Sold Nvidia, bought a power plant, 24-year-old alternative AI investor made $5 billion in one year

Title: Sold Nvidia, Bought Power Plants — 24-Year-Old Alternative AI Investor Made $5 Billion in One Year

Author: Dongcha Beating

Source:

Reprint: Mars Finance

Text｜Sleepy.txt

In February 2026, the hedge fund Situational Awareness LP filed its quarterly holdings report, showing that as of the end of Q4 2025, the fund’s U.S. stock holdings had a total market value of $5.517 billion.

Wall Street manages trillions of dollars in assets, so $55 billion is just a drop in the ocean. But this fund’s assets under management were less than $400 million just 12 months earlier, and its founder and chief investment officer is a young man born in 1999.

His name is Leopold Aschenbrenner. 24 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 rose by single digits.

What’s even more astonishing is his holdings. Open the quarterly report, and you won’t find any AI superstar companies that you often see headlines about. Instead, there are companies making fuel cells, Bitcoin miners just clawing their way back from bankruptcy, and chip giants being abandoned by the market.

He claims his fund invests in AI, but what’s in his portfolio looks more like a crazy person’s shopping list than an AI fund.

But this crazy person is actually one of the earliest and most profound people in the world to understand how AI will change the world. Before Wall Street, he was a researcher at OpenAI, working on how to ensure AI doesn’t go out of control when it becomes smarter than humans; later, he was expelled for saying things he shouldn’t have, writing a 165-page manifesto predicting a future that most would find absurd.

And then, he bet his entire net worth on it.

Breaking down $5.5 billion: What exactly did he buy?

To understand how talented Leopold Aschenbrenner is in investing, the most direct way is to open his holdings report and read it line by line.

His largest holding is Bloom Energy, with a market value of $876 million, accounting for 15.87% of the total portfolio.

This company makes fuel cells. More precisely, it produces “solid oxide fuel cells,” which can directly convert natural gas into electricity with high efficiency. Founder KR Sridhar was once an engineer on NASA’s Mars exploration program, named one of Fortune’s “Top Five Futurists Creating the Future.”

An AI fund has placed its biggest bet on an energy company.

According to Gartner’s forecast, the global power consumption of AI-optimized servers will soar from 93 TWh in 2025 to 432 TWh in 2030, nearly five times in five years. The U.S. data center power grid demand will nearly triple by 2030, reaching 134.4 GW. Meanwhile, the average age of U.S. power infrastructure exceeds 25 years, with many components between 40 and 70 years old, far beyond their designed lifespan.

In other words, the electricity AI needs exceeds what the entire grid can supply. And the grid itself is aging rapidly.

The most scarce resource in the AI era isn’t chips — it’s electricity.

Bloom Energy’s fuel cells can bypass this bottleneck. They don’t need to connect to the grid, generating power directly next to data centers, 24/7. In 2025, Bloom Energy secured a contract from CoreWeave to supply fuel cells for its AI data center in Illinois.

Speaking of CoreWeave, that’s Leopold’s second-largest holding.

He holds $774 million worth of call options on CoreWeave, plus $437 million in common stock, totaling over $1.2 billion, or 22% of the portfolio. CoreWeave is a GPU cloud service provider, transformed from a crypto mining farm.

In 2017, Mike Intrator and Brian Venturo and a few others pooled resources to mine Bitcoin. When the crypto crash hit in 2018, mining became unprofitable. But they had a bunch of GPUs. In 2019, they had an epiphany: GPUs can do more than mine — they can run AI.

So the company pivoted, transforming from a mining farm into an AI computing powerhouse. On March 27, 2025, CoreWeave IPO’d on NASDAQ, raising $1.5 billion at $40 per share. A company that emerged from mining, now a core supplier of AI infrastructure.

Leopold’s bet is on CoreWeave’s massive GPU inventory and its deep ties with Nvidia. In an era where computing power equals productivity, whoever controls GPUs is king.

But what’s truly baffling is his third-largest holding: Intel. With a market value of $747 million, all in call options, accounting for 13.54% of the portfolio.

In 2025, Intel was one of Wall Street’s most despised companies. Its stock was halved from its 2024 high, market share eroded by AMD and Nvidia, and CEOs changed repeatedly. Almost all analysts declared Intel finished.

Yet Leopold aggressively bought call options at this moment. A highly risky move — if right, it soars; if wrong, it’s zero.

What is he betting on? Just two words: foundry.

In November 2024, the U.S. Department of Commerce announced that Intel would receive up to $7.86 billion in direct funding through the Chips and Science Act. The goal? Make Intel a domestic chip foundry to compete with TSMC.

In the context of U.S.-China tech decoupling, America needs a “homegrown” chip maker. Intel may be behind, but it’s the only choice. Leopold isn’t betting on Intel’s technology — he’s betting on America’s national will.

His other holdings are even more interesting. Core Scientific, with $419 million; IREN, $329 million; Cipher Mining, $155 million; Riot Platforms, $78 million; Hut 8, $39.5 million.

All these companies share a common trait: they are Bitcoin miners.

Why would an AI fund invest in Bitcoin miners?

Simple — because Bitcoin miners have access to the cheapest electricity and largest data center sites in the U.S.

Core Scientific has over 1,300 MW capacity. IREN plans to expand by 1.6 GW in Oklahoma. These miners have long locked in the cheapest power sources worldwide through long-term power purchase agreements to survive fierce computing competition.

And now, what AI data centers need most is electricity and space.

In 2022, Core Scientific filed for bankruptcy due to the crypto crash. It restructured in January 2024, cutting about $1 billion in debt, and relisted on Nasdaq. Then, it signed a 12-year, over $10.2 billion contract with CoreWeave, transforming its mining farms into AI data centers. To fully pivot, Core Scientific even plans to sell all its Bitcoin holdings.

IREN (formerly Iris Energy) signed a $9.7 billion AI contract with Microsoft, receiving $1.9 billion in advance payments. Cipher Mining signed a 15-year lease with Amazon. Riot Platforms signed a 10-year, $311 million contract with AMD.

Overnight, Bitcoin miners became landlords of the AI era.

Now, let’s complete this puzzle.

Bloom Energy supplies power, CoreWeave provides GPU compute, Bitcoin miners offer space and cheap electricity, Intel offers domestic chip manufacturing. Plus, the fourth-largest holding, Lumentum ($479 million, optical components, key for interconnecting AI data centers), the ninth-largest, SanDisk ($250 million, data storage), and the eleventh, EQT Corp ($133 million, natural gas producer fueling fuel cells).

This is a complete AI infrastructure supply chain.

From power generation, transmission, chip manufacturing, GPU compute, data storage, to fiber optic interconnects — Leopold has bought into every link.

And he’s doing another thing that makes this logic even clearer. In Q4 2025, he completely liquidated Nvidia, Broadcom, and Vistra. These three were among the biggest winners of the 2024 AI rally.

He also shorted Infosys, one of India’s largest IT outsourcing firms.

Selling the hottest AI chip stocks, buying unloved power plants and mining farms. Shorting traditional IT outsourcing because AI programming tools are making programmers more efficient, reducing outsourcing demand.

Every move points to one conclusion: the bottleneck of AI isn’t software, it’s hardware; not algorithms, but electricity; not cloud models, but the physical world.

So, how did a 24-year-old form this understanding?

From East German doctor’s son to OpenAI rebel

Leopold Aschenbrenner was born in Germany, to parents who were both doctors. His mother grew up in East Germany, his father in West Germany; they met after the Berlin Wall fell. His family bears the mark of a historical rupture — Cold War, division, reunion. His obsession with geopolitical competition may have its roots here.

But Germany couldn’t keep him. He later said in an interview: “I really wanted to leave Germany. If you’re the most curious kid in class, wanting to learn more, teachers won’t encourage you — they’ll be jealous and try to suppress you.”

He called this phenomenon “High Poppy Syndrome”: the taller you grow, the more likely you are to be cut down.

At 15, he convinced his parents to let him fly alone to the U.S., to Columbia University.

15 and attending college — that’s unusual anywhere. But Leopold’s performance at Columbia turned “outsider” into “legend.” He majored in economics and math-statistics, winning awards like the Albert Asher Green Memorial Prize, Romine Economics Award, and becoming a Junior Phi Beta Kappa.

At 17, he wrote a paper on economic growth and existential risks. Renowned economist Tyler Cowen read it and said: “When I read it, I couldn’t believe it was written by a 17-year-old. If it were a PhD thesis at MIT, I’d be equally impressed.”

At 19, he graduated as valedictorian from Columbia — the highest honor for undergraduates. In 2021, still under the shadow of the pandemic, this 19-year-old German boy delivered the commencement speech at Columbia.

Cowen advised him: don’t pursue a PhD in economics.

He thought academia had become somewhat “decadent,” and encouraged him to do bigger things. Cowen also introduced him to Silicon Valley’s “Twitter oddball” culture — a group obsessed with AI, effective altruism, and humanity’s long-term fate.

After graduation, Leopold first joined the Forethought Foundation, researching long-term economic growth and existential risks. Then he joined the FTX Future Fund, founded by SBF, working alongside key figures in effective altruism like Nick Beckstead and William MacAskill. His title was “Economist at the Oxford University Global Priorities Research Institute.”

This experience was crucial. It meant that before entering AI, Aschenbrenner had spent years systematically pondering a fundamental question: what kind of event could truly change the course of human civilization?

Then he joined OpenAI.

The exact timing is unclear, but he was part of a special team — “Superalignment.” Founded on July 5, 2023, led by OpenAI co-founder Ilya Sutskever and team lead Jan Leike, its goal was to solve the alignment problem for superintelligence within four years — ensuring that an AI far smarter than humans would still follow human instructions.

OpenAI had promised to allocate 20% of its compute power to this team. But there was a gap between promise and reality.

Leopold saw troubling signs inside OpenAI. He submitted a security memo warning that the company’s safeguards were “severely inadequate” to prevent foreign governments from stealing key algorithm secrets. The response surprised him: HR called him in, saying his concerns about espionage were “racist” and “unconstructive.” Company lawyers questioned his views on AGI and his team’s loyalty.

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

The so-called “leak” was a brainstorming document about AGI safety he shared with three external researchers. Leopold said the document contained no sensitive info; sharing such drafts internally for feedback was normal.

A month later, Ilya Sutskever left OpenAI. Three days after that, Jan Leike also departed. The Superalignment team was disbanded, and OpenAI never delivered on its promise of 20% compute.

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

The irony is undeniable. But for Leopold, being fired was a kind of liberation. He was no longer employed, no longer had to be cautious in internal memos. He could speak openly to the world.

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

165 pages of prophecy

To understand Leopold’s investment logic, you must read this manifesto. Because his $5.5 billion holdings are the financial translation of these 165 pages.

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

This prediction sounded crazy in June 2024. But Leopold’s reasoning was straightforward: order of magnitude.

From GPT-2 to GPT-4, AI capabilities have undergone a qualitative leap, transforming from preschoolers into high schoolers. Behind this leap is roughly a 100,000-fold (five orders of magnitude) increase in effective compute. This growth comes from stacking physical compute, improving algorithms, and “unbounding” models to unleash capabilities.

He predicts that by 2027, a similar scale of growth will happen again. In terms of physical compute, training resources for cutting-edge models will be 100 times more than GPT-4. Algorithm efficiency will improve about 0.5 orders of magnitude annually, totaling roughly 100 times over four years. Plus, the “unbounding” effect will turn AI from chatbots into tool-using, autonomous agents — another order of magnitude jump.

Stacking three 100-fold increases results in another 100,000 times — a qualitative leap from high schooler to surpassing human intelligence.

What makes this article truly unsettling is the chain of consequences he derives from this prediction.

First: Trillion-dollar-scale compute clusters.

He writes that in the past year, Silicon Valley’s focus has shifted from $10 billion clusters to $100 billion, and now to trillion-dollar clusters. Every six months, the board’s plans add another zero. By the end of this decade, hundreds of millions of GPUs will be in operation.

This forecast sounded exaggerated in June 2024. But in January 2025, the Trump administration announced the Stargate project, a joint investment by SoftBank, OpenAI, Oracle, and MGX, planning to invest $500 billion over four years to build AI infrastructure in the U.S. The first immediate funding was $100 billion. Construction has already begun in Texas.

What Leopold wrote as “trillion-dollar clusters” became an official White House plan within half a year.

Second: Power crisis.

How much electricity do hundreds of millions of GPUs need? Leopold’s answer: increase U.S. power generation capacity by dozens of percentage points.

Data confirms his judgment. In 2024, Amazon, Microsoft, Google, and Meta’s capital expenditures exceeded $200 billion, up 62% from 2023. Amazon alone spent $85.8 billion, up 78%. In 2025, Amazon’s capex is expected to surpass $100 billion.

Most of this money is 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 plant.

Yes, the same plant that experienced the worst nuclear accident in U.S. history in 1979.

This plant will reopen in 2028, renamed the “Crane Clean Energy Center,” powering Microsoft’s data centers. Constellation CEO Joe Dominguez said: “Providing reliable, carbon-free power for critical industries like data centers requires sufficient, continuous energy — nuclear is the only energy source that can deliver that promise.”

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

Third: Geopolitical competition.

The most controversial part of the manifesto is Leopold’s near-Cold War language, framing the AGI race as a struggle for the “survival of the free world.” He criticizes top U.S. AI labs’ safety measures as “virtually useless,” calling for AI algorithms and model weights to be treated as national secrets.

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

These claims sparked fierce debate. Critics say he oversimplifies geopolitical complexity and uses panic narratives to justify unrestrained acceleration.

But some believe he’s telling the truth. Dario Amodei of Anthropic and Sam Altman of OpenAI share his view that AGI will arrive soon.

The true value of the manifesto isn’t whether its predictions are 100% accurate, but that it provides a comprehensive, actionable mental framework.

If AGI really arrives around 2027, what does the world need before then?

Massive compute power.

What does compute power need? GPUs.

What do GPUs need? Electricity.

Where does that electricity come from? Power plants, nuclear stations, Bitcoin mines with cheap power.

Where are chips made? TSMC.

But what if U.S.-China decoupling happens? Then it needs Intel.

How are data centers interconnected? Optical components — Lumentum.

Where is data stored? Storage — SanDisk.

See? That’s the logic behind his holdings report.

The manifesto is a map; the holdings are the route. Leopold translated this 165-page macro forecast into an investable portfolio. Every buy corresponds to a point in the manifesto; every sell reflects a market mispricing he perceives.

But a map alone isn’t enough. In real markets, you also need one thing: the ability to keep believing you’re right when everyone else says you’re wrong.

This skill was put to the ultimate test on January 27, 2025.

DeepSeek Shock

On January 27, 2025, the release of DeepSeek’s DeepSeek-R1 model sent shockwaves through Wall Street. Its performance approached OpenAI’s GPT-1, but at 20 to 50 times lower cost. Even more astonishing, its predecessor, DeepSeek-V3, was reportedly trained for less than $6 million using Nvidia H800 chips sanctioned and performance-limited by the U.S.

The market’s logic instantly collapsed.

If Chinese researchers can train top-tier models for just $6 million with limited chips, what’s the point of the billions spent annually by U.S. tech giants? Do the trillions of dollars in planned compute clusters still matter? Will GPU demand plummet?

Panic spread like wildfire. Nvidia’s stock plunged nearly 17%, losing $593 billion in market cap in a single day — the largest single-day loss in Wall Street history. The Philadelphia Semiconductor Index fell 9.2%, its worst since March 2020. Broadcom dropped 17.4%, Marvell 19.1%, Oracle 13.8%.

The decline started in Asia, spread to Europe, and finally exploded in the U.S. Nasdaq 100 components lost nearly $1 trillion in market value in one day.

Silicon Valley venture legend Marc Andreessen called DeepSeek a “Sputnik moment” for AI on Twitter, saying: “This is one of the most astonishing and impressive breakthroughs I’ve seen, and as an open-source project, it’s 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 was questioning the entire logic of AI infrastructure.

But according to Fortune, an investor from Situational Awareness LP revealed that during the panic sell-off, some large tech funds called to check in. The response was five words:

“Leopold says it’s fine.”

Why was Leopold so calm? Because, in his view, DeepSeek’s emergence not only didn’t overturn his logic but confirmed it.

His core argument in the manifesto: AI progress won’t slow down — it will accelerate.

Algorithm efficiency improvements are one of the three main engines driving AI. DeepSeek training models with less money and weaker chips proved that algorithm efficiency is skyrocketing. The higher the efficiency, the more valuable each GPU becomes, fueling greater demand rather than reducing it.

Using his framework: DeepSeek doesn’t prove “we don’t need so many GPUs,” but “each GPU becomes more valuable.” When you can train better models for less money, you don’t stop — you train more, bigger, stronger models.

Panic stems from fear of “demand disappearing.” But those who truly understand AI know that cost reductions never eliminate demand — they create bigger demand.

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

In investing, belief is the most scarce asset. Not because forming beliefs is hard, but because sticking to them when everyone else says you’re wrong is almost against human nature.

The End of the Physical World

Leopold Aschenbrenner’s story can of course be simplified as a genius teen’s rapid wealth. But if you only focus on the money, you miss the real value of this story.

What he did right was, while everyone was staring at code and model parameters, he looked at smokestacks at power plants, substation stations at mines, and fiber optic cables spanning continents.

In 2024, the world debated how powerful GPT-5 would be, how realistic Sora’s videos could get, and when AI would replace programmers. These discussions are important. But Leopold asked a deeper question: how much electricity do all these things need? Where does that electricity come from?

This seemingly simple question points directly to the biggest investment opportunity of the AI era.

AI is growing exponentially, but its physical infrastructure still lags behind the last century. Leopold saw this gap. Then he traced it back to the physical world’s end. Every step starts from a physical bottleneck, finds the company solving it, and bets.

This methodology isn’t new. During the California Gold Rush in the 19th century, the biggest profits weren’t made by gold miners but by sellers of shovels and jeans. Levi Strauss made his fortune then.

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

Because to do so, you need two skills: a deep understanding of technological trends — knowing AI’s development path and resource needs; and a concrete understanding of the physical world — knowing where electricity comes from, how to build data centers, and how to lay fiber.

The former requires experience in labs like OpenAI; the latter, willingness to study a bankrupt miner’s power contract.

Tech experts understand AI but not power markets. financiers understand markets but not the physical constraints of AI. Leopold has both.

But more important than ability is perspective.

His manifesto often quotes: “You can see the future first in San Francisco.” The subtext: the future isn’t evenly distributed.

The essence of investing is finding mispricings in a future that has already arrived but isn’t yet evenly distributed.

Leopold saw the AI capability curve firsthand at OpenAI. He knows GPT-4 isn’t the end but the beginning. He expects bigger models, more compute, and crazier capital inflows. Yet the market still debates whether “AI is a bubble.”

That’s the mispricing. His job is to turn this mispricing into $5.5 billion.