Martin Shkreli breaks down the Situational Awareness collapse: leverage math, margin calls, and Ken Griffin's shadow banking role

Situational Awareness collapse: leverage math, margin calls, and Ken Griffin's shadow banking role

Leopold Aschenbrenner's Situational Awareness fund ran at roughly 4x leverage on a reported $45 billion in assets under custody, which translates to approximately $120 billion in gross market value. That math is the story. A 25% drop in GMV doesn't sound catastrophic until you realize the equity underneath it collapses from ~$35 billion to ~$5 billion — at which point no prime broker stays patient.

Martin Shkreli, who says he had overlapping positions and heard rumors crystallizing late last week, estimates Anthropic represented roughly $10 billion of the fund's book. When margin calls came, Aschenbrenner reportedly contacted around 10 parties over the weekend to place the Anthropic stake, with the offer priced at an implied $1.1 trillion Anthropic valuation. Shkreli says about half of that stake may have been sold, though the buyer remains unclear. Shkreli and his team were separately offered a look at $100 million of Anthropic stock through what appeared to be an SPV — a structure he read as a possible probe to gauge appetite for a much larger sale.

The Citadel trade

Three firms were brought in to bid on the remaining book: Jane Street, Millennium, and Citadel. Citadel's bid won. Jane Street was reportedly an LP in the fund and did not bid, though may have taken Anthropic stock separately.

The buyer of the public book is said to have received an immediate mark-to-market gain of $3 to $4 billion on the trade. Shkreli frames Ken Griffin's positioning as deliberate brand-building — the same role Citadel played in the Amaranth blowup and others, stepping in where banks once would have absorbed the risk. Griffin is, in Shkreli's read, constructing himself as the financial system's buyer of last resort for the post-Buffett era. Citadel is reportedly up for the month of July, one of the few large funds to be so.

“They were at 45,000,000,000, you know, 10,000,000,000 of that is in Anthropic... So you have $30,000,000,000 of cash in your bank account and running 4x levered means you have a 120,000,000,000 gross market value. If your GMV drops 25%, your equity drops from 35,000,000,000 to 5,000,000,000. No prime broker is gonna let you keep 90,000,000,000 of gross market value... Three firms were bidding on the assets: Jane Street, Millennium and Citadel.”

How a forced liquidation actually works

Selling a position at this scale is not a market-sell button. Shkreli walks through the mechanics: selling "into the screens" quietly is possible in liquid names, but for concentrated AI infrastructure positions — some trading at ten days of average daily volume — hitting the open market would crater the price 50% or more before you're out. The typical path is a phone call to Goldman or Bank of America, who then advertise the order through their four-digit market maker ID. Sophisticated counterparties sniff out the seller by process of elimination: call Fidelity, call the ETFs, call the index funds. If everyone says they're not selling, it's probably the distressed fund.

Once the prime broker decides to act, it's no longer the fund manager's choice. The bank's calculus is simple: take a known loss now rather than risk a larger one later. Post-Archegos, prime brokers don't wait. The buyer who can actually absorb the book, Shkreli argues, needs the liquidity to hold for five years without blinking, because other funds will immediately short the portfolio trying to force another capitulation. Only a Citadel-scale balance sheet can credibly make that commitment.

The leverage acquisition process

At the scale Situational Awareness was operating, leverage comes from multiple prime brokers simultaneously. The prime broker's business model depends on it — a 1% financing spread on 4x leverage translates to roughly 400–600 basis points of margin on enormous capital bases. The risk desk's constraint is concentration and illiquidity, which is precisely where Situational Awareness was most exposed. Holding $10 billion in Anthropic — a private, illiquid position — inside a leveraged hedge fund structure is what Shkreli describes as a historically reliable warning sign. In his read, the moment a hedge fund starts treating private investments like a VC portfolio, it's usually the beginning of the end.

The AGI narrative as risk factor

Shkreli draws a direct line between Aschenbrenner's intellectual conviction on AGI and his inability to hedge or exit. Funds that get "spellbound by the narrative" — AGI in this cycle, dotcom in 2000, natural gas in Amaranth's case — typically ride it past the point where rational position sizing would have forced a trim. He compares Aschenbrenner to a succession of concentrated cycle believers: Ryan Jacob's internet fund in 2000, Gerald Tsai's Manhattan Fund in the 1960s, Cathie Wood more recently.

The broader damage may not be contained to one fund. Shkreli expects July numbers from other tech-focused hedge funds to show 30–40% drawdowns, with many running variants of the same AI infrastructure trade, often entered later and with more leverage as they tried to catch up. The gross exposure across the sector could be five to ten times Situational Awareness's $100 billion GMV alone.

Kelly criterion and the universal overbetting problem

Shkreli closes with a structural argument that applies well beyond this episode: almost every trader, retail or institutional, bets two to ten times the Kelly-optimal position size. The Kelly criterion — developed by Bell Labs researcher John Kelly — defines the optimal bet as your edge minus the reciprocal of that edge. At a 55% win rate, the optimal bet is 10% of capital. Most funds behave as though they're running at four or five times that implied edge, and simulation shows that even a 60/40 edge leads to ruin if the position size is wrong. Shkreli says he learned this painfully from watching a former SAC Capital manager who kept 90% of his book in cash, made tiny trades, never had a down quarter across more than 20 years, and compounded at roughly 30% annually — then immediately ignored all of it when he had his own capital to deploy.

Aschenbrenner reportedly told Dwarkesh Patel in a recent appearance that there was "100x left" before AGI, having already made roughly 20x since inception. The collapse came before he could stay in the game to find out.