Sell Nvidia, Buy Power Plant: 27-Year-Old AI Investor Earns $5 Billion in One Year

Article by Sleepy.txt

In February 2026, hedge fund Situational Awareness LP submitted its quarterly holdings report, which revealed that as of the end of the fourth quarter of 2025, the fund's total market value of U.S. stock holdings was $5.517 billion.

On Wall Street, where trillions of dollars in assets are managed, $5.5 billion is just a drop in the ocean. However, this fund had a management size of less than $400 million just 12 months ago, and its founder and chief investment officer is a young man born in 1999.

His name is Leopold Aschenbrenner. He is 27 years old.

In 12 months, he grew this fund from $383 million to $5.517 billion, an increase of over 14 times. During the same period, the S&P 500 saw only single-digit growth.

What's even more surprising is his holdings. When you open the quarterly holdings report, you won't find any of the AI star companies that usually make financial news headlines. Instead, there are companies working on fuel cells, Bitcoin miners who just survived bankruptcy, and a chip giant abandoned by the entire market.

He claims his fund is focused on AI investments, but this doesn't look anything like a typical AI fund's holdings; it looks more like a madman's shopping list.

Yet, this "madman" happens to be one of the earliest and most profound minds in the world to understand how AI will change the world. Before joining Wall Street, he was a researcher at OpenAI, responsible for contemplating how to ensure that AI doesn't go rogue once it surpasses human intelligence. Later, he was ousted for speaking out of turn, wrote a 165-page manifesto, and predicted a future that most people found absurd.

Later on, he went all-in with his entire fortune.

Breaking Down $5.5 Billion: What Did He Actually Buy?

To grasp how much of a genius Leopold Aschenbrenner is when it comes to investing, the most direct way is to open his holdings report and read through it line by line.

His largest holding is Bloom Energy. With a holding value of $876 million, it represents 15.87% of the total position.

This company specializes in fuel cells. More precisely, it deals with something called "solid oxide fuel cells," which can directly convert natural gas into electricity with extremely high efficiency. Its founder, KR Sridhar, was once an engineer for NASA's Mars exploration program and was named by Fortune magazine as one of the "Top 5 futurists inventing tomorrow."

An AI fund placed its biggest bet on a power company.

According to Gartner's projections, global AI-optimized server electricity consumption will skyrocket from 93 terawatt-hours in 2025 to 432 terawatt-hours in 2030, nearly a fivefold increase over five years. The electricity demand of U.S. data centers is set to nearly triple by 2030, reaching 134.4 gigawatts. The average age of the U.S. power infrastructure is over 25 years, with many components aged between 40 and 70 years, far exceeding their design life.

In other words, AI requires more electricity than the entire grid can provide, and the grid itself is on the brink of collapse.

In the AI era, the scarcest resource is not chips, but electricity.

Bloom Energy's fuel cells happen to bypass this bottleneck. They do not need to connect to the grid, generating power directly next to data centers, running 24/7. In 2025, Bloom Energy secured a contract from CoreWeave to provide fuel cells for its AI data center in Illinois.

Speaking of CoreWeave, this is coincidentally Leopold's second-largest holding.

He holds $7.74 billion worth of CoreWeave call options, along with $4.37 billion in common stock, totaling over $12 billion, representing 22% of the total portfolio. CoreWeave is a GPU cloud provider that transitioned from cryptocurrency mining farms.

In 2017, Mike Intrator, Brian Venturo, and a few others came together to mine Bitcoin. By 2018, the crypto market crashed, and mining was no longer profitable. However, they had a bunch of GPUs on hand. In 2019, they had a revelation: GPUs could do more than mine; they could run AI.

So the company pivoted, shifting from mining to being an arms dealer of AI compute power. On March 27, 2025, CoreWeave went public on the Nasdaq, raising $1.5 billion at a price of $40 per share. A company that crawled out of a mining farm became a key supplier of AI infrastructure.

What Leopold saw in CoreWeave was its extensive GPU holdings and deep ties with NVIDIA. In an age where compute power is productivity, whoever has the GPUs reigns supreme.

However, what truly baffles people is his third largest position: Intel. With a holding value of $747 million, all in call options, it accounts for 13.54% of the total portfolio.

By 2025, Intel had become one of the most despised companies on Wall Street. Its stock price was halved from the 2024 peak, market share was eroded by AMD and NVIDIA, and CEOs came and went in succession. Almost every analyst was declaring Intel's demise.

But it was precisely at this time that Leopold doubled down on Intel with bullish options. This was an extremely aggressive move, where being right would lead to a huge payoff, and being wrong would mean a complete loss.

What was he betting on? Just two words: semiconductor manufacturing.

In November 2024, the U.S. Department of Commerce announced that Intel would receive up to $7.86 billion in direct funding through the Chip and Science Act. The sole purpose of this money was to transform Intel into a domestic semiconductor foundry to compete with TSMC.

Against the backdrop of U.S.-China tech decoupling, the U.S. needed a "local" player to produce chips. Despite Intel's lagging position, it was the only choice. Leopold was not betting on Intel's technology but on America's national resolve.

His subsequent positions are even more intriguing. Core Scientific, with a $419 million holding; IREN, $329 million; Cipher Mining, $155 million; Riot Platforms, $78 million; Hut 8, $39.5 million.

These companies share a common trait: they are all Bitcoin mining companies.

Why would an AI fund invest in a bunch of Bitcoin miners?

It's simple. These Bitcoin miners have access to the cheapest electricity and largest data center spaces across the U.S.

Core Scientific boasts over 1,300 megawatts of power capacity. IREN plans to expand to 1.6 gigawatts in Oklahoma. To survive in the intense mining competition, these miners have long secured the cheapest power resources globally through long-term power purchase agreements.

And now, what AI data centers lack the most is precisely power and space.

In 2022, Core Scientific filed for bankruptcy due to the crypto market crash. It emerged from restructuring in January 2024, shedding about $1 billion in debt and relisting on the Nasdaq. It then signed a over $10.2 billion, 12-year contract with CoreWeave to transform its mining facilities into AI data centers. To fully pivot, Core Scientific even plans to sell all its Bitcoin holdings.

IREN (formerly known as Iris Energy) then signed a $9.7 billion AI contract with Microsoft, receiving a $1.9 billion prepayment. Cipher Mining entered a 15-year lease agreement with Amazon. Riot Platforms inked a 10-year, $311 million deal with AMD.

Overnight, Bitcoin miners became the landlords of the AI era.

Now, let's piece this puzzle together.

Bloom Energy provides power, CoreWeave offers GPU computing power, Bitcoin mining firms provide space and cheap electricity, Intel contributes onshore chip manufacturing capability in the U.S. Add to that the fourth-largest position Lumentum ($479 million, optical components, a core part of interconnecting AI data centers), ninth-largest position SanDisk ($250 million, data storage), eleventh-largest position EQT Corp ($133 million, a natural gas producer supplying fuel for fuel cells).

This is a complete AI infrastructure supply chain.

From generation to transmission, chip manufacturing to GPU power, data storage to fiber interconnect. He bought every piece of it.

And he did one more thing to make this logic even clearer. By the fourth quarter of 2025, he completely divested from Nvidia, Broadcom, and Vistra. These three companies happened to be the biggest stars in the 2024 AI market rally.

He also shorted Infosys, one of India's largest IT outsourcing companies.

Selling the hottest AI chip stocks, buying unwanted power plants and mines. Shorting traditional IT outsourcing because AI programming tools are making developers more efficient, squeezing the demand for outsourcing.

Every transaction points to the same conclusion: the bottleneck of AI is not in software, but in hardware; not in algorithms, but in power; not in cloud models, but in the physical world.

So here's the question: how did a 27-year-old young man develop this set of insights?

From the Son of an East German Doctor to the Rebel of OpenAI

Leopold Aschenbrenner was born in Germany, with both parents being doctors. His mother grew up in former East Germany, his father from West Germany, and they met after the fall of the Berlin Wall. This family itself carries a mark of historical rupture—Cold War, division, reunification. His later obsession with geopolitical competition may perhaps find its initial seed here.

But Germany couldn't hold on to him. He later said in an interview, "I really wanted to leave Germany. If you are the most curious kid in the class, wanting to learn more, teachers won't encourage you, they will envy you, try to suppress you."

He referred to this phenomenon as the "Tall Poppy Syndrome," where whoever grows tall will be cut down.

At the age of 15, he persuaded his parents and flew alone to the United States to attend Columbia University.

Studying at a university at the age of 15 is unusual anywhere. But Leopold's performance at Columbia turned "unusual" into "legendary." He double majored in Economics and Mathematics-Statistics, won all the awards he could, such as the Albert Asher Green Memorial Prize, the Romine Economics Prize, and became a Junior Phi Beta Kappa Honors Society member.

At 17, he wrote a paper on economic growth and existential risk. Renowned economist Tyler Cowen said after reading it, "When I read it, I couldn't believe a 17-year-old wrote it. If this were a MIT doctoral dissertation, I would also be impressed."

At 19, he graduated from Columbia University as the Valedictorian. This is the highest honor for an undergraduate at the university. In 2021, while the world was still in the shadow of a pandemic, a 19-year-old German stood at Columbia's graduation ceremony representing all graduates.

Tyler Cowen gave him some advice: don't pursue a Ph.D. in Economics.

Cowen felt that the academic field of Economics had become somewhat "decadent" and encouraged him to do greater things. Cowen also introduced him to Silicon Valley's "Twitter eccentric" culture, a group fascinated by AI, effective altruism, and the long-term future of humanity.

After graduation, Leopold first went to the Forethought Foundation to research long-term economic growth and existential risk. He then joined the FTX Future Fund founded by SBF, working with key figures of the effective altruism movement, Nick Beckstead, and William MacAskill. His title is "Economist affiliated with the Oxford Global Priorities Institute."

This experience was crucial. It meant that before entering the AI industry, Aschenbrenner had spent several years systematically pondering a question: What kind of event could fundamentally alter the course of human civilization.

Later, he joined OpenAI.

The specific timing is unclear, but he joined a special team — the Superalignment team. This team was formed on July 5, 2023, with OpenAI co-founder Ilya Sutskever and Alignment team lead Jan Leike at the helm. The goal was to solve the alignment problem for superintelligence within four years, ensuring that an AI much smarter than humans would still do what humans say.

OpenAI had pledged to devote 20% of its compute to this team. But between promise and reality gaped a chasm.

Leopold saw things within OpenAI that troubled him. He submitted a security memo to the board, warning that the company’s security measures were "woefully insufficient" to guard against a foreign government stealing critical algorithmic secrets. The company’s response surprised him. HR called him in, saying his concerns about espionage were "racially motivated” and “not constructive.” The company’s lawyers grilled him on his views on AGI and his team loyalty.

In April 2024, OpenAI fired him for “leaking confidential information.”

The so-called “leak” was a brainstorming document on AGI safety measures shared with three external researchers. Leopold argued that the document contained no sensitive information and that sharing such documents internally for feedback was standard practice.

A month later, Ilya Sutskever left OpenAI. Three days after, Jan Leike followed suit. The Superalignment team was disbanded, and the promised 20% compute allocation by OpenAI never materialized.

A team studying “how to control superintelligence” was dismantled by the very company creating superintelligence.

The irony of the situation cannot be overstated. But for Leopold, being fired turned into a form of liberation. He was no longer employed by anyone, no longer needing to carefully phrase his words in internal memos. He could say what he truly thought, to the world.

On June 4, 2024, he published a 165-page article on a website called situational-awareness.ai. The title was “Situational Awareness: The Decade Ahead.”

The Prophecy of Page 165

To understand Leopold's investment logic, you must first decipher this tome. Because that $55 billion position is the financial translation of these 165 pages of text.

The core argument of the tome can be summarized in one sentence: AGI (Artificial General Intelligence) is very likely to be achieved in 2027.

This assessment sounded like madness in June 2024. But Leopold's method of argumentation is straightforward: think in orders of magnitude.

From GPT-2 to GPT-4, AI's capability underwent a qualitative leap, transforming from a preschooler into an intelligent high schooler. Behind this leap is roughly a 100,000-fold (5 orders of magnitude) increase in effective computation. This growth arises from the stacking of computational power, improvements in algorithmic efficiency, and the unleashing of capabilities through model "unshackling."

His prediction is that by 2027, a similar scale of growth will occur again. In terms of computational power, the computing resources used to train cutting-edge models will be 100 times greater than those for GPT-4. In algorithmic efficiency, there will be an annual improvement of approximately 0.5 orders of magnitude, totaling about 100-fold over four years. Coupled with the gain from "unshackling," allowing AI to evolve from a chatbot into an entity that can use tools and act autonomously, another order of magnitude leap will take place.

Triple 100-fold increases combined result in another 100,000-fold leap, another qualitative leap. From surpassing humans to beyond.

What truly captivates readers in this article is the series of consequences derived from this prediction.

The first consequence: a trillion-dollar-level computation cluster.

He wrote, over the past year, the Silicon Valley discourse has shifted from a $100 billion computation cluster to a $1 trillion cluster, and most recently to a trillion-dollar cluster. Every six months, there is an added zero in the boardroom plans. By the end of this decade, there will be hundreds of millions of GPUs in operation.

This prediction sounded exaggerated in June 2024. But in January 2025, the Trump administration announced the Stargate project, jointly invested in by SoftBank, OpenAI, Oracle, and MGX, planning to invest $500 billion over four years in building AI infrastructure in the United States. The first tranche of funds deployed was $100 billion. Construction has already begun in Texas.

What he wrote in the Book of Ten Thousand Words as the "Trillion-Dollar Cluster" became the White House's official plan half a year later.

The second consequence: an electricity crisis.

How much power do hundreds of millions of GPUs need? Leopold's answer is: The U.S. needs to increase its electricity production capacity by several tens of percentage points.

The data confirmed his assessment. In 2024, the capital expenditures of the four companies Amazon, Microsoft, Google, and Meta exceeded 200 billion dollars, a 62% increase from 2023. Among them, Amazon alone spent 858 billion dollars, a 78% increase. In 2025, Amazon's capital expenditure is expected to exceed 1 trillion dollars.

Most of this money was spent on data centers and power infrastructure.

Microsoft even did something unimaginable ten years ago: it signed a 20-year power purchase agreement with Constellation Energy to restart the Three Mile Island nuclear power plant.

Yes, it's the Three Mile Island that had the most severe nuclear accident in U.S. history in 1979.

This nuclear power plant will reopen in 2028, renamed the Crane Clean Energy Center, dedicated to powering Microsoft's data centers. Constellation Energy's CEO Joe Dominguez said, "Providing power to critical industries including data centers requires abundant, carbon-free, and reliable energy every hour of every day, and nuclear energy is the only energy source that can consistently deliver on this promise."

When a software company starts restarting a nuclear power plant, you know that electricity has shifted from an infrastructure problem to a strategic resource issue.

The third consequence: a geopolitical race.

The most controversial part of the Book of Ten Thousand Words is when Leopold, using language reminiscent of the Cold War, defines the AGI race as a struggle for the survival of the "free world." He harshly criticizes the security measures of the top AI labs in the U.S., calling them virtually non-existent. He urgently calls for AI algorithms and model weights to be treated as the nation's top secrets.

He even predicts that the U.S. government will eventually have to launch a national-level AGI project similar to the "Manhattan Project."

These arguments have sparked intense debates. Critics argue that he oversimplifies the complexity of geopolitics and uses a panic narrative to justify unchecked rapid development.

But some also believe that he spoke the truth. Anthropic's Dario Amodei and OpenAI's Sam Altman also believe that AGI will soon become a reality, just like him.

The true value of Leopold's Wanyanshu does not lie in whether its predictions are 100% accurate, but in providing a comprehensive, actionable mindset.

If AGI really arrives around 2027, then before that,

what does the world need? It needs massive computing power.

What does computing power need? It needs GPUs.

What do GPUs need? They need electricity.

Where does the electricity come from? From power plants, nuclear power plants, and Bitcoin mining farms with cheap electricity.

Where are the chips manufactured? At TSMC.

But what if there is a decoupling between the U.S. and China? Then you will need Intel.

How do data centers interconnect? They need optical components like Lumentum.

Where is the data stored? It needs storage like SanDisk.

See, this is the logic of that position report.

Wanyanshu is the map, and the position report is the route. Leopold translated this 165-page macro outlook into an investment portfolio that can be bet with real money. Each buy corresponds to a point in the Wanyanshu. Each sell corresponds to an assumption where he believes the market is mispriced.

But having just the map is not enough. In the real market, you need one more thing: to continue believing that you are right when everyone else says you are wrong.

This ability was severely tested on January 27, 2025.

DeepSeek Impact

On January 27, 2025, the release of DeepSeek's DeepSeek-R1 model plunged the entire Wall Street into panic. The performance of this model was close to OpenAI's o1, but the cost of use was 20 to 50 times cheaper. What was even more shocking was that the training cost of its predecessor model, DeepSeek-V3, was reportedly less than $6 million, using the sanctioned and performance-limited Nvidia H800 chip.

The market's logic collapsed in an instant.

If the Chinese can train a top model with $6 million and a stripped-down chip, what does the tens of billions of dollars spent annually by American tech giants amount to? Do those trillion-dollar computational cluster plans still make sense? Will GPU demand experience a cliff-like drop?

Panic spread like a plague. NVIDIA's stock price plummeted nearly 17%, evaporating $593 billion in market value in a single day, marking the largest single-day market value loss on Wall Street in history. The Philadelphia Semiconductor Index plunged by 9.2%, marking the largest single-day drop since the panic in March 2020. Broadcom fell by 17.4%, Marvell fell by 19.1%, Oracle fell by 13.8%.

The downturn started in Asia, spread to Europe, and finally exploded in the United States. Just the constituents of the Nasdaq 100 index saw nearly a trillion dollars in market value evaporate in a single day.

Silicon Valley venture capital godfather Marc Andreessen called DeepSeek AI's "Sputnik moment" on Twitter, stating, "This is one of the most astonishing and impressive breakthroughs I've ever seen, and as an open-source project, it is a gift to the world."

For Leopold's fund, this day should have been a disaster. His holdings are all in AI infrastructure stocks, and the market is now questioning the entire logic of AI infrastructure.

However, according to Fortune magazine, an investor from Situational Awareness LP revealed that on that day, during the market's panic selling, large tech funds called to inquire about the situation. The response they received was five words:

"Leopold says it's fine."

Why was Leopold so calm? Because in his view, the emergence of DeepSeek not only did not overturn his logic but instead affirmed it.

There is a core argument in his lengthy thesis: the progress of AI will not slow down; it will only accelerate.

The improvement in algorithm efficiency is one of the three major engines driving AI development. DeepSeek trained a stronger model with less money and weaker chips, proving that algorithm efficiency is rapidly increasing. The higher the algorithm efficiency, the more potent AI can be produced with the same computing power, stimulating more demand for computing power rather than reducing it.

Using the framework in his Magnum Opus: DeepSeek did not prove that "we don't need that many GPUs," but rather proved that "every single GPU became more valuable." When you can train better models for less money, you don't stop; you train more, bigger, and stronger models.

The panic stemmed from the fear of "demand disappearing." But those who truly understand AI know that cost reduction never eliminates demand; it only creates greater demand.

Leopold bought against the panic. The market quickly proved him right. NVIDIA and the entire AI sector rebounded rapidly in the following weeks, surpassing pre-collapse levels.

In the world of investing, belief is the scarcest asset. Not because forming belief is hard, but because persisting in belief when everyone tells you you're wrong is almost contrary to human nature.

The End of the Physical World

Leopold Aschenbrenner's story could certainly be simplified into a feel-good tale of a teenage genius striking it rich. But focusing solely on the money would waste the true value of this story.

What he truly did right was, while everyone was fixated on the code on screens and model parameters, shifting his gaze to the smokestacks of power plants, the substations of mining facilities, and the fiber-optic cables spanning continents.

In 2024, the world was abuzz with discussions about how powerful GPT-5 would be, how lifelike Sora could generate videos, and when AI would replace programmers. These discussions are certainly important. But Leopold posed a more fundamental question: How much power do these things need? Where does the power come from?

While this question may sound too basic, it is precisely this simple question that points to the greatest investment opportunity of the AI era.

AI is growing at an exponential rate, yet the physical infrastructure supporting it remains in the past century. Leopold saw this rift. Then, along this rift, all the way back to the end of the physical world. Every step started from a physical bottleneck, finding the companies to solve that bottleneck, and then betting on them.

The essence of this methodology is not actually new. During the California Gold Rush of the 19th century, the ones who made the most money were not the gold prospectors but the ones selling shovels and denim jeans. Levi Strauss made his fortune back then.

But knowing this truth is one thing; executing it in the AI era is another.

Because to execute it, you need to have both capabilities at the same time: one is a deep understanding of technological trends, knowing the development path of AI and the resource requirements; the other is a specific awareness of the physical world, knowing where electricity comes from, how data centers are built, and how fiber optics are laid.

The former requires you to have spent time in OpenAI's lab, the latter requires you to be willing to squat down and study the power contract of a bankrupt mining company.

Technologists understand AI but not the electricity market. Financial professionals understand the market but not the physical constraints of AI. Leopold happens to have both.

But more important than ability is perspective.

A line from his lengthy writing is often quoted: "You can see the future first in San Francisco." The implication of this line is: the future is not evenly distributed.

The essence of investment is to find price discrepancies in a future that has already arrived but is not yet evenly distributed.

Leopold has seen firsthand the ability curve of AI in OpenAI's lab, he knows that GPT-4 is not the end but the beginning, he knows that there will be larger models, more computing power, and crazier capital investments to come. Meanwhile, the market is still debating whether "AI is a bubble."

That's the discrepancy. What he has done is turn this discrepancy into $5.5 billion.