Aschenbrenner Raises $5B Fund After OpenAI Firing — AI Safety Bet

In April 2024, OpenAI fired a 22-year-old researcher named Leopold Aschenbrenner. The stated reason was leaking internal documents. Aschenbrenner says the real reason was a memo he wrote to the board arguing the company was dangerously under-prepared for Chinese espionage targeting its AI research.

Eighteen months later, Aschenbrenner manages $5.5 billion.

This is not a redemption arc. It is something more specific: what happens when someone who genuinely believes AGI is coming in three years puts that belief into a financial instrument.

Who Is Leopold Aschenbrenner

Start with the biography because it is genuinely unusual.

Aschenbrenner entered Columbia University at age 15. He graduated at 19 as class valedictorian with a degree in economics and mathematics-statistics. By conventional metrics, he was already extraordinary. In 2023, at 22, he joined OpenAI's Superalignment team — the group tasked with figuring out how to control AI systems far more capable than current ones. The team was led by Jan Leike and Ilya Sutskever, two of the most respected researchers in the field.

He lasted about a year.

In April 2024, OpenAI terminated his employment. The company said it was over an information leak, a brainstorming document Aschenbrenner had shared with three external researchers for feedback. Aschenbrenner says this was a pretext. His account: he had written a detailed memo to OpenAI's board arguing that the company's security practices were dangerously inadequate given that Chinese state actors were actively trying to steal frontier AI research. He documented specific vulnerabilities. HR gave him a warning. A few weeks later, he was fired.

OpenAI denied the memo was the reason. Aschenbrenner says they told him directly that it was.

Both things can be true or neither can. What happened next is documented.

The Manifesto That Built the Fund

Two months after being fired, in June 2024, Aschenbrenner published a 165-page document titled "Situational Awareness: The Decade Ahead." He posted it online expecting a few thousand readers. It got hundreds of thousands.

The document is dense, serious, and alarming. It makes three core arguments:

AGI is coming faster than almost anyone in mainstream discourse believes. Aschenbrenner argues that extrapolating from current trajectory, AGI — AI that can perform any cognitive task a human can — arrives by 2027 or 2028. Not a rough approximation of human intelligence. AGI.

The US-China competition over AI is the most important geopolitical contest of the century. Whoever achieves AGI first gets a decisive economic and military advantage. OpenAI and Anthropic are not just companies. They are strategic national assets. Their security should reflect that.

Current AI safety practices are dangerously inadequate. The same security gaps he documented at OpenAI exist across the industry. Foreign intelligence services are actively exploiting them.

The document is not a financial prospectus. It is a geopolitical and technical argument. But every investor who read it had the same thought: if this is true, what do you buy?

Aschenbrenner had already figured that out.

The Fund: Situational Awareness LP

By late 2024, Aschenbrenner had founded Situational Awareness LP, a hedge fund named after the essay. The initial raise was approximately $225 million. The investors included Patrick and John Collison (founders of Stripe, collectively worth over $10 billion), Nat Friedman (former GitHub CEO), and Daniel Gross (partner at various Silicon Valley ventures). These are not credulous investors. They read the 165-page document and wrote cheques.

The fund's Q4 2024 SEC 13F filing showed a US equity portfolio of approximately $254 million.

The Q4 2025 13F filing, filed in February 2026, showed $5.52 billion in US equity positions across approximately 30 holdings.

That is a 21x increase in reported equity exposure in roughly 12 months. The fund now counts West Coast founders, family offices, institutions, and endowments among its investors. Aschenbrenner has invested almost all of his own net worth in the fund.

The Bet: Power and Compute, Not Models

Here is the contrarian thesis that drove those returns.

Every retail investor and major fund manager in 2024 asked the same question about AI: should I buy Nvidia? Should I buy OpenAI when it IPOs? Should I buy Microsoft, which owns a chunk of OpenAI? Should I buy the model companies?

Aschenbrenner asked a different question: what are the actual physical constraints on AGI arriving by 2027?

His answer: electricity and compute.

Training frontier AI models requires enormous amounts of both. The companies building the models (OpenAI, Anthropic, Google DeepMind, Meta) will compete ferociously for GPU clusters and power capacity. That competition benefits whoever owns the infrastructure, not whoever owns the model weights.

The most undervalued infrastructure, by this thesis: Bitcoin miners.

This sounds counterintuitive until you understand what Bitcoin miners actually are. They are companies that own massive power contracts, huge parcels of land with grid connections, and large amounts of computing hardware. The hardware depreciates. The power contracts and grid connections do not.

When AI labs need to build new data centres fast, they face two bottlenecks: getting grid connections (takes 3-7 years for new connections in the US) and acquiring land near grid infrastructure. Bitcoin miners already have both. They are pre-positioned AI infrastructure plays that the market was pricing as crypto plays.

The Specific Holdings

The Q4 2025 13F filing made the thesis concrete with numbers:

Core Scientific (CORZ) is the largest holding at approximately $418.7 million. It is the largest Bitcoin miner by hash rate in the United States. It emerged from bankruptcy in 2023 and has been converting some of its facilities to AI data centre colocation. It already has contracts with CoreWeave, the GPU cloud provider backed by Nvidia. Aschenbrenner is betting it converts more.

Riot Platforms (RIOT) at approximately $78.1 million. One of the largest publicly traded Bitcoin miners. 1.1 gigawatts of power capacity under development in Navarro County, Texas.

Hut 8 (HUT) at approximately $39.5 million. Canadian Bitcoin miner expanding into US AI infrastructure with a partnership with US Bitcoin Corp.

Bitdeer Technologies (BTDR) at approximately $20 million. Significant operations in Norway (cheap hydropower) and Texas.

CleanSpark (CLSK) at approximately $16.6 million. 22 facilities across the US focused on sustainable power.

Bitfarms (BITF) at approximately $16.2 million. Canadian miner with low-cost Quebec hydropower.

The pattern: every significant position is in a company that owns large power contracts in the United States or has access to low-cost power. The Bitcoin mining business is secondary. The power infrastructure is the asset.

Additionally, Aschenbrenner reportedly holds positions in IREN, Cipher Mining, and Applied Digital, smaller miners with similar power infrastructure profiles.

The fund has also, at various points, held short positions in semiconductor companies. In late 2025 the fund reported being long Intel while shorting Nvidia, TSMC, and Broadcom — a contrarian view that the AI chip supply chain will restructure in ways that benefit domestic US manufacturers over the current Taiwan-dependent supply chain. This is consistent with the geopolitical thesis in Situational Awareness: if the US-China AI race intensifies, supply chain independence becomes strategically critical.

Why This Worked

The 21x growth in equity exposure from 2024 to 2025 reflects a combination of new capital inflows and actual returns. The Bitcoin mining stocks Aschenbrenner bought were significantly undervalued as AI infrastructure plays when he started buying.

Core Scientific was trading near $1 in mid-2024. By early 2026 it is trading above $10. The CoreWeave colocation contract announced in mid-2024 was the catalyst — it proved the thesis that Bitcoin miners could convert to AI data centres profitably.

The broader narrative also shifted. Every major cloud provider announced massive data centre expansion plans in 2025. Microsoft ($80 billion), Google ($75 billion), Amazon (over $100 billion in capex, since extended to $138 billion through the OpenAI AWS deal). All of it requires power. All of it competes for grid connections. Everyone started understanding what Aschenbrenner understood a year earlier.

What This Means For AI Investors

The obvious question: is this trade still live, or did the market catch up?

The fund's continued expansion (the Q4 2025 filing shows no significant selling) suggests Aschenbrenner believes it is still live. His public thesis is that AGI arriving by 2027-2028 means demand for compute and power will increase by orders of magnitude, not percentage points. Current infrastructure is not remotely adequate. The market has started pricing the trade but has not finished pricing it.

The less obvious question: what does the Situational Awareness thesis mean for developers?

If Aschenbrenner is right about the AGI timeline, every software application built on current AI capabilities is going to need to be redesigned within 24 months to take advantage of capabilities that do not exist yet. The developers who understand that transition earliest build the applications that matter.

He was fired for taking AI safety seriously at a company that was not ready to. The market paid him $5.5 billion for taking the same view on AI capability seriously when investors were not ready to.

That pattern tends to repeat.