Leopold Aschenbrenner's Situational Awareness AI fund filed its most recent 13F disclosure showing a concentrated US equity book in memory stocks before a 67% drawdown. The fund, managed by the former OpenAI researcher turned allocator, built positions across the memory semiconductor complex in late 2024. The filing confirms what family offices suspected in February: the AI infrastructure trade had become a levered bet on a single layer of the stack.

The 13F shows exposure tilted toward US-listed memory names at a moment when Chinese AI labs were quietly designing around expensive high-bandwidth memory. Aschenbrenner's thesis—that frontier model scaling would require exponential growth in memory capacity—was directionally correct but mispriced the velocity of architectural innovation. DeepSeek's late-January release of a reasoning model trained on older, cheaper hardware triggered immediate re-rating across the memory complex. The fund's positions unwound without the liquidity cushion that marks institutional-grade risk management.

This matters because Aschenbrenner is not a retail momentum chaser. He wrote the most widely circulated memo on AI timelines in venture capital history, then raised capital on the strength of that research. His allocator base includes family offices and fund-of-funds operators who trusted the technical depth. A 67% loss on a concentrated book signals either catastrophic position sizing or a fundamental misread of how quickly AI infrastructure demand can pivot. Both explanations are expensive for the Limited Partners who wired capital based on the OpenAI pedigree.

The filing also clarifies the timeline. Memory stocks peaked in mid-December 2024, began their slide in early January, then collapsed after DeepSeek's January 20 announcement. Aschenbrenner's fund was still holding when the sector broke. That suggests either conviction that held past the technical breakdown or portfolio construction that made exits impossible at scale. Neither scenario improves the forward risk profile for the strategy.

Allocators should watch three things. First, whether Situational Awareness AI files a Schedule 13G in the next forty-five days indicating activist-style stakes that would explain the concentration and the illiquidity. Second, whether any of the memory names on the 13F report secondary offerings in Q2, which would confirm that institutional holders are using new issuance to exit. Third, whether family offices that backed Aschenbrenner begin requesting separate managed accounts with explicit sector caps, which would appear in Form ADV amendments by mid-Q2.

The sector rotation is no longer a risk. It already happened. The question now is whether AI-themed allocators learned to separate thesis depth from position sizing discipline.