Leveraged exchange-traded funds tied to individual stocks have crossed a threshold that transforms them from a retail hazard into a potential systemic event. More than 450 leveraged and inverse single-security ETFs have launched since 2022, a category that now commands over $150 billion in assets and is stacked overwhelmingly on a tightly correlated cluster of AI and semiconductor names. Roughly a quarter of all new US ETF filings are now leveraged or inverse products. The concentration and scale mean these instruments no longer fail one at a time—they share a common fuse tied to the outlook for AI capital spending that lifted all of them simultaneously.

The mechanics of these products create a predictable cycle of destruction. A buyer of a 2x single-stock ETF is not a counterparty to the trade but raw material for a loan from the future that the product's mechanics are built to call. A stock levered 2x daily can only keep rising forever, which it will not; fall, in which case leverage turns a 30% correction into a catastrophe; or chop sideways, in which case volatility drag quietly evaporates the fund. Two roads lead to ruin, and the third does not exist. The question is no longer whether retail investors will be mauled—that outcome is close to consensus—but whether the damage will radiate beyond them.

The answer lies in following the money out of the investor's pocket. Some of it vanishes through volatility drag, an arithmetic cost of resetting leverage along a bumpy path that no one pockets. The rest flows to fees, financing spreads on the swaps that manufacture the leverage, and insurance. Banks issuing total-return swaps to ETF sponsors buy protection against the underlying stocks collapsing, largely through cliquets—chains of options that reset to the current price at intervals and pay out leg after leg as a stock falls. The cost of these instruments has more than tripled for the hottest names over three months. Opportunistic trading adds another layer: end-of-day ETF rebalancing is large and predictable, so market makers front-run it and skim the difference.

The banks are not holding the crash risk—they bought protection. The trouble, if it comes, will land on whoever sold it to them. The cliquets and swaps are absorbed by yield-reaching asset managers and hedge funds, who collect large premiums in exchange for wearing the tail risk. They are short a book of correlated AI gap risk, and the fatter the premium grows, the more of them are tempted in. This is the next group in line after retail—the carry-hungry institutions that wrote the crash insurance. It is the oldest structure in finance, the one that detonated in the 2018 volatility blowup and, in a grander key, in 2008, when monoline insurers who sold cheap insurance on mortgage-backed securities were vaporized when the tail risk arrived.

Could an AI crash reach the big banks? No single name is large enough, the derivatives are disclosed, and the resets are daily. But the products do not fail one at a time because they share a fuse—the same outlook for AI capital spending. When the trend breaks, they rebalance in the same direction, into the same closing auction, on the same afternoon. Nomura Holdings Inc. estimates leveraged ETFs already generate some $9 billion of rebalancing demand for every 1% market move. Barclays Plc. puts recent US rebalancing at several times its long-run average. That is not one exploding cigar—it is a box of them wired to a single fuse.

Regulators have begun to register concern. South Korea's top financial regulator has openly regretted approving 16 leveraged chipmaker ETFs in May, saying the negative side-effects of the products have grown significantly. Europe offered a cleaner warning: a 3x short AMD product there was wiped out in a single day's rally. The US does have a guardrail that has capped leverage in these products. Under Rule 18f-4, the Securities and Exchange Commission anchors a leveraged fund's risk test to the asset it tracks, which is why every 3x and 5x single-stock application has been rejected and the ceiling holds at 2x.

But that leaves open the danger of 450 individually compliant products, every one pointed at roughly the same trade, selling together. And the correlated pile is about to grow: SK Hynix lists on Nasdaq on July 10, and the moment a hot AI name trades in New York, the issuers race to lever it. The system is stable for now in the way a row of dominoes is stable—every piece individually hedged, the eventual loss already sold to the retail buyer who mistakes a paper gain for profit and to the yield-hungry fund that wrote the crash insurance. The only open variable is how tall the pile grows before the AI trend finally breaks.