s long-term AI obligations Private credit market size: $2 to $3 trillion - Described as a lightly regulated market funding parts of the AI/software ecosystem Troubled loans at top private credit funds: highest level since 2017 - Financial Times reporting on deteriorating loan quality Private credit defaults: record high in July - Fitch data cited in the transcript Single large fund's troubled loans: 7% of its loan book - Example of stress in private credit portfolios Railway mania investment: close to half of Britain’s GDP by 1850 - Historical analogy for overinvestment in transformative infrastructure Railway share collapse: about two-thirds lost by 1850 - Shows how useful technology can still produce investor losses Dot-com market declines: S&P 500 fell by about half; Nasdaq lost nearly 80% - Historical bubble comparison Amazon peak-to-trough decline: about 90% - Example of a future winner that still devastated peak buyers Recovery time for Amazon investors: about 10 years to break even - If bought at the 1999 peak

Pivotal Quotes: "This technology is real, and these stocks are a good investment at these prices, are two completely different statements." — Speaker: Core warning that usefulness of AI does not imply current valuations are justified "Being right cost you a decade." — Speaker: On buying Amazon at the dot-com peak and waiting years to recover "You don't have to attend every party." — Speaker: Final takeaway urging investors not to chase the hottest trade with all their wealth

Implications: Listeners should treat AI as both a real technology and a crowded trade. The likely best defense is broad diversification, lower expectations, and some exposure to less fashionable assets that can hold up if the AI boom cools.