The Rhyme: AI's Biggest Fund Sells Everything & Its 1998 Echo

"History doesn't repeat… but it rhymes." — Mark Twain

The smartest fund in artificial intelligence just sold everything it owned. Situational Awareness LP, the $22 billion hedge fund built by 24-year-old former OpenAI researcher Leopold Aschenbrenner, dumped its entire public equities portfolio in a single block trade to Ken Griffin's Citadel on Thursday after margin calls from Goldman Sachs, JPMorgan, and Bank of America. The fund had returned 439% through June on a four-times-leveraged bet that AI infrastructure stocks would outrun the Magnificent Seven. Then July happened, and the stocks that made him a billionaire on paper turned the leverage against him overnight.

SA peak AUM: ~$22B | Leverage: 4x | Returns thru June: +439% | Fund age: 22 months | Buyer: Citadel | 30-yr yield: 5.24%

Twenty-eight years ago, another genius fund with a different thesis and the same fatal flaw made the same phone call.

The office was on the third floor of a low-slung building in Greenwich, Connecticut, the kind of place that looked nothing like what it was. No marble lobby, no oak-paneled conference rooms. Just screens, models, and the quiet hum of money multiplying itself. In February 1994, John Meriwether, the former vice chairman of Salomon Brothers and the man whose bond arbitrage desk had generated 80 to 100 percent of the bank's global profits through the late eighties and early nineties, sat down and began trading with $1.01 billion in capital.

The partners he assembled read like the guest list for a Nobel ceremony, because two of them were. Myron Scholes and Robert C. Merton had built the Black-Scholes model, the formula that turned options pricing from guesswork into science. They would share the Nobel Prize in Economics in 1997, one year before their fund nearly broke the global financial system. Long-Term Capital Management did one thing: it found bonds that were priced slightly apart from where the math said they should be, and it bet the gap would close. Government bonds against corporate bonds. Italian against German. The spreads were tiny, so the fund used leverage to make them matter. By early 1998, LTCM held $4.7 billion in equity and $129 billion in assets. The leverage ratio was 25 to 1.

For four years it worked. Annualized returns after fees ran 21 percent the first year, 43 the second, 41 the third. Investors begged to get in. Meriwether turned most of them away and actually returned $2.7 billion in late 1997 because the fund didn't need the money. Then Russia defaulted on its domestic debt on August 17, 1998, and the world stopped acting the way the models said it should. Money fled to the safest assets on earth. Every spread LTCM was straddling blew wide open at once. With 25 to 1 leverage, a four percent move against the book wiped out the equity. By the end of August the fund had lost $1.85 billion in a single month.

By September 22, equity had crumbled to roughly $600 million against well over a hundred billion in positions. The fund could not sell without cratering the markets it was selling into. Victor Haghani, one of the partners, said it felt like someone else had their exact portfolio, three times as large, and was dumping all of it at once. The next morning, the Federal Reserve Bank of New York called 14 Wall Street banks into a room and told them they were going to write checks. They assembled $3.625 billion. It was not a gift. It was a tourniquet.

Alan Greenspan cut the fed funds rate a week later, from 5.50% to 5.25%. He cut again on October 15 in a rare inter-meeting surprise, to 5.00%, and again on November 17, to 4.75%. The S&P 500, which had peaked near 1,190 in July and bottomed around 923 on October 8, rallied 33% by year's end to close near 1,229. The trades LTCM had been forced to unwind at fire-sale prices came back. The 14 banks made money on the bailout. Meriwether started a new fund the following year.

Both funds were right about the thesis and wrong about the leverage. Both learned that markets can stay irrational longer than you can stay solvent, even when you are the smartest person in the room.

Here is what happened after Greenwich ran out of air. The S&P 500 bottomed on October 8, 1998, at 923, down 22% from its July high near 1,190. Greenspan's three rate cuts brought the fed funds rate from 5.50% to 4.75% in less than two months. Money flooded back into risk assets. By December 31, the index had climbed to roughly 1,229, erasing the entire decline and adding to it.

The LTCM trades that were liquidated at fire-sale prices recovered. The spreads converged exactly as the models predicted. The thesis was right. The leverage was the problem. But the aftermath carried a longer tail that nobody wanted to talk about. The rate cuts Greenspan delivered to calm the crisis poured fuel on the dot-com rally that was already running hot. The Nasdaq, which bottomed near 1,500 in October 1998, doubled to 3,000 within a year and hit 5,048 by March 2000 before collapsing 78%. The medicine for the 1998 crisis became the disease of 2000.

That is the pattern that matters now. If the Aschenbrenner liquidation is the last forced seller, if Citadel buying the book at a discount marks the low for AI infrastructure names like CoreWeave, Bloom Energy, and SanDisk, then those stocks are cheaper today than they were yesterday for reasons that have nothing to do with fundamentals. When a portfolio hits the tape at four in the morning because a prime broker sent a margin call, price discovery is not happening. Liquidation is happening. And the difference between those two things is where the edge lives.

But Warsh is not Greenspan. There are no rate cuts coming to cushion the landing. The bond market is screaming that inflation is not finished. And the AI infrastructure trade is still crowded in funds that have not yet gotten their phone call from Goldman.

In 1998, forced selling by the smartest fund on Wall Street marked the low, not the beginning. But Greenspan had three rate cuts to give. Warsh has none. Watch who else gets a margin call before you assume the clearing is done.

The 30-year Treasury yield has now closed above 5% for 12 consecutive sessions, the longest streak since 2007. The last time long-term yields stayed this elevated for this long, the housing market was six months from cracking. The year to study is 1994, when the bond vigilantes last broke a sitting president's fiscal agenda and forced Bill Clinton to pivot. They are trying it again.