Leopold Aschenbrenner shorts $90 billion in NVIDIA, shifts focus to energy and infrastructure

Key Points Summary

Leopold Aschenbrenner, regarded as one of the world’s most aggressive AI investors, is shorting NVIDIA, ASML, and Oracle with approximately $9 billion in notional exposure on public markets, while redirecting capital toward deeper AI infrastructure and model assets such as power, memory, data center networks, and Anthropic.

The two hosts believe this does not mean the AI bubble has burst, but rather signals a rotation in infrastructure spending from “chip-first” to “energy, network, and data center construction-first,” especially after NVIDIA recently completed a $25 billion bond offering and Anthropic’s valuation was pushed higher—making the market implications of this assessment rapidly intensify.

Summary of insightful perspectives

The classic "selling shovels" trade in AI has become too crowded, and Leopold’s recent position changes are signaling exactly that.

His assessment was not that AI infrastructure has reached its peak, but that certain layers within the infrastructure stack—particularly semiconductors and traditional popular assets—have become overly crowded.

· If the question becomes where the funds will go next, there are two answers. The first and most direct is that they will flow into the next true infrastructure bottleneck—such as electricity, memory, and data center networks. The second answer is the mysterious investment that was only revealed a few weeks ago.

He has consistently bet on highly infrastructure-oriented areas, investing in both optical companies and power-related companies.

· If he is cautious about NVIDIA, the capital will flow into areas like power and memory; at the same time, he also wants to invest directly in the "mine" itself, rather than just continuing to buy the "shovels"—Anthropic is his favorite mine.

The issue isn't whether NVIDIA will continue to make profits, but why a company with extremely high profit margins and substantial cash on hand would still need to borrow an additional $25 billion externally.

If a company simultaneously engages in large-scale stock buybacks and significantly increases dividends while also taking on debt in the same month, it clearly isn’t borrowing due to a lack of cash. A more plausible explanation is that this represents access to cheap capital, and the financing approach during this AI-driven market rally is undergoing a subtle shift.

· "The real bottleneck is no longer just GPUs, but electricity, memory, data center networks, and the ability to actually build these things."

Even if you raise more money, you can't build data centers fast enough, ramp up memory chip production quickly enough, or immediately expand the power grid, transmission lines, and related infrastructure. There aren't enough people on the ground, and approvals, regulations, and various procedures are holding you back.

Whoever can build the data center will make the money.

As GPU scales grow larger, copper wires will become hotter and energy losses will increase, leading to poor efficiency; in such cases, fiber optics become the next logical upgrade direction.

In high-bandwidth, short-distance transmission scenarios, copper is almost the only material people truly want to use. Only when it becomes unsuitable—such as over long distances or due to excessive heat—do we switch to fiber optics. As a result, there is currently very strong market demand for a combination of copper and fiber optics.

Copper futures have recently been very strong, essentially because everyone needs it—it's the most critical foundational material for short-distance, high-bandwidth transmission, with fiber optics being the next step.

Copper remains the most critical material for short-distance, high-bandwidth transmission, but once the distance increases or the heat becomes too high, you must switch to fiber optics.

· The next wave of capital will flow into infrastructure companies that may not sound glamorous.

I’ve always been bullish on energy, because even if AI demand slows down, energy itself remains a global necessity, and this demand will only continue to grow.

The single trend that will continue to rise in any scenario is our demand for energy, electricity, and power—these companies are the ones I’m most willing to hold long-term.

The companies I most want to follow are those that Jensen is investing in and that also align with Leopold’s logic. Right now, the closest investment I’m tracking is Marvell.

· The best long-term positions aren't necessarily the most popular chip companies, but rather the power infrastructure that cannot be avoided under any macroeconomic scenario.

Leopold's AI Investment Portfolio

Josh Kale: Leopold Aschenbrenner, this 24-year-old specialist in AI investing, is now almost universally regarded by the market as the world’s top AI investor. Rumors suggest his fund’s nominal assets under management have surpassed $20 billion. When we reviewed Ejaaz’s post a month ago, the fund’s size was only $13.7 billion—meaning it has nearly doubled every quarter.

We’ve identified several significant new developments from his most recent investment moves. In our last episode, we discussed his portfolio, and the most surprising aspect was that he had taken a short position in a company nearly everyone knows—NVIDIA, the world’s most valuable company and the hottest name in AI. Many people are puzzled: why would he establish a short position exceeding $9 billion against such a company?

We’ve now obtained a new clue that might explain this situation: NVIDIA is raising funds—through bond issuance. On the surface, this seems illogical; why would a company of NVIDIA’s massive scale and extremely high profit margins need additional capital after just securing $25 billion in cash? Today, we’ll explore this alongside Leopold’s portfolio, discussing why he’s been so successful, what he’s focusing on next, and what NVIDIA’s fundraising truly signifies.

Ejaaz Ahamadeen: Let me provide some background first. Leopold Aschenbrenner was previously a researcher at OpenAI and raised a fund about one and a half to two years ago; the initial size was relatively small—I recall it was around $200 million—but according to his most recent 13F filing, the publicly disclosed holdings of this fund are now valued at $13.7 billion.

So the market naturally wonders which positions he has taken, what his core investment thesis is, and where his next major trade will land.

To understand this, first know that a month ago, Leopold was very optimistic about the entire AI sector, particularly favoring the "selling shovels" logic—such as NVIDIA and other upstream hardware suppliers.

But about a month ago, the market realized he wasn’t as bullish on the semiconductor sector as previously thought. He still remains optimistic about memory, power, and other true bottlenecks, and possibly also about new cloud providers—but he specifically lacks confidence in NVIDIA, the world’s most valuable company. More precisely, he has taken approximately $9 billion in short positions across NVIDIA, ASML, Oracle, and other companies widely seen as primary beneficiaries of AI infrastructure.

The logic behind shorting NVIDIA

Ejaaz Ahamadeen: As soon as this news broke, many people began to worry, wondering if the AI bubble was about to burst. After all, on the surface, NVIDIA’s GPUs are still selling strongly, and demand hasn’t shown any clear signs of weakening—so what’s really going on?

Later, we uncovered several new leads, the most important of which is that NVIDIA has just raised $25 billion externally through bond financing. This means it is not solely using its own cash reserves but is instead taking on additional leverage. The question then arises: Why would the world’s most profitable, highest-margin, and strongest-cash-flow company need to borrow $25 billion externally?

Josh Kale; and initially, they only planned to raise $20 billion, but ended up increasing it to $25 billion, with subscriptions exceeding three times the target. Last time, when we discussed this portfolio, we mentioned not to worry about a bubble, as these companies, despite their massive capital expenditures, generate sufficient revenue to theoretically support their expansion through their own balance sheets.

But this is the first time since 2021 that NVIDIA has clearly turned to off-balance-sheet financing rather than directly using its own cash reserves. I recall it currently has over $12 billion in cash on hand. Putting all of this together creates a strange tension: on one side, Leopold is shorting the stock, while on the other, NVIDIA—seemingly flush with unlimited cash and profits—is still issuing debt. So what’s really going on?

NVIDIA Bond Financing Breakdown

Josh Kale: Ejaaz, could you help us break down this transaction itself? Because this isn’t a typical financing—it’s a bond issuance. Ultimately, NVIDIA’s balance sheet now has an additional $25 billion, and the interest rate seems to be quite low.

Ejaaz Ahamadeen: I’ve presented both explanations. NVIDIA already had approximately $13.7 billion in cash on its balance sheet, meaning it could have simply used its own funds. So why raise external financing? The simplest analogy is buying a house. Many people, even if they have the full amount available, still choose to take out a loan because they can use their own capital for other purposes—and if borrowing costs are low enough, it’s actually more advantageous.

Over the past few years, the interest rate environment has not been favorable, but if you’re NVIDIA—one of the world’s most valuable and sought-after companies—you can borrow at very favorable terms. This $25 billion bond offering has maturities ranging from 2 to 30 years, effectively representing extremely low-cost capital, with interest rates nearing those of U.S. Treasury yields.

Moreover, this financing was reportedly oversubscribed by about four times, meaning $85 billion in capital sought to enter the $25 billion offering, allowing NVIDIA to essentially choose its investors at will. If we take the official statement at face value, NVIDIA’s explanation is that this is primarily a financial arrangement to repay and refinance a portion of its existing debt. Google performed a very similar transaction a few weeks ago and also did so in February this year. So, you can certainly accept this explanation as a form of financial optimization.

But on the other hand, it’s hard to ignore: over the past six weeks, NVIDIA, Amazon, Google, and several other hyperscale cloud providers have nearly all been raising external capital—some through bond issuances, others through stock sales. Leopold’s perspective may not be entirely unfounded—could this be a sign that the bubble is beginning to loosen, that the house of cards is starting to wobble? However, if you look solely at financial structures, this does not yet clearly point to danger.

Josh Kale: I agree. Shorting NVIDIA with a $9 billion position is an extremely large bet. But during our research, we also noticed something else: On May 18, NVIDIA’s board authorized an additional $80 billion in share repurchases and increased the dividend from $0.01 to $0.25 per share—a 25-fold increase.

If a company simultaneously engages in large-scale stock buybacks and significantly increases dividends while also taking on debt in the same month, it clearly isn’t borrowing due to cash shortages. A more plausible explanation is that this reflects access to cheap capital, signaling a subtle shift in how financing is being conducted during this AI-driven market surge. Everyone wants to participate in these capital maneuvers, and NVIDIA has recognized that issuing bonds is even cheaper than other financing options—so it simply went ahead with it. At least for now, NVIDIA remains in strong financial health.

Why did he rebalance his portfolio?

Josh Kale: This brings us back to another question: What exactly is Leopold thinking? Why has his judgment changed? The stock price chart you just showed also indicates that NVIDIA’s recent performance hasn’t been particularly strong, but it hasn’t been terrible either. It’s still the world’s largest company with a market cap close to $5 trillion, and a 7% drop in a month is negligible against the backdrop of other AI stocks surging.

Ejaaz Ahamadeen: I don’t believe NVIDIA will disappear. I think its GPUs, including the new CPU product line launched just weeks ago, will perform exceptionally well. Demand for AI products is currently exponentially oversupplied, and NVIDIA remains the primary core infrastructure provider capable of meeting this demand.

But I truly feel that the classic AI "selling shovels" trade has become too crowded, and Leopold’s recent position changes are signaling exactly that. Reviewing his latest 13F reveals that his bearish positions are heavily skewed toward the semiconductor sector, including NVIDIA, ASML, Oracle, and several other infrastructure-level companies.

At the same time, he heavily invested in memory, power, and new cloud technologies. This suggests his view is not that AI infrastructure has reached its peak, but rather that certain layers within the infrastructure stack—particularly semiconductors and traditional popular assets—have become overly crowded.

If the question becomes where the funds will go next, there are two answers. The first is the most direct: the funds will flow to the next real infrastructure bottleneck—such as electricity, memory, and data center networks. The second answer is the mysterious investment that was only revealed a few weeks ago.

Accidentally Exposed Anthropic Position

Josh Kale: This was the most surprising thing to me—I only found out yesterday after you mentioned it, and my first reaction was, “That’s impossible.” Could it really be that 20% of Leopold’s fund, “Situational Awareness,” is allocated to Anthropic’s equity? External rumors suggest that this company makes up about one-fifth of Leopold’s fund position, and outlets like The Wall Street Journal and several others have reported this, with insiders close to the transaction confirming it.

This became a completely unexpected card in his portfolio.

Because 13F filings only disclose publicly traded holdings and do not include private equity, and Anthropic is precisely a significant piece of non-public equity. This is precisely why people have begun to understand why the market has valued his portfolio at $20 billion.

If 20% of the fund is allocated to Anthropic and he invested in early 2025, the returns from Anthropic alone have been equivalent to seven years' worth of growth in just one year. This shift would require a significant revision of our understanding of his entire portfolio.

Ejaaz Ahamadeen: Yes. He first invested in Anthropic through private channels or funds around March 2025, when Anthropic was valued at approximately $60 billion. Based on the most recent valuation, it is now valued at $965 billion.

This represents an almost 15-fold increase. According to the algorithm we showcased in today’s episode, the liquidity portfolio disclosed in his most recent 13F filing is valued at $13.7 billion. Adding the Anthropic position mentioned in the Wall Street Journal report would increase this by approximately $7 billion, bringing the total assets under management to $20 billion.

How exaggerated is this? Bill Ackman, a top investor with three to four decades of experience in the market, has a Pershing Capital fund of roughly $20 billion. Leopold has only been in this game for a year and a half, is just 24 years old, and has virtually no meaningful investment experience.

But he made some incredibly astute predictions—crazy as it is, he had already written all of this down in advance. When he launched the fund a year and a half ago, he published a 65-page AI essay titled “Situational Awareness,” in which he laid out nearly the entire logic, including how capital would rotate from semiconductors and certain infrastructure segments to other bottlenecks. The market is now unfolding exactly along this trajectory, which is truly astonishing.

The next wave of infrastructure momentum

Ejaaz Ahamadeen: So this also tells me where the next wave of capital will flow. If he’s cautious about NVIDIA, the money will shift toward power and memory; at the same time, he wants to invest directly in the “mine” itself, rather than just continuing to buy the “shovels”—Anthropic is his favorite mine.

Josh Kale: This does seem like a new trend, and he’s still ahead of most people. Over the past 12 months, everyone has been looking for bottlenecks in AI—rare metals, memory, RAM, and so on—and the market has already moved on those. Those assessments weren’t wrong, because that rally did indeed happen.

But now, the valuations of those previously seen as bottlenecks are gradually becoming more reasonable. People have gained a better understanding of these companies’ business models, market potential, and future revenues, so much of their value has already been priced in. In the next round, we’re more focused on where the next wave of capital will flow.

You mentioned land, electricity, server racks, and physical infrastructure—this direction seems correct. Because if we think about what AI truly needs most, the answer is increasingly becoming the capability to build physical infrastructure. Look at xAI, or more accurately, at SpaceX, which is now publicly traded: its core revenue doesn’t come from rockets themselves, but from AI infrastructure development.

Looking at this latest deal with Anthropic and Google, the value generated has already surpassed the combined total of Starlink, Starship, and the entire satellite business. Clearly, there is immense demand and immense value here. The question then becomes: who can actually build these things?

SpaceX is clearly one answer. Its stock price reached $230 after hours last night, implying a valuation of approximately $3.1 trillion. We’ll dedicate a full episode this week to SpaceX because its recent surge is extraordinary—it just completed the acquisition of Cursor, and its valuation has now hit $3 trillion; Elon’s daily earnings even surpass the total amount Warren Buffett earned over his entire career.

Who will benefit from the next round?

Josh Kale: We’re interested in identifying which companies are best at building this kind of hardware infrastructure—specifically, those developing the “machines that make machines.” Aligned with Leopold’s focus and the broader trend, we believe the next wave of capital will flow here. So, Ejaaz, in practice, which companies will this rotation actually benefit?

Ejaaz Ahamadeen: Many will be infrastructure companies that don’t sound glamorous. One name that has come up frequently over the past month is Marvell. A few weeks ago at the Computex conference in Taiwan, Jensen Huang directly stated on stage that it will be the next trillion-dollar company.

Three months before he made this statement, NVIDIA had just invested $1.5 billion in Marvell. I’m already having trouble determining whether this constitutes insider trading or market manipulation, because after his comment, the stock price rose another 70%.

It’s easy to conclude right now that AI infrastructure has peaked; however, if you compare it to historical financial crises, such as 2008, the hallmarks of extreme leverage, complex financial engineering, and systemic manipulation have not yet fully emerged in this cycle.

There are two key differences. First, the products these companies are producing today are genuinely being purchased by real customers—unlike during the dot-com bubble or the financial crisis, there is no lack of solid, real demand. Second, due to physical constraints, we simply cannot infinitely increase leverage, as the entire system is bottlenecked by human labor and construction capacity.

Even if you raise more money, you can't build data centers fast enough, scale up memory chip production sufficiently, or immediately expand the power grid, transmission lines, and related infrastructure. There aren't enough people on the ground, and approvals, regulations, and various procedures are holding you back.

So I actually think this gives investors an advantage. Since you already know that the hottest chips and shovel-selling trades are too crowded, money will next flow into power and data networks—companies like Astera Labs—and then into other related sectors.

What you should really be thinking about is when these contracts will begin to pay off, when these fabs will actually be built, when SpaceX’s rockets will launch AI satellites into orbit, and when we’ll be able to start using solar power to train AI models.

The timeline dictates the betting rhythm. At least, that’s the framework I’ve been using for my own investments—though this is not investment advice. I view it this way because over the past year and a half, we’ve clearly witnessed capital shifting from broad AI stocks toward semiconductors and infrastructure plays.

Josh Kale: If you look further at this composite chart, you’ll find that this narrative is clearly reflected in his portfolio allocation. By category, what’s his largest holding? Power and energy. Next is memory, followed by cloud and GPU miners—the most tangible infrastructure.

He wants to hold companies like CoreWeave, new cloud providers, as well as miners that have transitioned into cloud computing. What he wants to own is the physical infrastructure, because he believes this is the true bottleneck. You just mentioned it—there are still many finer details involved, such as actual construction, hardware manufacturing, and the challenges of building data centers themselves, all of which are extremely difficult.

If there’s one major bottleneck, it might even be regulatory approvals. Who’s addressing these issues? SpaceX wants to move data centers into space, and Tesla aims to use humanoid robots to solve labor shortages. But both of these solutions are still far off. In the short to medium term, there are abundant untapped opportunities—and that’s precisely the direction Leopold is betting on.

Advantages of optical modules and fiber optics

Josh Kale: I’d also like to add a detail we didn’t fully explore earlier. For those looking to dig deeper and seek additional yield, many of his insights are hidden in optics and the underlying tech stack. Ejaaz, you’ve been researching this recently—could you walk us through his approach?

Ejaaz Ahamadeen: If you look at these positions on his screen, CoreWeave and Iron are essentially among the top-tier new-generation cloud service providers. Simply put, they’re similar to Amazon Web Services, except that while AWS provides cloud services to internet companies, these companies offer ready-made GPU infrastructure specifically for AI firms.

They handle everything for you—GPUs, networking, deployment—so AI companies don’t have to worry about underlying infrastructure and can focus directly on training models and accessing computing power. CoreWeave and Iron have been among the largest concentrated positions since he built his stake, delivering the highest returns.

It’s also worth noting that he still holds these two companies as his largest positions today. This further indicates that, in his view, this investment is far from over. Moreover, he has privately invested in Core Scientific, a company that can help unlock CoreWeave’s infrastructure capacity—effectively adding another layer of leverage to his CoreWeave position.

Beyond these, consider companies like Coherent and Lumentum, which are essentially suppliers related to fiber optics and optical connectivity. In the simplest terms, for semiconductors and GPUs to communicate with each other, the traditional method typically requires a large number of copper wires.

The issue is that as GPU scales grow larger, copper wires become hotter, energy losses increase, and efficiency drops significantly. In this context, fiber optics become the next logical upgrade: they enable faster data transmission, offer better cost efficiency, and allow companies providing inference and training compute power to earn more. You’ll find that his investments have consistently focused on foundational infrastructure—backing both optical companies and power-related firms. It may not sound glamorous, but in my view, this is precisely where the money is truly flowing right now.

Josh Kale: I find copper particularly interesting because I’ve only recently realized how critical it is for short-distance data transmission. In many high-bandwidth, short-range scenarios, copper is almost the only material people truly want to use. It’s only when it becomes unsuitable—such as over long distances or due to excessive heat—that you switch to fiber optics. As a result, there’s now strong market demand for a combination of copper and fiber optics, which is why watching copper as an asset is so compelling.

Copper futures have recently been very strong, essentially because everyone needs it—it’s the most critical foundational material for short-distance, high-bandwidth transmission, with fiber optics being the next step.

Thinking at a more fundamental level, the materials sector has always been fascinating. At the very bottom of everything, the core raw materials needed to achieve intelligence are what truly matter—copper is one, lithium is another, and there are many more. We really should do a dedicated episode on materials. Perhaps Leopold hasn’t reached that level yet, but we might be able to spot the next rotation before he does.

Josh Kale: If you keep going down the stack, you could even visit copper mines to see how these things are actually made. But returning to the core assessment, I believe the next rotation will indeed shift from those seemingly smaller bottlenecks toward truly difficult challenges—namely, hardware and large-scale data center construction.

Whoever can build out data centers will make the most money. We’ve already seen how much money SpaceX has made due to the immense demand for data centers. Whoever can bring more data centers online faster and provide sufficient power and GPUs will earn the most. This is essentially the direction Leopold is betting on now.

Is a bubble forming?

Josh Kale: In summary, we don’t believe we’re currently in a bubble burst phase. Leopold’s position changes seem more like rotation rather than a full retreat. So, should we still follow him?

Ejaaz Ahamadeen: I admit, when I first saw his 13F, my initial reaction was that it was insane for him to be shorting a company that is the most valuable in the world and has demand booked all the way through 2029. But now, seeing this financing round, I’m starting to think that if NVIDIA continues to take on more external debt—or even sells shares in the future—and if this trend persists, Leopold might end up being right again.

If that’s true, his fund could eventually surpass the world’s top traders and best investment funds. He’s truly been winning consistently—it’s hard not to be impressed.

Josh Kale: But another important point is that his entire life so far has been spent only going long—he has never truly been tested by large-scale selling. As we mentioned earlier, achieving 30x returns and surviving in the market for 30 years are two entirely different things.

If he can truly sustain this growth while also learning when to sell, how to manage risk, and how to use hedging to protect himself, it becomes even more formidable. We’re already beginning to see the early signs of this capability. That $9 billion short position wasn’t achieved by directly shorting $9 billion in cash—it was executed through options and leverage, not a one-to-one naked short. Regardless, this development is definitely worth continued observation.

Energy is the core bet.

Josh Kale: If you had to pick one stock from his entire portfolio that you'd most want to buy, which one would it be?

My own answer is energy stocks. I’ve always been bullish on energy because, even if AI demand slows, energy itself remains a global necessity, and this demand will only continue to grow.

Even without considering AI, we still need more energy and more electricity. Companies like Bloom Energy that enhance power generation and delivery capabilities are the areas I’m most excited about, because they most closely resemble hedge-like bets. The one consistent trend that will continue to rise regardless of circumstances is our growing demand for energy, electricity, and power—these companies are the ones I’m most willing to hold long-term.

Ejaaz Ahamadeen: My answer is a bit of a cheat. The companies I most want to follow are those that Jensen is investing in, while also aligning with Leopold’s logic. The closest I’ve come to copying trades is Marvell. Although it’s not a company Leopold publicly holds, it strongly aligns with his bets on fiber optics and power, and Jensen has already invested $1.5 billion in it.

I’ve noticed a pattern: whenever Jensen invests in a company through NVIDIA—whether it’s Intel, CoreWeave, or another—its price tends to keep rising afterward. That’s why my current position is roughly here. I also hold some CoreWeave myself, since both Jensen and Leopold are extremely bullish on it.

Josh Kale: Marvell has risen 270% over the past six months. This might be a solid rule of thumb: when people like Jensen—or even someone with massive influence like Trump—publicly say to buy a stock, it’s often worth taking a serious look.

This signal has been proven multiple times in the past to have significant upside potential. Whether it’s Intel or Marvell, these cases demonstrate that they truly understand what they’re talking about and also have the ability to influence the outcomes of these companies. So this market move has been truly wild.

I hope it continues. Based on current conditions, it’s very likely to do so. At least for now, we remain bullish and optimistic, and will continue to assess daily developments.

Josh Kale: Is there anything else you'd like to add about Leopold's portfolio update?

Ejaaz Ahamadeen: I’d really like to hear what those who are skeptical think. If, after listening to our analysis above, you feel we’re completely wrong or have misunderstood something, please feel free to point it out.

Yesterday, I stared at the news about NVIDIA’s $25 billion funding round for a long time, originally intending to find fault. But if you look purely at the financial logic, it actually makes sense.

Why not borrow this nearly risk-free, low-cost money? Borrowing to expand is clearly more sensible than selling your own equity, as it allows you to retain more of your future earnings.