Fundraising as reported by the Financial Times; fund positioning and performance context from public 13F filings and prior coverage (Fortune).
Leopold Aschenbrenner’s AI-native hedge fund built one of the most structurally coherent AI trade architectures on record — and the July rout is testing whether coherence is enough.
What Happened
The Financial Times reported this week — corroborated by Bloomberg — that Situational Awareness LP is seeking to raise fresh capital. The timing, following the chip-and-AI-infrastructure rout that accelerated around July 28, has drawn attention. But the fund’s context warrants precision before interpretation. Situational Awareness launched in 2024 with a roughly $1.5 billion raise, founded by Leopold Aschenbrenner, the former OpenAI researcher whose 2024 essay of the same name became the definitive AGI-imminence manifesto in Silicon Valley. Since launch, the fund has grown past $20 billion in assets under management. Its most recent 13F filing disclosed a notional book of approximately $13.7 billion — up roughly 148% quarter over quarter — a figure representing disclosed long US equity positions and option notionals, not a fundraising total. Jane Street is among its investors. By multiple accounts the fund was up sharply in 2026 and many times over since inception.
The fund’s disclosed Q1 2026 positioning was, on paper, constructed for exactly this kind of semiconductor dislocation. Its 13F showed approximately $8.46 billion in notional PUT exposure bet against chipmakers — Nvidia, the SMH semiconductor ETF, Broadcom, AMD, Micron, ASML, and TSMC — the precise cluster of names that cratered in the KOSPI-led chip rout. On the long side: crypto miners including Riot and IREN, power infrastructure play Bloom Energy, AI-compute names CoreWeave and Nebius, and approximately 20% of assets in Anthropic.
The FT reporting does not specify the size of the new raise, and critically, it does not settle whether the move is opportunistic — deploying fresh capital into dislocated AI-infrastructure names — or defensive, managing liquidity after a sharp drawdown in the long book. Both readings are consistent with the public facts. Neither is established.
The key insight: The ~38% figure circulating in coverage is a third-party reconstruction of Situational Awareness LP’s disclosed 13F long positions only — not the fund’s net asset value. A 13F captures long US equity holdings and notional option exposure; it does not show the profit and loss on short bets. If the ~$8.46 billion in semiconductor put exposure performed as designed during the July chip rout, it likely offset — or generated gains against — the same selloff that crushed the longs. The fund’s actual net performance through the rout is genuinely unknown from public data. Treating the 38% as the fund’s return is a category error.
The Structural Read
Situational Awareness LP was built around one of the most internally coherent investment theses in the AI cycle: AGI is coming, the bottleneck is physical, and the real value accrues to whoever controls power and compute — not the chip fab. That’s why the fund went long crypto miners, energy infrastructure, and data-center operators while shorting the chipmakers it viewed as over-owned relative to the actual value-capture layer. The memory wall thesis — that chip architectures face a fundamental bandwidth constraint as AI models scale — underpinned the short side. The long side was a direct bet on the infrastructure buildout documented in Beyond NVIDIA’s Moat.
What the July rout exposed is the reflexivity built into this trade. The infrastructure names — miners, Bloom Energy, CoreWeave — had soared precisely because the AI buildout thesis was being priced in. When chip-cycle fear hit, the market didn’t discriminate between chipmakers and the infrastructure that depends on them. The same narrative that drove the longs up unwound them simultaneously. Being right about the long-run trajectory of AI offered no protection against correlated de-risking. This is the market-side mirror of the capex-versus-returns stress test the week delivered through the KOSPI, a hawkish Fed, and mega-cap earnings — the same theme mapped in The Map of AI Redrawn.
The Anthropic stake adds a separate dimension. Approximately 20% of assets sit in Anthropic equity — a position that connects directly to the circular dynamics explored in the Microsoft-Anthropic mark and the circular AI economy. Anthropic is not mark-to-market in the same way as the public book; its valuation is a private mark. That structural feature means the disclosed 13F long book reconstruction and the actual fund NAV diverge in ways that external observers cannot fully reconcile.
Map of AI — Reflexivity Thesis
The AI trade has no safe layer during a correlated unwind
In the Map of AI framework, the nine layers of the AI stack — from silicon to applications — are structurally distinct in terms of who captures value. But during a sentiment-driven de-risking event, correlations collapse toward one. A fund that correctly identified which layers would win long-term can still experience correlated drawdowns when the market reprices the entire buildout narrative simultaneously. The hedge (short chips) and the long (power, compute infrastructure) are both children of the same AI capex supercycle. When that cycle is questioned, the internal hedge becomes imperfect.
Three Implications
IMPLICATION 1 — The Fundraise Is Structurally Ambiguous, and That Matters
A fund raising into a dislocation is either deploying opportunistically into cheaper versions of assets it already holds — which is bullish on the thesis — or managing liquidity after a hard drawdown in the public book. The Financial Times reporting does not settle which. For investors trying to read the fundraise as a signal about the AI infrastructure trade, that ambiguity is the signal. It means the trade’s most sophisticated native practitioners are themselves operating under genuine uncertainty about timing, not just direction.
IMPLICATION 2 — 13F Reconstructions Are Structurally Incomplete for Complex Books
The wide coverage of the ~38% long-book drawdown illustrates a recurring problem in AI-fund analysis: 13F filings were designed for long-only equity disclosure, not for funds that run substantial short option books. When a fund’s most important risk positions — $8.46 billion in notional semiconductor puts — are the exact inverse of the narrative dominating coverage, external reconstruction of performance is not just incomplete; it is directionally misleading. The analytical standard for AI-native hedge funds needs to acknowledge this ceiling explicitly.
IMPLICATION 3 — The Aschenbrenner Thesis Is a Stress Test for the Entire AI Buildout Narrative
Situational Awareness LP is not just a fund — it is, in effect, a tradeable expression of the AGI-is-coming thesis that has shaped AI investment since 2024. Its positioning (short chip overcapacity, long physical compute and power infrastructure, long the lab most aligned with safety-AGI research) maps almost exactly onto the structural view that underpins the broader AI capex supercycle. If the July rout is a capex-versus-returns inflection — not a blip — then the fund’s post-rout fundraise is also a data point on whether institutional capital still believes the buildout thesis at current prices. That question has no clean answer yet.
