Leopold Aschenbrenner's Situational Awareness hedge fund peaked at nearly $45 billion in assets under management before being forced to unwind most of its positions last week. The 24-year-old former OpenAI researcher is already back raising fresh capital.
The Thesis: Maximum Conviction, Maximum Leverage
Aschenbrenner launched Situational Awareness in 2024 after being fired from OpenAI's Superalignment team. His thesis was straightforward: artificial general intelligence is coming faster than markets expect, compute infrastructure is the bottleneck, and the companies building it will capture extraordinary value. He ran the fund at 4x leverage, concentrated in CoreWeave, compute infrastructure plays, and crypto miners.
For 18 months, the trade worked. The fund grew from single-digit billions to $45 billion at its peak, riding the largest capital expenditure cycle in technology history. SpaceX alone is now building toward 10 gigawatts of compute capacity by 2027, according to SemiAnalysis, a buildout that could generate $300 billion in annual revenue. The physical infrastructure thesis was not wrong.
The Unwind: Leverage Kills
What collapsed was not the thesis but the portfolio construction. At 4x leverage, even a modest drawdown in AI infrastructure stocks triggers margin calls that cascade. When positions are concentrated and levered, the unwind becomes self-reinforcing. Forced selling in CoreWeave and related names created additional downward pressure, which created additional margin calls.
The sequence is familiar. Long-Term Capital Management in 1998. Archegos Capital in 2021. Directional conviction amplified by leverage creates a fragile structure. The trade can be right on a 3-year horizon and still blow up in a single week.
The Bearish Read
Aschenbrenner's unwind is a data point, not a thesis invalidation. AI capex is real: Microsoft, Google, Amazon, and Meta are collectively spending over $200 billion annually on infrastructure. But the market for AI infrastructure stocks had priced in perfection. When Google DeepMind's leadership exodus sent Alphabet shares down more than 5% last week, it revealed how thin the margin of safety had become in AI-adjacent names.
The risk now is that forced selling from leveraged AI funds creates a negative feedback loop in the very stocks that underpin the compute buildout thesis. If CoreWeave, Nvidia suppliers, and energy plays face persistent selling pressure from fund unwinds, the stocks can decouple from fundamentals for quarters.
The Bitcoin Angle
Aschenbrenner's fund held positions in Bitcoin miners alongside its AI compute bets, a logical pairing given that both mining and AI training compete for the same GPU and power infrastructure. Forced selling in MARA, RIOT, and CleanSpark creates a temporary dislocation. Bitcoin itself, trading 49% below its all-time high at roughly $63,000, is not directly affected by the unwind. But miner stocks face pressure from a seller who does not care about price.
The broader lesson is structural. When conviction trades get sized at 4x, the unwind damages everyone in the ecosystem, even assets with sound fundamentals.
Follow @UTXOMacro for daily Bitcoin, macro, and AI breakdowns.