Banks Are All That's Left Standing: Hedge Funds Dumped the Rest After a 19-Month-Worst Day

On Nov 5 the S&P 500 fell 4.36% to 7,353.73, its worst day since April 2025, and the Nasdaq Composite dropped 5.52% to 24,315.00. Every one of the 11 S&P sectors closed lower, information technology worst at 6.6%, and the semiconductor ETF lost roughly 12% in a day that matched the COVID-19 crash. On Nov 6 the selling moved east: South Korea's KOSPI triggered a 20-minute circuit breaker before closing 3.9% lower at 5,174.84, and the Nikkei fell as much as 7% intraday.

The trigger was political, not operational. A Democratic midterm sweep revived the prospect of a national data-center moratorium — Senator Bernie Sanders has called for one — and with it the risk of a slower AI build-out, which strategists described as a broader economic slowdown with reduced prospects for the AI cycle. The cross-asset signature of forced selling followed: the VIX closed at 28.18 after touching 29.64, the 10-year Treasury yield ended the day at 4.837%, bitcoin fell 6.1% to $75,657, and crude held above $91 while gold slipped to $4,055.

Hedge funds have cut gross equity exposure 7.8% from its recent peak, the largest reduction in the Morgan Stanley prime-brokerage series since it began in 2016. Funds are now underweight the broad AI trade for the first time since early 2023, and on the heaviest day they liquidated owned shares roughly 3.5 times faster than they opened shorts. That is a crowd in retreat, not a crowd that has finished selling: AI holdings still sit in the top 17% of the past three years, so the crowding has been thinned, not washed out.

The clearest evidence of the rotation came from a fund making it in public. On Nov 5, Samlyn Capital — a roughly $7 billion fund run by Rob Pohly — announced a restructuring that included layoffs of senior analysts and pivoted its book toward financial stocks. The Wall Street Journal separately reported that Samlyn, Engineers Gate and Walleye cut staff, and that October produced the highest rate of stock-picking funds unwinding positions in more than a decade.

Read the dispersion, not the average. The two banks that beat most — Goldman Sachs and Morgan Stanley — earned it from the volatility that is now liquidating their clients. The two that missed, JPMorgan and Bank of America, carry the rate-sensitive balance-sheet trade the rotation is supposed to reward, and Bank of America took a $1.2 billion US regulatory penalty on top of a 17% fixed-income decline. That makes the new crowded long a bet on rate stability and persistent volatility at the same time — two conditions that can both break in the same week.

The mechanism matters. Net interest income is the ballast: JPMorgan earned $27.0 billion in the quarter, Citigroup $17.0 billion, Bank of America $16.4 billion and Wells Fargo $12.5 billion, with the 10-year Treasury yield ending Nov 5 at 4.837%. Trading revenue is the payload: it exists only while clients are panicking and re-trading. So depends on the panic continuing — the same quarter that justifies buying banks is a bet that the crowd's distress does not stop, which is a much thinner foundation than the sector's headlines suggest.

The credit channel is where this becomes a loss rather than a mark-down. Hedge funds have been massive sellers of AI-related debt, according to Andrzej Skiba of Royal Bank of Canada's BlueBay unit, describing a capitulation inside the space. A portfolio of roughly $500 million of data center loans went to auction with bids due Nov 19 and a $10 million minimum, including a $25 million slice backing a New Mexico site of OpenAI's Stargate venture with Oracle and SoftBank Group, plus loans tied to Vantage Data Centers and CyrusOne. The first clean price for AI infrastructure credit is being set in a loan auction, not on an exchange.

The funding math is what the auction is testing. Treasury's own data show aggregate net leverage of 0.28x across the tech ecosystem against 1.3x for the S&P 500 excluding financials, with roughly $600 billion of tech-related debt projected for 2026 and more than $700 billion for 2027, including an estimated $735 billion of 2027 tech debt issuance against about $1.3 trillion of total capital raising. Low net leverage does not make the buildout funding-neutral — it makes the whole thing dependent on continued access to debt markets, and the crowd that supplied that appetite is now in the auction as a seller.

Near-term, the calendar is unusually dense. The Nov 19 auction bids are the first public mark on AI infrastructure credit, January brings fourth-quarter bank earnings that will show whether trading revenue survives the crowd's shrinkage, and Korean authorities have already stepped in to calm volatility after the circuit breaker — a backstop that can slow a deleveraging without reversing it.

Over one to three years the consequences are structural rather than tactical. Hedge fund gross exposure normalising means smaller prime-brokerage and financing books for the same banks now being bought as a refuge. If the AI capex cycle survives the new Congress, upstream earnings stay intact — SK hynix's storage order book is booked through 2029 — while the multiple keeps compressing. And as the hedge funds dump AI debt, private credit and bank balance sheets absorb it, which sets the terms of the next leg of the buildout.

The trade that survived the rout is the trade the rout created. Hedge funds liquidated almost everything they owned, and what they did not liquidate is financials, in public, from a fund that restructured around bank stocks while cutting staff. The AI basket meanwhile has no marginal buyer: leverage is 7.8% below peak, net buying since 2025 is erased, and mutual funds sit in the 10th percentile of sensitivity to the Magnificent Seven. Banks are therefore the only crowded long left standing, which is precisely the risk — the sector's record trading revenue came from the positioning now collapsing, and its rate sensitivity makes it the worst place to be long if the Fed turns.