After suffering a historic 67% drawdown in July, Situational Awareness, managed by former OpenAI researcher Leopold Aschenbrenner, reinvested in AI infrastructure assets in September. Market data shows that in early September, the fund purchased seve

2026-10-16

After suffering a historic 67% drawdown in July, Situational Awareness, managed by former OpenAI researcher Leopold Aschenbrenner, reinvested in AI infrastructure assets in September. Market data shows that in early September, the fund purchased seven assets—AMD, Bloom Energy, CoreWeave, Intel, SK Hynix, SanDisk, and the Roundhill Memory Chip ETF—using customized call options, paying approximately $315 million in option premiums, corresponding to an initial Delta exposure of approximately $1.1 billion. This investment comes just over a month after the leverage crisis in July. At that time, the fund held a highly concentrated position in AI infrastructure stocks and used leverage to amplify its exposure. The sharp drop in related assets triggered margin calls, forcing the fund to significantly reduce its holdings. Subsequently, Situational Awareness reduced leverage and shifted from stocks and total return swaps to fully paid call options to mitigate the risk of margin calls and forced liquidation due to falling stock prices. However, if the option price does not reach its strike price before expiration, the invested premium could still be significantly reduced or even wiped out. It's worth noting that the assets reinvested in September largely overlapped with those heavily invested in July, covering AI computing, memory, data center computing power, and power. Entering October, CoreWeave, SanDisk, Bloom Energy, and AMD experienced significant sell-offs again, with the market refocusing on AI capital expenditure returns, data center financing costs, and the risks of high valuations. This volatility is insufficient to determine the final profit or loss of Leopold's options trading, but its reinvestment indicates that its assessment of the long-term demand for AI infrastructure remains unchanged.