Leopold Fund Q1 Holdings Analysis: A Full-Scale Bet on AI Infrastructure

One of the world's most knowledgeable AI investors just updated his portfolio.

I've dug through the data for you, and the conclusion is more aggressive than I imagined.

---

First, who is this person?

Leopold Aschenbrenner, former senior researcher at OpenAI.

In 2024, he wrote a long essay titled "Situational Awareness," predicting that AI would trigger industrial-scale capital mobilization by 2026, electricity would become the biggest bottleneck, and energy and infrastructure stocks would soar.

Now it's 2026, and his predictions have almost all come true.

After leaving OpenAI, he directly founded an AI-themed hedge fund called Situational Awareness LP.

---

In the United States, institutions managing over $100 million in assets must publicly disclose their U.S. stock holdings to the SEC every quarter.

On May 18, 2026, they filed their latest Q1 2026 report (as of March 31). I'm interpreting it for you right away.

---

I. 15 Months, from $250M to $13.68B

First, look at the complete growth trajectory of this fund since its inception:

• End of 2024 (start): $250 million
• Q1 2025: $1 billion (4x in 4 months)
• Q2 2025: $2.1 billion
• Q3 2025: $4.1 billion
• Q4 2025: $5.5 billion
• Q1 2026 (latest): $13.68 billion

15 months, a 54x growth.

This isn't ordinary fund growth; this is the process of one person fully monetizing his own understanding.

And just last week, he submitted his latest Q1 2026 holdings report, which I'm interpreting for you right away.

---

II. What He's Really Betting On — AI Infrastructure

Excluding options, his core stock holdings are:

① Bloom Energy (BE) $880 million A leader in fuel cells, specifically providing power for data centers. The hotter the AI computing demand, the more voracious data center power consumption becomes, and the more Bloom Energy profits. Since his position, this stock has risen from $135 to $256, up +89%.

② SanDisk (SNDK) $720 million An independently listed storage company. Training and inference for large AI models require massive storage, making SanDisk a direct beneficiary. Since his position, it has risen from $635 to around $1,350, up +113%.

③ CoreWeave (CRWV) $560 million NVIDIA's largest cloud computing customer, specializing in GPU cloud services. Leopold believes that the "users" of computing power are a better investment than the "manufacturers." Since his position, it has risen from $77 to around $104, up +34%.

④ IREN Limited (IREN) $400 million A dual-purpose company in Bitcoin mining + AI computing. Right after his disclosure, IREN announced a 5-year computing power contract with NVIDIA, with NVIDIA investing up to $2.1 billion. Since his position, it has risen from $34 to around $52, up +52%.

⑤ Core Scientific (CORZ) $390 million Data centers + Bitcoin mining, already signed contracts with hyperscale cloud providers to transition to AI computing. Since his position, it has risen from $15 to around $24, up +63%.

⑥ Applied Digital (APLD) $320 million An infrastructure company focused on building AI data centers, selecting sites with ultra-low electricity prices. Since his position, it has risen from $24 to around $41, up +71%.

---

III. Bitcoin Mining: Across-the-Board Significant Increases

Another notable change this time — mining stocks saw almost across-the-board significant increases in holdings.

• CleanSpark (CLSK): Number of shares held increased by +648%
• Bitfarms (BITF): Increased by +188%
• Riot Platforms (RIOT): Increased by +87%
• Bitdeer (BTDR): Increased by +92%

Why?

Leopold's logic is clear: These companies are essentially large-scale consumers of low-cost electricity. They compete for the same resources as AI data centers — site selection, power contracts, construction capabilities — all core elements of AI computing infrastructure.

He's betting on Bitcoin miners not because he's bullish on Bitcoin, but because these companies control the scarcest resources in the AI era — cheap electricity and already-built data center facilities.

---

IV. How to Interpret the $8.8 Billion in Put Options?

This is the most puzzling part of the report.

While buying AI infrastructure stocks, he simultaneously used $8.8 billion worth of Put options to short almost all chip companies:

• VanEck Semiconductor ETF (SMH): Put $2.04 billion
• NVIDIA: Put $1.57 billion
• Oracle: Put $1.07 billion
• Broadcom: Put $1.01 billion
• AMD: Put $970 million
• TSMC, ASML, Micron... all included

There are two interpretations:

Interpretation ① (Long-term bearish): He believes these chip stocks are severely overvalued, and regardless of strong AI demand, their prices will correct first.

Interpretation ② (Short-term hedge): The timeline is the key clue.

On February 28, 2026, the U.S. and Israel launched a military strike against Iran, triggering a war. Throughout March, global markets plummeted. The South Korean stock market plunged 18% in four days, with Samsung and SK Hynix losing $500 billion in market value. The semiconductor supply chain was severely impacted — Strait of Hormuz blockade, helium supply disruption, soaring tungsten prices.

And his 13F snapshot date is precisely March 31 — the peak moment of war panic.

My judgment: This $8.8 billion in Puts is most likely a short-term hedge position built for war risk, not a strategic long-term bearish bet on semiconductors.

Reason: A ceasefire agreement was reached on April 8. After the ceasefire, SMH rose from $370 to $556, up nearly 50%.

A fund that is long-term bearish on semiconductors is unlikely to continue holding $8.8 billion in Puts through such a rally.

A reasonable inference: He most likely gradually closed these Put positions after the ceasefire, possibly also capturing some profits during the war period.

---

V. The Long Bets Were Right, and the Puts Were Short-Term Wisdom

Looking back at this portfolio, Leopold's core judgments are astonishingly accurate:

All his long positions, in the 7 weeks after the end of March, not a single one fell, all rose significantly.

Bloom Energy +89%, SanDisk +113%, Applied Digital +71%, Core Scientific +63%, IREN +52%...

The logic is validated: The "users and builders" of AI infrastructure have completely outperformed the "manufacturers" of chips in this round.

The Put options were likely a textbook case of geopolitical risk hedging — buying insurance when war breaks out, collecting insurance when a ceasefire is reached, and using the profits to continue betting on the long side.

---

Summary: What is he betting on?

Leopold's core framework is just one sentence:

In the AI era, what's truly scarce isn't chips; it's electricity, computing power, storage, and the infrastructure that can make these things run at scale.

The logic behind these positions:

• Power and cooling: Bloom Energy, Solaris Energy
• GPU cloud and computing power: CoreWeave, Applied Digital, IREN
• Storage: SanDisk
• Low-cost power infrastructure: The entire Bitcoin mining basket

He's not betting that "AI will succeed"; he's betting that "for AI to succeed, these infrastructure bottlenecks must be solved first, and the companies solving these problems will make money first."

So far, his bet is paying off.

The next update will be in August 2026, when the Q2 13F will tell us: whether he has closed his Put positions, and whether he continued adding to his positions after this surge.

---

References

1. Situational Awareness: The Decade Ahead (Leopold Aschenbrenner, 2024.06) — The original long essay repeatedly referenced in the text, predicting AI industrial-scale capital mobilization and the electricity bottleneck. It's the primary source for understanding this fund's investment framework.