Everyone says this time is different until the math says it is not. If it is truly a revolution, it should withstand a capital cycle, a recession scare, and a crowd that bought the same story at the same time. The paradox of the AI trade is simple: the stronger the narrative, the more fragile the system that grows around it.
The highest priests in tech and finance are starting to admit the obvious: valuations in Big Tech have run well ahead of tested cash flows. That is not the core risk. Valuation excess is a symptom, not the disease. The structural fragility is in the feedback loops we built around the theme—index concentration, capex reflexivity, and model-driven herding. Once growth expectations set the cadence of spending and index flows, small disappointments propagate like cracks in tempered glass. Federal Reserve Vice Chair Michael Barr flagged the core mechanism: speed and automaticity. When machines and mandates move together, errors scale. That is how clustered volatility becomes systemic even without a banking crisis.
Call it the tyranny of the path. Investors anchor on long-term total addressable market and dismiss near-term drawdowns as noise. That is linear thinking in a non-linear game. In game theory terms, we are in common knowledge territory: everyone knows everyone else believes AI is the only game in town. This boosts correlation and weakens the market’s shock absorbers. Passive flows chase winners, models train on similar signals, and volatility becomes endogenous. Jim Cramer says embrace the dips; the Kelly criterion says beware the sequence of returns. High volatility with occasional large drawdowns compounds poorly for levered players and for anyone with a fixed spending obligation. The geometric return is what pays pensions and endowments, and it is penalized by variance. That is fragility hiding in plain sight.
Goldman Sachs points to the next hinge: a cooling of capital spending after a year of breakneck data center and chip investment. The capital cycle is ruthless. It rewards scarcity and crushes abundance. Semiconductors know this dance: booms seed busts via overcapacity. The dot-com era did not end because the internet failed; it ended because orders slowed from unsustainable levels and inventories met reality. Today’s AI buildout is running into old-economy constraints—power, transformers, transmission lines, permitting. When lead times stretch, procurement over-orders. When constraints ease, the bulge reverses and suppliers face an air pocket. That is how narratives that are correct in the long run still produce brutal drawdowns in the short run. If capital spending is the heartbeat of the theme, even a normalization, not a collapse, can reset equity prices that capitalized peak momentum.
A durable theme does not immunize weak unit economics. Training and inference costs remain heavy, and pricing power is unproven across broad enterprise use. Many deployments are subsidies in search of scale. In software, gross margins face a hidden tax from compute and model calls that vendors cannot always pass through. The belief in winner-take-most is doing the heavy lifting in equity multiples; the base rate says diffusion is slower, margins compress under competition, and regulators tax the rents. Nobel laureate Paul Romer warns that the hype cycle around platform technologies rhymes with crypto: bold claims and underpriced friction. Economist David Rosenberg calls out the dot-com parallel: great technology, poor entry points, and a long valley between promise and profits. The history of general purpose tech is not a straight line up. It is an S-curve with a long flat section that punishes investors who bought the vertical.
The index now leans on a handful of mega-cap names with AI as the shared growth vector. That concentration amplifies macro shocks. If capex cools, earnings misses synchronize, and the correlation structure spikes. In 2008, the plumbing failed because leverage hid in structured products and off-balance-sheet vehicles. Today, the fragility is different but rhymes: vendor financing in the supply chain, multi-year cloud commitments, power purchase agreements, and capex partnerships that look like debt in a downturn. When cash flows wobble, obligations do not. Fragility is the mismatch between flexible revenue and fixed commitments. Financial history is clear on this point: it is not the average growth rate that breaks systems; it is the tails and the timing of cash needs.
The market confuses fast with safe. AI accelerates decision loops, but that speed cuts both ways. Fast diffusion compresses the period when incumbents earn excess returns from scarcity. Fast feedback turns crowd behavior into herding. The Fed’s Barr put it bluntly: speed plus automation leads to concentration of risk. Automated trading and model-based capital allocation magnify moves because signals are correlated. When everyone optimizes to the same objective function, the system loses diversity. In nature and engineering, systems that lack diversity fail catastrophically. The antifragile alternative is messy: redundancy, slack, and optionality. None of those screen well in an efficiency-obsessed bull market, which is why they pay when conditions turn.
Invert the problem. Suppose AI is as transformative as its boosters claim. What breaks first? Power economics will reset, squeezing margins across compute-heavy businesses. Pricing will converge toward marginal cost, especially as open models improve. The beneficiaries may be outside the indices everyone owns: grid infrastructure, boring materials, niche suppliers with bottleneck control. Meanwhile, the glamour cohort carries the cost of scaling and the burden of expectations. If the story disappoints, the unwind is more direct: capex cuts, earnings downgrades, and a correlation shock as everyone tries to narrow the same trade at once. Either tail is adverse for crowded longs. Investors dusting off the dot-com playbook—diversification, valuation discipline, real cash flow over adjusted metrics—have the right instinct, but the implementation matters. Avoiding concentration risk is not the same as owning equal slices of the same theme across ten tickers.
None of this argues for abandoning innovation or shorting progress. It argues for respecting probability and base rates. The market has priced the upside scenario with precision and the downside with hope. Hope is not a hedge. The discipline is straightforward: separate the technology’s destiny from the securities that represent it; treat capex booms as cyclical, not structural; measure unit economics before narratives; and remember that path matters more than destination. The market’s reflex is to buy the dip. Sometimes the correct trade is to step aside and let variance do its work. Embracing volatility sounds brave on television. In portfolios, it is often a wager against compounding.
The AI era may well justify the excitement. The question a stoic would ask is whether the system built around that excitement can withstand its own success and its inevitable stumbles. Markets break on what everyone knows for sure that just ain’t so. Today, everyone knows AI will save the index. That is precisely the kind of knowledge that deserves a discount.
