The Intern Who Shook OpenAI

In 2024, Leopold Aschenbrenner — a 23-year-old researcher on OpenAI's Superalignment effort — left the lab and published Situational Awareness: The Decade Ahead, a long thesis arguing that advanced AI is arriving faster than most institutions expect and that national-scale dynamics will shape who benefits.¹ Within months he launched Situational Awareness LP (SALP), positioning it as an investor with a deep technical read on where the value chain will accrue.²

What the fund is (and isn't). SALP isn't a quant "AI that trades" shop; it's a fund that invests because of AI. Think "picks-and-shovels" plus selective platform bets: semiconductors, power and datacenter infrastructure, model and tooling layers, and a small number of exposure bets in public names and late-stage private deals.² Its early disclosures show a focus on the AI supply chain more than splashy app-layer speculation.³

Capital and performance. Public filings show SALP starting life in 2024 with a few-hundred-million under management (SEC-tracked; ~$383M as of Dec 31, 2024).³ By mid-2025, major financial press reported more than $1.5B AUM raised rapidly, with ~47% returns in the first half of the year — a number that significantly outpaced broad indices during the same period.² ⁴ ⁵ Those results came in a market still digesting AI demand, suggesting SALP timed the "infrastructure super-cycle" well.² ⁴

The principles behind the strategy.

• Asymmetric timeline view. The core thesis from Aschenbrenner's essay — that meaningful AGI-level capabilities are plausible this decade — pushes the portfolio toward capacity bottlenecks: compute, power, memory, and networking.¹ ²
• Concentration where edges exist. Early SALP commentary and holdings imply concentration rather than spray-and-pray. The edge is informational (deep technical networks; constant thesis-testing with builders) rather than purely factor exposure.² ³
• Risk controls via hedges. Coverage notes describe long–short elements and macro hedges (e.g., semis or power cyclicality) to dampen drawdowns if AI exuberance cools.² ⁴
• Governance & safety as an investable theme. The fund treats AI reliability, alignment, and MLOps safety as its own category (monitoring, interpretability, compliance), consistent with the "infrastructure + guardrails" view from his writing.¹ ²

Why it worked (so far). First, the macro was on-side: hyperscalers pulled forward capex for GPUs, data centers, and grid upgrades, lifting suppliers across the chain. Second, SALP's narrow map of where value accrues avoided the noisiest app stories and rode the parts of the stack with the cleanest fundamentals.² ⁴ Third, the credibility loop mattered: a widely read thesis, a network of top operators, and quick scaling of AUM brought deal flow and co-investors that many new funds can't access.¹ ² ⁵

For executives, the lesson isn't about replicating a hedge fund. It's about where to place your own bets: capacity, reliability, and the systems that let AI scale safely. The money followed those fundamentals for a reason.