Announcement
Mar 14, 2026
The Fired OpenAI Researcher Who Turned $225M Into $5.5B — By Betting Against the Hype
The Essay That Started Everything
In 2024, Leopold Aschenbrenner was a 23-year-old researcher on OpenAI's superalignment team. By early 2026, he was running a $5.5 billion hedge fund. The gap between those two facts tells you everything about where AI is actually headed — and where the real money is flowing.
Before the fund, there was the thesis. In mid-2024, Aschenbrenner published Situational Awareness: The Decade Ahead — a 165-page monograph arguing that AGI could arrive as soon as 2027. It wasn't a blog post or a Twitter thread. It was a rigorous, deeply technical document that laid out the trajectory of AI scaling, the geopolitical implications, and the infrastructure requirements in granular detail.
Shortly after, he was fired from OpenAI. The details remain murky, but the timing was notable: a young researcher publishes a sweeping public thesis about the pace of AGI development, and his employer — one of the most secretive AI companies on earth — shows him the door.
What happened next was more interesting than the firing itself.
From Thesis to Fund
Aschenbrenner didn't retreat. He launched Situational Awareness LP, a hedge fund built entirely around the thesis he'd laid out in public. The fund was seeded by a who's-who of Silicon Valley: Nat Friedman (former GitHub CEO), Daniel Gross, and Patrick and John Collison — the founders of Stripe.
The initial raise was $383 million by early 2025. By October 2025, the fund had grown to $1.5 billion, posting 47% gains in the first half of the year alone. By February 2026, according to Fortune, assets had ballooned to $5.5 billion across roughly 30 holdings in US equities.
Those numbers alone would make headlines. But the portfolio composition is what makes this story worth studying.
The Thesis: It's Not About the Chips
Here's where most investors — and most businesses — are getting it wrong.
The prevailing AI trade for the last three years has been straightforward: buy NVIDIA, buy the model companies, buy the software layer. Aschenbrenner's argument is that this phase is already priced in. The bottleneck has moved.
If AGI is arriving on the timeline he projects, the constraint isn't compute chips. It's everything those chips need to actually run: power generation, physical connectivity, and data center capacity.
Aschenbrenner sold his NVIDIA position entirely. Instead, he built a portfolio around the infrastructure that makes AI physically possible.
Three Layers of Infrastructure
Layer 1: Power Generation — AI data centers consume staggering amounts of electricity, and the existing grid cannot keep up. Building new grid capacity takes three or more years. Bloom Energy manufactures solid oxide fuel cells that can deliver power to a data center in as little as 90 days. Other positions include Vistra and Constellation Energy.
Layer 2: Connectivity and Hardware — The models need high-speed optical interconnects, networking infrastructure, and specialized hardware to move data between thousands of GPUs at scale. Positions like Lumentum, Coherent, Broadcom, CoreWeave, and Intel address this layer.
Layer 3: The New Data Centers — Perhaps the most contrarian bet. Bitcoin mining operations like Core Scientific, IREN, and Applied Digital are pivoting to AI workloads. The infrastructure already exists — cooling systems, power connections, physical facilities. Repurposing them is faster and cheaper than building from scratch.
The Numbers in Context
A 24-year-old with no prior fund management experience raised hundreds of millions from some of the sharpest capital allocators in technology, built a concentrated portfolio around a single macro thesis, and generated returns that put most established hedge funds to shame.
The $225M-to-$5.5B trajectory represents one of the fastest asset growth stories in recent hedge fund history. The 47% first-half return in 2025 wasn't driven by a lucky options trade — it was the result of a structural thesis playing out exactly as predicted.
What This Means for Businesses Building With AI
Infrastructure is the real dependency. If you're building AI into your business, you're implicitly making a bet on power availability, data center capacity, and network infrastructure. When inference demand outstrips available compute — and it will — the companies with secure infrastructure access will have a decisive advantage.
The picks-and-shovels logic applies to operations, not just investing. The businesses that win with AI won't be the ones chasing the latest model release. They'll be the ones that built robust, infrastructure-aware systems that can actually run reliably when demand spikes.
Concentration risk is real. If your entire AI strategy depends on a single cloud provider or a single model API, you're exposed to exactly the kind of infrastructure constraints Aschenbrenner is betting on. Diversification at the infrastructure level isn't paranoia. It's preparation.
The biggest opportunities in AI aren't in AI itself. They're in what makes AI physically possible.
Sources: Fortune (March 5, 2026), "Situational Awareness: The Decade Ahead" by Leopold Aschenbrenner (2024), barebone.ai portfolio analysis.
Changelog