Citadel cleared $4bn of a rescued AI book in three weeks. The discount was the trade.

The discount is where this trade lives. Citadel took the book at about 10% below marks in late July, after Leopold Aschenbrenner had already tried to raise cash by selling positions one at a time and found it difficult ^[5][6]. Work backwards from the disposals. If $4bn of block trades cleared more than 80% of the portfolio, the whole thing was worth roughly $5bn or less ^[2], which puts the discount somewhere near $500m before a single price moved ^[3]. That is what the absence of competing bids costs, and it is close to what Ken Griffin told investors in his letter, that Citadel was the only firm capable of executing a solution of that scale in that timeframe ^[14].

The execution detail is the useful part. Roughly 100 blocks over three weeks works out to an average near $40m a trade ^[1] and something like $1.3bn a week ^[6]. A portfolio whose liquidation the market had been afraid of went out the door at that pace because the concentration sat in names with deep two-way markets. SanDisk and Micron, among SA's larger holdings, rebounded quickly once the overhang was gone ^[9].

The failure structure deserves more attention than the rescue. SA was long AI-related stocks and short the software companies AI was expected to damage ^[4]. When the correction that began in June intensified in July ^[15], SanDisk and Bloom Energy fell more than 50% while Adobe, a short, went up ^[7]. Both legs were expressions of the same forecast, so the short book was not protection, it was the same bet with the sign reversed. Margin calls and forced selling followed ^[8].

Assets had passed $20bn inside two years ^[3][4]. An equity book of roughly $5bn is about a quarter of that ^[4], which tells you either how much had already been liquidated before Citadel arrived or how much of the risk was never in long equities at all. Griffin, who rarely writes to investors ^[13], reported a 6% July for the flagship fund while rivals lost money over the same period ^[11]. Nobody has published a realised number on the SA book; the market simply assumes it was large ^[10]. Just under a fifth of the acquired portfolio has not been reported sold ^[5], and that remainder is the only part of the position still carrying price risk.