Jim Rickards says AI companies are sitting on hidden debt that could trigger a systemic crisis. The full career arc behind this warning sits in the Jim Rickards dossier.
He has been here before. In 1998, he was in the room when the last one almost happened.
What LTCM Was
Long-Term Capital Management was a hedge fund with a brain trust that read like a Nobel Prize committee meeting — Myron Scholes, Robert Merton, and John Meriwether were not gamblers but mathematicians who built models to find tiny pricing differences in bond markets. The trades were small winners, but small winners do not make you rich unless you bet big on them, so LTCM borrowed heavily.
At the end of 1997, the fund was holding about $30 in debt for every $1 of capital, according to the Federal Reserve History archives. That was the stable number. In practice, some of their derivative positions pushed leverage past 100:1. The fund had $125 billion in assets on a capital base that was a fraction of that.
The returns were spectacular — 43 percent in 1995, 41 percent in 1996. Then the floor dropped out.
The Analogy That Matters
Here is the simplest way to understand LTCM’s setup.
Imagine a guy who finds a penny on the ground every time he walks down the street. Good business, reliable, math checks out — except he borrowed $100 for every $1 he had, so he could pick up $100 worth of pennies per walk.
One trip goes wrong and he drops a nickel — he does not lose a penny, he loses everything. That was LTCM: the trades worked individually, but the leverage killed them collectively.
The Collapse
August 1998. Russia defaulted on its debt and devalued the ruble. LTCM’s models said this was essentially impossible. The fund had built its entire book around convergence — the assumption that spreads would narrow. Instead, they blew out in every direction.
LTCM lost 44 percent of its value in August alone. The problem was not the loss but what the loss exposed — LTCM had positions with 14 major banks as counterparties, and if LTCM went down those banks took the hit, and if those banks took the hit the dominoes started falling. The Federal Reserve did the math and realized a fire sale of LTCM’s positions would destabilize global markets.
The Rescue
On September 23, 1998, after a day of frantic negotiation that included a last-minute bid from Warren Buffett that fell through due to legal issues, 14 banks agreed to inject $3.625 billion into LTCM in exchange for 90 percent of the fund.
Jim Rickards was the principal negotiator. He sat across the table from the Nobel laureates who founded the fund and hammered out the terms. He represented the banks — the counterparties who had the most to lose. The LTCM rescue story covers what that negotiation did to his worldview; what follows is the leverage lesson he took from it.
This is not a detail Rickards drops into every newsletter. It is the foundation of his entire worldview on risk.
What Rickards Learned
The LTCM experience taught him something that most economists never learn in a classroom. It is the same lesson he later carried into the 2006 CIA warning that preceded Lehman by 18 months.
Systemic risk does not come from the obvious threats. It does not come from the thing everyone is watching. It comes from hidden leverage — debt that is off-balance-sheet, buried in derivatives, hidden in counterparty arrangements that nobody can see until it is too late.
The banks that lent to LTCM thought they were safe with collateral, margin calls, and model validation, but what they did not have was visibility into the total picture — every bank saw its own exposure while no bank saw the system. That is exactly what Rickards is arguing about AI companies today.
His thesis, laid out in his current AI Black Paper campaign, is that the big AI players — the hyperscalers, the foundation model startups, the infrastructure builders — are carrying massive hidden debt — not bonds or bank loans, but off-balance-sheet commitments: compute leases, data center construction contracts, GPU financing structures, revenue-sharing guarantees. The kind of obligations that look like operating expenses on paper but function like debt in a downturn.
The numbers are staggering. The AI infrastructure buildout is projected to run into the trillions over the next five years. A lot of that is financed through vehicles that look nothing like the bond market but carry the same risk. If the revenue does not materialize — if AI adoption plateaus or the killer app takes longer than expected — the margin calls on those structures would cascade in a pattern similar to LTCM’s. Rickards is not arguing that AI is a bubble; he is saying the leverage behind AI is invisible, and invisible leverage is what he has spent his career hunting. The MoneyGPT argument lays out the book where he connected this lesson to algorithmic finance.
The Honest Take
Rickards’ LTCM experience gives his AI debt thesis credibility that most newsletter writers cannot match. He is not reading about leverage in a textbook. He negotiated the resolution of the most famous leverage crisis in history. When he says off-balance-sheet debt can blow up without warning, he is speaking from direct experience.
But his tendency to see LTCM-like risk everywhere is also his blind spot. Not every leveraged structure is LTCM. LTCM blew up because it was levered 30:1 into a single correlated bet on convergence. An AI company with a $5 billion GPU lease and a $50 billion valuation involves a different time horizon and counterparty structure. The underlying asset — compute capacity that can be repurposed — is not a bond spread that evaporates overnight.
The leverage and structures are real, and whether the market has priced in the risk he describes is a different matter. Whether the market has priced in the risk he is describing is a different matter — one that the bond market itself is still working through. More from Rickards in the Guru Files.