What if the smartest trade in a mania is abstention? The message is unglamorous, but it travels well across centuries. Whether it was 1720’s South Sea investors or the dot-com faithful, the outcome was the same when cash flows failed to catch up to promises. The latest warning about exuberance in AI-linked equities is not novel. What is new is the engine propelling it: debt. When hype meets leverage, the system’s tolerance for error shrinks. This is not a call to fear. It is a call to measure load, stress, and failure modes before the bridge looks crowded.
Burry’s caution lands because it taps a real anxiety. But the problem with bubbles is not prediction; it is coordination. In game theory terms, this is a timing coordination game with career risk layered on top. Many know the music will stop. Few want to be the first to leave the dance floor. The Keynesian beauty contest—guess what others will think others will think—magnifies the delay. Scott Galloway is right on one point: timing is brutal. The goal is not calling the peak, but shrinking exposure to fragility. In markets, what breaks you is less often valuation and more often the path: liquidity vanishes, margins compress, and funding channels seize. Bubbles are processes, not events. They fail at the weakest seams.
The pivot from equity and free cash flow to debt to fund AI infrastructure is a tell. Equity is a shock absorber; debt is a constraint. Too much of the wrong capital at the wrong time converts a growth story into a solvency story. The telecom buildout in 1999 offers a warning: massive capex paid with borrowed money created overcapacity and a wave of defaults. Today’s AI buildout demands expensive chips, power, and real estate. If that is increasingly financed in bond and private credit markets—often with covenant-lite terms—it imports rollover risk into core technology narratives. Rates are not near zero anymore. Interest coverage matters. When the level of rates does the underwriting rather than the project’s returns, fragility builds. Changing one variable—say, slower AI monetization—can force costly refinancings and diluted equity just as sentiment cools.
Layer onto this a more modern accelerant. The IMF has warned that AI and machine learning can automate procyclicality. If models trained on recent data are embedded in credit underwriting, risk targeting, and market making, they can push everyone into the same trades at once. Correlations jump when it matters. Order books thin. Feedback loops form. George Soros called it reflexivity: expectations shape actions which shape outcomes which validate expectations—until the loop flips. The adoption of similarly trained models across funds, banks, and corporate treasuries is not diversification; it is an unseen concentration. When the same pattern detectors chase the same signals, volatility does not get shared, it gets synchronized. That is how small errors become system events.
The AI stack looks diversified from the surface. Underneath, it is narrow. One primary chip vendor. A handful of hyperscalers. Grid constraints in a few regions. Limited high voltage interconnect capacity. Long lead times on substations and cooling. Even export rules are a variable. This is not a judgement on any one firm. It is a simple engineering observation: a system with key single points of failure is brittle. If a chip shipment is delayed, if a data center’s power upgrade slides, if a regulation curtails a key component, what happens to revenue projections levered into debt-financed buildouts? In probabilistic terms, the tails are fatter than models assume. Investors should care less about next quarter’s beat and more about the resilience of the supply chain that produces those beats.
The story says scale wins. Sometimes it does. But scale comes with physics. Compute costs are nonlinear. Power and cooling do not obey spreadsheet growth rates. The assumption that unit economics will glide down a cost curve fast enough to justify today’s multiples may be correct for a few, not for many. Historically, great technologies often birthed poor investments when capital crowded in too fast—railroads in the 1840s, electrification in the 1910s, fiber optic networks in the late 1990s. Yale researchers studying real time bubble dynamics point to explosive price behavior that decouples from fundamentals as a recurring pattern. Analysts calling this cycle an everything bubble see the same signals: stretched valuations, hype-led funding, leverage creeping into the cap table. You do not need to know the minute the music stops to recognize the room is over capacity.
There is another quiet loop at work. Companies issue debt to buy back stock, tightening float and juicing per-share metrics. Passive vehicles then buy more of what has gone up, regardless of price. Private credit fills gaps that banks step back from, often without the same transparency. All three are rational choices on their own. Together, they are an accelerant. Rising prices invite more mechanical buying that invites more issuance that invites more price insensitive demand. In a falling tape, the flow reverses. The crowding that helped on the way up hurts on the way down. The fragility is not only in company balance sheets; it is in the market’s plumbing. Liquidity is a fair weather friend. When spreads gap and dealers step back, good names get sold with the bad because the seller needs dollars, not nuance.
Antifragility is not a buzzword. It is a design choice. In this cycle it would look like net cash balance sheets, flexible capex plans, and revenue tied to real demand rather than promotional sizzle. It would look like vendors with diversified suppliers, data centers with secured long term power, and customers who pay in cash rather than options on future profitability. For investors, it is small bet sizes when the edge is uncertain, dry powder in reserve, and strategies that do not depend on continuous liquidity. The Kelly criterion teaches that overbetting even a good edge courts ruin. Here, the edge is debated, the distributions are wide, and the correlations are unstable. A system built to gain from volatility is one that can survive a few bad coin flips without begging the pit boss for a loan.
Markets love speed. Balance sheets love durability. The AI story may be transformational over a decade. That does not make every balance sheet financing that story sound over the next two years. The paradox of this moment is simple: the faster capital chases the future, the more pressure it puts on the present’s plumbing. If you must be involved, underwrite the funding path as hard as you underwrite the technology. Look through the beauty contest to the cash conversion cycle. Momentum can reward patience-free strategies for a while, until it punishes them all at once. Avoiding crowded, levered trades is not fear. It is discipline. Sometimes the strongest position in a coordination game is to set your own tempo and refuse plays where the downside is irreversible and the upside is already priced as certain.
