Pickaxes and shovels of Ashenbrenner - ForkLog

# Aschenbrenner's Picks and Shovels

How a former OpenAI employee built one of the fastest-growing AI funds

In May 2026, many traders were waiting for the quarterly 13F filing from the hedge fund Situational Awareness — its founder Leopold Aschenbrenner, a former OpenAI employee, had caught Wall Street's attention with one of the most profitable AI portfolios of the last two years.

The filing deadline passed on May 15, but the document did not appear. Investors assumed the fund had requested confidentiality from the U.S. Securities and Exchange Commission (SEC), allowing it to hide positions for up to a year. More likely, managers were simply waiting until the last day — a standard practice for funds whose portfolios are watched by thousands of traders.

On May 18, the report finally came out — and revealed a radical strategy update: Aschenbrenner strengthened his positions in AI infrastructure, while simultaneously opening a large short position against semiconductor companies.

To understand why the portfolio of 24-year-old Aschenbrenner attracts so much attention, it is worth going back two years — to the firing, the manifesto, and the bet on Bitcoin miners converting facilities into data centers.

Disclaimer ForkLog is not responsible for readers' investment decisions.

Wunderkind from Berlin

Leopold Aschenbrenner was born in Germany to a family of doctors. He graduated from the John F. Kennedy School in Berlin, attended Columbia University, and in 2021 became the valedictorian — the top graduate of his class.

At age 17, he received a grant from economist Tyler Cowen through the Emergent Ventures program. Cowen called him an "economics wunderkind."

In 2022, Aschenbrenner joined the FTX Future Fund — the charitable arm of Sam Bankman-Fried's crypto exchange. He witnessed the company's collapse from the inside and left before its bankruptcy. He later recalled:

"We were a tiny team, and in one day everything collapsed and became associated with a giant fraud. It was incredibly hard."

In 2023, Aschenbrenner joined OpenAI's Superalignment team — a division led by Ilya Sutskever and Jan Leike that worked on the problem of controlling superintelligent AI. There, he co-authored the scientific paper Weak-to-Strong Generalization.

At the same time, Aschenbrenner wrote a memo to OpenAI's board of directors, warning about the risks of industrial espionage by China and calling the company's security system "glaringly insufficient." Earlier, the NYT reported that in early 2023, a hacker broke into the startup's internal messaging systems and stole information about the design of its AI technology.

In spring 2024, Aschenbrenner was fired for leaking corporate data. He called the decision politically motivated and characterized the startup's approach to security as "not robust enough to protect against theft of key secrets if foreign actors infiltrate the company." A month later, Sutskever and Leike left OpenAI, after which the company disbanded Superalignment.

In June 2024, Aschenbrenner published the essay "Situational Awareness," which caused widespread resonance in the AI industry.

Situational Awareness in a Hedge Fund

The central argument of the essay: artificial general intelligence (AGI) will arrive by 2027, the world is not ready for it, and companies building physical infrastructure for computing are the most undervalued assets on the market.

On Dwarkesh Patel's podcast, Aschenbrenner explained the investment logic through the scale of necessary infrastructure. In 2022, the GPT-4 training cluster consumed roughly 10 megawatts and cost about $500 million. AI compute scales at half an order of magnitude per year: by 2024, the largest cluster already required 100 megawatts and cost billions. Aschenbrenner described the further trajectory:

"By 2026 — a gigawatt cluster, the size of a large nuclear reactor. Tens of billions of dollars. By 2028 — 10 gigawatts, more energy than most states consume. By 2030 — a trillion-dollar cluster, 100 GW, more than 20% of all electricity production in the US. And that's just the training cluster."

That's only model training — inference will require several times more power. Aschenbrenner compared the advantage in the AGI race to the technological gap of the coalition in the First Gulf War:

"Western forces had a 100:1 casualty ratio. Better sensors on tanks, more accurate missiles, GPS, stealth. A 20-30 year gap — and they simply crushed the opponent. Superintelligence applied to a broad range of R&D could compress a century of technological progress into less than ten years."

In September 2024, Aschenbrenner founded the hedge fund Situational Awareness LP. Anchor investors include Stripe co-founders Patrick and John Collison, former GitHub CEO Nat Friedman, and Safe Superintelligence co-founder Daniel Gross. The minimum entry threshold is $25 million, with a two-year lock-up period.

According to The Wall Street Journal, as early as the first half of 2025, the fund showed a 47% return after fees — versus roughly 6% for the S&P 500 over the same period. The main contribution came from bets on energy and data centers.

Fortune wrote that by early 2026, the fund's publicly disclosed exposure in U.S. assets had grown to roughly $5.5 billion, compared to about $225 million at the end of 2024. It should be noted: this is not about confirmed net profit, but about the size of disclosed positions in 13F filings — the figure includes asset value growth, new capital inflows, and possible use of leverage.

What the Fresh 13F Showed

Form 13F is a quarterly report that funds with more than $100 million in assets are required to file with the SEC.

Situational Awareness's previous filing for Q4 2025 recorded 29 positions totaling $5.5 billion with a minimal number of option trades. The new report looks fundamentally different: 42 positions with disclosed exposure of $13.67 billion — nearly triple in one quarter.

Portfolio structure by instrument type: 66% comes from put options, 10% from calls, and 24% from direct stock ownership.

The fund opened put positions on the largest chip manufacturers — all appearing in the filing for the first time:

• VanEck Semiconductor ETF (SMH) — $2.04 billion;
• Nvidia — $1.57 billion;
• Oracle — $1.07 billion;
• Broadcom — $1.01 billion;
• AMD — $969 million;
• Micron — $584 million;
• TSMC — $535 million;
• ASML — $494 million;
• Intel — $159 million.

Total — $7.46 billion in puts against the semiconductor sector. None of these positions were in the previous filing.

At the same time, Aschenbrenner increased investments in stocks of companies in the energy, data center, and data storage sectors:

• Bloom Energy — $879 million (largest long in the portfolio);
• SanDisk — $724 million;
• CoreWeave — $556 million;
• IREN — $401 million;
• Core Scientific — $389 million;
• Applied Digital — $320 million.

The fund also increased positions in Bitcoin miners: Riot Platforms (+87% shares), CleanSpark (+648%), Bitdeer (+92%), Bitfarms (+188%). New names include Hive Digital and T1 Energy.

Fully exited from the portfolio: Lumentum, Hut 8, Cipher Mining, Coherent, EQT, and Tower Semiconductor. One of the most notable rotations: a call option on Intel worth $747 million was replaced with a put option worth $159 million.

A separate detail — call options on individual names that the fund holds simultaneously with puts on the sector as a whole:

• Micron — $422 million;
• SanDisk — $389 million;
• TSMC — $355 million;
• CoreWeave — $141 million;
• Bloom Energy — $55 million.

The combination of puts on the index and calls on selected stocks means that Aschenbrenner is not simply betting against chip makers. He is building a position that benefits if the sector as a whole pulls back, while individual companies — those closer to physical infrastructure and data storage — continue to grow.

What the Puts Hide

Form 13F does not disclose strikes and expiration dates of options. The amounts stated in the filing are the market value of the shares controlled by the puts (notional value), not the cost of the contracts themselves. For "far" or deep out-of-the-money options, the difference can be enormous: the fund could have controlled $7.5 billion in stocks while paying far less for the contracts.

Blogger Jason's Chips highlighted this nuance. In his view, the puts could be a short-term hedge against geopolitical risks — for example, escalation of conflict with Iran — rather than a full-fledged bearish bet on semiconductors. In that case, the real money invested in puts is significantly less than the stated $7.5 billion.

Another caveat matters. The 13F reflects the portfolio state on the last day of the quarter — March 31, 2026. Nearly seven weeks passed between the reporting date and publication. During that time, the semiconductor sector rose noticeably, meaning the put positions likely partially depreciated. However, Aschenbrenner could have closed them or increased them — the current portfolio is by definition unknown.

Nevertheless, the structure is telling in itself: the largest longs — Bloom Energy, SanDisk, and CoreWeave — indicate where exactly Aschenbrenner sees the next bottleneck in the AI industry.

Long Shovels

The name of the classic "picks and shovels" strategy dates back to the California Gold Rush: while prospectors went broke chasing gold, sellers of tools grew rich. Aschenbrenner applies the same logic to the AI boom, but goes a level deeper.

He does not buy model creators like OpenAI or Anthropic. He does not make an outright bullish bet on chip developers — Nvidia or AMD. Instead, the fund invests in what models and chips physically cannot function without: electricity, cooling, fiber optics, land plots with grid connections, and power contracts.

Bloom Energy, the fund's largest position, manufactures solid oxide fuel cells that can power a data center independently, without connecting to the strained power grid. Standard connection to utility grids takes more than five years, while Bloom Energy installation begins generating power within 90 days of setup.

SanDisk represents a bet on NAND memory. AI inference generates streams of data that need to be stored somewhere: each processed token relies on a stack of HBM and enterprise SSDs. Demand for storage grows in parallel with computation, but attracts far less investor attention.

CoreWeave serves as the "neocloud" for AI workloads. Contracts with Anthropic and OpenAI have turned the company into a de facto third hyperscaler for GPU computing after AWS and Azure.

Analyst Jim Lue adds insider context: Aschenbrenner worked at OpenAI in 2023–2024, when the startup used CoreWeave's infrastructure, and saw its software stack from the inside. Since then, OpenAI moved its AI infrastructure in-house and now only takes "bare metal" from CoreWeave. But for the long tail of AI startups without the resources for their own stack, CoreWeave's cloud remains one of the few accessible options.

Miners as a Bet on AI

A separate cluster in the portfolio is Bitcoin miners. These companies already own what is hardest to build from scratch: land plots near power lines, active electricity contracts, and powerful cooling systems. Converting such a facility for AI hosting is cheaper and faster than building a new data center.

Computations for neural networks generate 2–5 times more revenue per kWh consumed than maintaining the Bitcoin network. According to CoinShares, the total volume of AI hosting contracts in the industry has already exceeded $70 billion.

IREN is a vertically integrated miner combining its own capacity, energy assets, and GPU cloud. Riot Platforms owns a site in Corsicana, Texas, with over 1 GW of capacity — the type of infrastructure that today is worth more as AI compute power than as hashrate. CleanSpark, whose position grew by 648%, remains a pure Bitcoin miner, but the ownership logic is the same: cheap energy plus land and the ability to switch to HPC hosting at any moment.

Aschenbrenner is consistently increasing this part of the portfolio: Bitdeer (+92%), Bitfarms (+188%), a new position in Hive Digital. All these companies share one thing: they control energy infrastructure, the scarcity of which determines the growth rate of the AI industry.

Physics vs. Valuations

The portfolio formula looks like this: longs in electricity, fiber optics, campuses, and long-term power contracts; shorts in chip manufacturers, whose valuations, according to Aschenbrenner, already price in the best-case scenario.

Over the past 18 months, the fund has bet exclusively on electricity, memory, compute power, and physical sites. This strategy has made Situational Awareness one of the most profitable funds in the world, and Aschenbrenner continues to adhere to it.

But the appearance of $7.5 billion in puts signals: Aschenbrenner believes that semiconductor companies, on which Wall Street has been betting for the last two years, have already priced in all the good that can happen to them. Chip makers' margins risk shrinking amid competition and overproduction, while physical constraints — energy, cooling, grid connections — remain unresolved.

Text: Sasha Kosovan