NewsFlash Articles Data Fundraising Skill&API

US Stock Market's "Architect of the Version" Latest Rebalancing Analysis: $9 Billion Short on NVIDIA, Bullets Fired at the Power and Memory Sector

Read this article in 45 Minutes

The best long-term play is not necessarily the hottest chip company.

Original Video Title: Leopold Aschenbrenner says 「No More Stocks!」
Original Video Source: Limitless Podcast
Original Text Translation: DeepTech TechFlow

Key Points Summary

Leopold Aschenbrenner, seen as one of the world's most radical AI investors, is simultaneously shorting NVIDIA, ASML, and Oracle with a nominal position of around $9 billion in the public markets while shifting funds towards electricity, memory, data center networks, and deeper AI infrastructure and model assets like Anthropic.

The two hosts believe that this does not mean the AI bubble has burst, but rather signifies a rotation signal in infrastructure investing from "chip-first" to "energy, network, data center construction-first," especially as NVIDIA just completed a $25 billion bond offering and with Anthropic's valuation soaring, the market implications of this judgment are rapidly amplifying.

Highlights of Insightful Views

Leopold's Core Trading Logic

· "The 'most classic' 'selling shovels' trade in AI has become too crowded, and Leopold's recent positioning changes convey that signal."

· "His judgment is not that AI infrastructure has peaked, but that certain levels of the infrastructure stack, especially semiconductors and traditional popular targets, have become overly crowded."

· "If the question becomes where the money will flow next, there are two answers. The first, most direct answer is flowing towards the next real infrastructure bottleneck, which is electricity, memory, data center networks, these links. The second answer is the mysterious investment that was exposed just weeks ago."

· "What he's actually betting on has always been very infrastructure-oriented, investing in both optics companies and electricity-related companies."

· "If he is cautious about NVIDIA, then the funds will go towards electricity, memory, and such; at the same time, he also wants to invest directly in the 'mine' itself, rather than continuing to just buy 'shovels,' and Anthropic is his favorite 'mine'."

Signal from NVIDIA's Financing

· "The issue is not whether NVIDIA will continue to make money, but why a highly profitable company with a large cash reserve would still need to borrow an additional $250 billion externally."

· "If a company is aggressively buying back stock, significantly increasing dividends in the same month, and borrowing money at the same time, it's clearly not borrowing out of necessity. A more reasonable explanation is that this is cheap funding, and the financing method for this AI boom is starting to undergo a slight shift."

The Next Wave of AI Infrastructure Dividend

· "The real bottleneck is no longer just the GPU, but rather power, memory, data center networks, and the actual ability to build all of these things."

· "Even if you raise as much money as possible, you won't be able to build data centers fast enough, expand memory chip capacity enough, or immediately scale up the power grid, electrical lines, and related infrastructure. There aren't enough people on the ground, and various approvals, regulations, and procedures are hindering you."

· "Whoever has the ability to build data centers will take the money and run."

Photonics, Copper, and Optical Fiber

· "As GPU scale continues to increase, copper wires will become hotter, energy consumption will rise, efficiency will deteriorate, and optical fiber will become the next upgrade path."

· "In high-bandwidth short-distance transmission scenarios, copper is almost the only material everyone really wants to use. Only when it becomes unsuitable, such as when the distance is too far or the heat is too high, will they switch to optical fiber, so the market's demand for a combination of copper and optical fiber is very strong right now."

· "Copper futures have recently been strong fundamentally because everyone needs it; it is the most critical basic material for short-distance high-bandwidth transmission, while optical fiber is the next step."

· "Copper is still the most critical material for short-distance high-bandwidth transmission, but once the distance extends or the heat becomes too high, a shift to optical fiber is necessary."

· "The next round of funding will flow to infrastructure companies that may not sound glamorous."

Why Energy Is the Safest Bet

· "I have 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 rise."

· "A single trend that will continue to rise in any scenario is our demand for energy, electricity, and power. These are the companies I am most willing to be long on for the long term."

· "The person I most want to shadow is someone like Jensen who combines investing with Leopold's logic. So right now, the closest target I am considering following is Marvell."

· "The best long-term positions are not necessarily in the hottest chip companies, but in those that are essential to the power infrastructure and cannot be bypassed in any macroeconomic scenario."

Leopold's AI Investment Portfolio

Josh Kale: Leopold Aschenbrenner, a 24-year-old specializing in AI investments, is now almost universally regarded as the world's top AI investor. Rumor has it that the AUM of his fund has exceeded $20 billion. When we saw Ejaaz's post a month ago, the fund's AUM was only $13.7 billion, which means it has basically doubled every quarter.

This time, we have obtained some very important new changes in his recent investment trends. In the last issue, we discussed his portfolio, and the most surprising thing at the time was that he was actually shorting a company that almost everyone is familiar with, which is NVIDIA, the world's largest market cap and the most hyped AI company. Many people couldn't understand why he would take over a $9 billion short position on such a company.

Now we have received a new clue that may explain this. NVIDIA is actually raising funds, and it is through debt financing. On the surface, this seems very illogical, why would a company as massive and highly profitable as NVIDIA need another $25 billion in cash just after completing a $25 billion cash raise? Today, we want to discuss why Leopold can earn so much, what he is looking at next, and what this funding round by NVIDIA really means, while integrating it with Leopold's investment portfolio.

Ejaaz Ahamadeen: First, a bit of background for everyone. Leopold Aschenbrenner used to be a researcher at OpenAI and raised a fund about a year and a half to two years ago. Initially, the fund was not very large, probably around $200 million from what I remember, but based on his most recent 13F, the fund's publicly disclosed holdings are now valued at $13.7 billion.

So naturally, the market wants to know, what positions has he taken, what is his core investment thesis, and where will the next big trade land.

To understand this, you need to know that up until about a month ago, Leopold was actually very optimistic about the entire AI track, especially bullish on the "selling shovels" logic, such as NVIDIA and other GPU and upstream hardware vendors.

However, about a month ago, the market found that he was not as bullish on the semiconductor side. He still favored memory, power, and other real bottleneck areas, and may have also been positive on new cloud vendors, but he specifically did not have a positive view on the world's most valuable company, NVIDIA. More specifically, he had a total of about $9 billion in short positions spread across NVIDIA, ASML, Oracle, and a few other companies seen as core beneficiaries of AI infrastructure.

Logic Behind Shorting NVIDIA

Ejaaz Ahamadeen: When this news came out, many people started to worry, wondering if the AI bubble was about to burst. After all, on the surface, NVIDIA's GPUs are still selling well, and demand has not shown any significant weakening, so where is the problem?

Later, we uncovered a few new clues, with the most important one being that NVIDIA had just raised $25 billion through bond financing. This means it did not just use its own cash but leveraged itself further. So the question arises: Why would a company that is the most profitable, with the highest profit margins and strongest cash flow, need to borrow $25 billion externally?

Josh Kale: And initially, they only planned to raise $20 billion, but it ended up being increased to $25 billion, and subscriptions were oversubscribed by over 3 times. Last time we discussed this investment portfolio, we said, don't worry about the bubble yet because even though these companies have huge capital expenditures, their revenue is high enough to theoretically support expansion using their own balance sheets.

But this is the first time since 2021 that NVIDIA has significantly raised funds externally rather than directly using its cash reserves. I remember it currently has around $12 billion in cash reserves. Putting all this together creates a strange tension: on one hand, Leopold is shorting, and on the other hand, NVIDIA seems to have unlimited cash and profits, yet they are still issuing bonds. So what exactly is happening?

Breakdown of NVIDIA's Bond Financing

Josh Kale: Ejaaz, can you help us break down this transaction? Because this is not your typical fundraising; it's a bond issuance. Essentially, NVIDIA has added another $25 billion to its balance sheet, and the interest rate seems to be very low.

Ejaaz Ahamadeen: I'll lay out two explanations. NVIDIA had about $13.7 billion in cash on hand, meaning it could have easily spent its own money. So why opt for external financing? The simplest analogy is buying a house. Many people, even if they have the full amount, still choose to take out a loan because their own capital can be used elsewhere, and if the cost of borrowing is low enough, it's actually more cost-effective.

In recent years, the interest rate environment hasn't been very favorable. But if you're NVIDIA, one of the world's most valuable and sought-after companies, you can borrow money on quite favorable terms. This $25 billion bond issuance has maturities ranging from 2 to 30 years, making it almost cheap money, with interest rates approaching U.S. Treasury yields.

Furthermore, this financing was oversubscribed by about 4 times, meaning there was $85 billion looking to enter the $25 billion tranche, and NVIDIA could almost freely choose investors. If we only consider the official explanation, NVIDIA states that this is mainly a financial arrangement to repay and refinance some existing debt. Google did something very similar a few weeks ago and also in February. So you can certainly accept this explanation and see it as financial optimization.

But the other side is hard to ignore: in the past month and a half, NVIDIA, Amazon, Google, and several other hyperscale cloud companies have almost all resorted to external financing. Some issued bonds, some sold stocks. Leopold's view may not be entirely unreasonable. Could this be a sign of the bubble starting to deflate or the house of cards starting to shake? However, if you only look at the financial structure, this matter does not yet point to danger.

Josh Kale: I see it the same way. A $9 billion short position on NVIDIA is indeed a very large position. But while doing our research, we also noticed another thing: on May 18, the NVIDIA board just authorized an additional $80 billion buyback, and increased the dividend from $0.01 per share to $0.25, a direct 25-fold increase.

If a company in the same month is heavily buying back stocks, significantly increasing dividends, and then going into debt, it's clearly not borrowing out of necessity. A more reasonable explanation is that this is cheap funding, and the financing method for this AI rally is undergoing a slight shift. Everyone wants to participate in these capital operations, and NVIDIA has also realized that issuing bonds to raise money is even cheaper than other financing methods, so it just went ahead and did it. At least for now, NVIDIA itself is still doing well.

Why Did He Rebalance

Josh Kale: This brings us back to another question. What was Leopold thinking? Why did his assessment change? The stock price chart you just showed also indicates that NVIDIA has not been performing particularly well recently, but it hasn't been terrible either. It is still the world's largest company with a market cap close to $5 trillion. A 7% drop in a month is nothing significant, especially compared to the skyrocketing AI stocks.

Ejaaz Ahamadeen: I don't think NVIDIA is going away. I believe both its GPU, including the CPU product line launched just a few weeks ago, will perform exceptionally well. The demand for AI products is currently exponentially high, and the core machine provider truly capable of meeting this demand is still primarily NVIDIA.

However, I do feel that the most classic "selling shovels in a gold rush" trade in AI has become too crowded, and Leopold's recent portfolio changes convey this signal. Just look at his recent 13F filing; his bearish positions are notably short on the semiconductor line, such as NVIDIA, ASML, Oracle, and several other infrastructure-level companies.

At the same time, he has heavily invested in areas like memory, power, and new cloud technologies. This indicates that his assessment is not that AI infrastructure has peaked, but that certain levels of the infrastructure stack, especially semiconductors and traditional popular targets, have become too crowded.

If the question shifts to where the money will flow next, there are two answers. The first is the most straightforward, which is flowing towards the next real bottleneck in the infrastructure, namely power, memory, data center networks, and so on; the second answer is the mysterious investment that was exposed just a few weeks ago.

Unexpected Exposure of Anthropic Position

Josh Kale: This is the most surprising to me. I only found out about it yesterday after you mentioned it, and my initial reaction was disbelief. Could it be that 20% of Leopold's fund "Situational Awareness" is actually allocated to Anthropic's equity? There are now rumors circulating that this company represents approximately one-fifth of Leopold's fund's holdings. Reports from The Wall Street Journal and several other media outlets claim this, and individuals very close to the deal have also confirmed it.

This has become a card in his portfolio that was completely unexpected by the market.

Because 13F can only disclose public market holdings and does not reveal private equity, and Anthropic is precisely a large chunk of off-market equity. It is for this very reason that everyone has begun to understand why the external world has raised the valuation of his portfolio to $20 billion.

If 20% of the fund is in Anthropic, and he probably invested in it in early 2025, the return for that year was as if he had lived through seven years with Anthropic. This change will cause a significant revision in our understanding of his entire investment portfolio.

Ejaaz Ahamadeen: Yes. He first invested in Anthropic through a private channel or fund around March 2025 when Anthropic was valued at about $600 billion. Now, based on the latest valuation round, it has been pegged at $965 billion.

This is close to a 15x increase. Based on the algorithm displayed in today's program, the liquidity portfolio value disclosed in his most recent 13F filing was $13.7 billion. If we add the portion of the Anthropic position reported in The Wall Street Journal, it could increase by another $7 billion, bringing the total AUM to $20 billion.

How extraordinary is this? Bill Ackman, a top investor who has been in the market for three to four decades, with Pershing Capital, also has a size of about $20 billion. Leopold entered this game just a year and a half ago, and he is only 24 years old, with almost no meaningful investment experience.

But he has made some extremely amazing judgments, and the crazy thing is that he actually wrote it all out in advance. When he launched the fund a year and a half ago, he published a 65-page AI essay "Situational Awareness," laying out the entire logic almost completely, including how funds will rotate from the semiconductor and some infrastructure links to other bottleneck constraints. The market is now developing along this line, which is truly amazing.

The Next Wave of Infrastructure Rally

Ejaaz Ahamadeen: So, this also tells me where the money will flow next. If he is somewhat cautious about NVIDIA, then the funds will go to places like power and memory; at the same time, he also wants to invest directly in the "mine" itself, rather than continuing to just buy the "shovels", and Anthropic is his favorite mine.

Josh Kale: This indeed looks like a new trend, and he is still ahead of most people. Over the past 12 months, everyone has been looking for the AI bottleneck, rare metals, memory, RAM, and so on, and the market has experienced a rally. These assessments were not entirely wrong because that rally did happen.

But now, those areas considered bottlenecks are gradually approaching a reasonable valuation. People now have a better understanding of these companies' business models, market space, and future revenue, so much of the value has actually been priced in. In the next round, what we are more concerned about is where the money will continue to flow.

You just mentioned land, power, casing, physical infrastructure, and that direction seems to be correct. Because if we think about the most important thing for AI, the answer increasingly resembles physical construction capability. Look at xAI, or more accurately, look at SpaceX, which is now listed; its core revenue is not the rockets themselves but the AI infrastructure development.

Looking at its recent transaction with Anthropic and Google, the value it has brought has already exceeded the sum of Starlink, Starship, and the entire satellite business. There is obviously enormous demand and value here. So the question becomes, who can truly build these things?

SpaceX is clearly one answer. After-hours last night, its stock price had reached $230, corresponding to a valuation of approximately $3.1 trillion. We will have a dedicated episode on SpaceX this week because its recent trend has been too exaggerated; just completing the acquisition of Cursor, it is now valued at $3 trillion, and Elon is making money at a rate that even surpasses Warren Buffett's entire career earnings in a day.

Who Will Reap the Benefits of the Next Round

Josh Kale: What we care about is which companies are best at building this kind of hardware infrastructure, skilled at developing those "machines that make machines." Combining Leopold's direction with the overall major trend, we believe that funds will move in this direction next. So Ejaaz, which companies will ultimately benefit from this rotation in reality?

Ejaaz Ahamadeen: Many of them will be those seemingly unsexy infrastructure companies. One name that has been frequently mentioned in the past month is Marvell. A few weeks ago at the Computex event in Taiwan, Jensen Huang directly stated on stage that this would be the next trillion-dollar company.

And just three months before he made this statement, NVIDIA had just invested $1.5 billion in Marvell. I'm starting to wonder whether this could be considered insider trading or market manipulation because after he said this, the stock price went up by another 70%.

Right now, I think it's too easy to directly judge whether the AI infrastructure has peaked. But if you compare it to historical financial crises, such as 2008, that highly leveraged, financial engineering, and systemic manipulation vibe hasn't fully appeared in this round yet.

The key differences lie in two aspects. First, the products companies are producing today are actually being purchased by real users. Whether it's the dot-com bubble or the financial crisis period, there wasn't such a solid real demand. Second, constrained by physical laws, we actually can't leverage infinitely now because the entire system is constrained by manpower and construction capacity.

No matter how much you raise, you can't build data centers fast enough, expand memory chip production capacity enough, or immediately scale up the power grid, power lines, and related infrastructure. There aren't enough people on the ground, and approvals, regulations, and various procedures are also hindering you.

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 flow next into electricity, data networks, such as companies like Astera Labs, and then into other related areas.

What you really need to think about is when these contracts will start to pay off, when these semiconductor fabs will actually be built, when SpaceX's rockets can launch AI satellites into space, and even when we can start using solar energy to train AI models.

The timeline determines the pace of the bet. At least that's the framework I'm using for my investment, but of course, this is not investment advice. The reason I see it this way is that over the past year and a half, we have witnessed funds flow from broad AI stocks to semiconductor and infrastructure trades.

Josh Kale: If you continue to look at this portfolio chart, you'll actually see that this story has already been clearly written into his holdings structure. By category, what is his largest allocation? It's electricity and energy. Next is memory, followed by cloud and GPU miners, which are the most tangible forms of infrastructure.

He wants to hold new cloud players like CoreWeave, and he also wants to hold miners who have transitioned to cloud computing power. What he wants to own is these physical infrastructure entities because he believes this is where the real bottleneck lies. As you just mentioned, of course, there are many more detailed aspects, such as actual construction, hardware manufacturing, and the construction of data centers, all of which are extremely challenging.

If we're talking about the biggest bottleneck, it might even be the approval process. So who is addressing these issues? SpaceX wants to move the data center into space, and Tesla wants to use a humanoid robot to solve labor problems. But both of these things are still a long way off. In the short to medium term, there are actually a lot of blank opportunities, and this is exactly the direction Leopold is heading.

The Advantages of Photonic Modules and Optical Fiber

Josh Kale: I would like to add a detail that we didn't discuss before. For those who want to dive deeper and find more alpha, many of his clues are actually hidden in optics and the lower layers of the tech stack. Ejaaz, you've been researching this recently, can you talk about his approach?

Ejaaz Ahamadeen: If you look at his positions on the screen, CoreWeave and Iron are basically the top-tier new cloud providers. Simply put, they are a bit like Amazon Web Services, except AWS provides cloud services to internet companies, while these companies provide ready-made GPU infrastructure to AI companies.

They help set up everything for you—GPU, networking, deployment—so that AI companies don't have to worry about the underlying infrastructure and can directly train models and access computing power. CoreWeave and Iron have been one of the largest concentrated positions since he started building his portfolio, and have also brought the highest returns.

Moreover, it's worth noting that he still has these two companies in his largest positions today. This also indicates one thing: in his view, this transaction is far from over. Furthermore, he has privately invested in Core Scientific, a company that can help unleash CoreWeave's infrastructure supply capabilities. In a sense, he has added another layer of leverage to CoreWeave.

In addition to these, if you look at companies like Coherent and Lumentum, they are essentially suppliers of fiber optics and optical connectivity. Put simply, when communicating between a semiconductor and a GPU, the traditional approach usually relies on a large amount of copper wire.

The issue is, as GPU scales grow, copper wire will become hotter, energy loss will increase, efficiency will become poor, and optical fiber will become the next upgrade path in this scenario. It can achieve faster data transmission, higher cost efficiency, and allow companies providing inference and training power to earn more money. So you'll find that what he's actually betting on has always been very infrastructure-oriented—investing in these optical companies as well as power-related companies. It may not sound very sexy, but in my opinion, this is where the money is really being deployed now.

Josh Kale: The thing about copper is also very interesting to me because I recently realized how crucial it is for short-range data transmission. In many high-bandwidth short-distance transmission scenarios, copper is almost the only material everyone really wants to use. Only when it becomes unsuitable, such as when the distance is too far or the heat is too high, do they switch to fiber optics, so the market now has a very strong demand for a combination of copper and fiber optics, which is why observing this copper transaction is very interesting.

Copper futures have recently been performing strongly, essentially because everyone needs it; it is the most crucial foundational material for short-range high-bandwidth transmission, with fiber optics being the next step.

At a more fundamental level, thinking about materials has always been very interesting. Underneath all the underlying layers, in essence, is the pursuit of intelligence, the most core need for which raw materials are required. Copper is one, so is lithium, among many others. We really should do a dedicated materials feature. Maybe Leopold hasn't even reached that level yet, but we are able to see the next round of rotation ahead of time.

Josh Kale: If we keep going deeper to the stack, we could even go directly to the copper mine to see how these things are made. But returning to the core judgment, I think the next round of rotation is indeed shifting from those seemingly smaller bottlenecks to the truly challenging things, which are hardware and large-scale data center construction.

Who has the ability to build data centers will take away the money. We have already seen how much money SpaceX has made because of the strong demand for data centers. Whoever can bring more data centers online faster, who can provide enough power and GPUs, will make the most money. This is basically the direction in which Leopold is currently detained.

Is a Bubble Emerging?

Josh Kale: In summary, we do not believe that we are already in the bubble burst stage. Leopold's position seems more like a rotation rather than a complete retreat. So, should we continue to follow him?

Ejaaz Ahamadeen: I admit that when I first saw his 13F, my initial reaction was that this guy actually has a bearish view on the world's most valuable company, with demand projected all the way to 2029, which seemed outrageous. But now, seeing this financing round, I am starting to think that if NVIDIA continues to increase its external debt in the future, or even potentially issue shares, if this trend continues, then Leopold might once again be right.

If that's the case, his fund may eventually surpass the world's top traders and best hedge funds. He has truly been winning, and it's hard not to be impressed.

Josh Kale: However, there's another crucial point to consider. His life so far has mainly involved going long, without truly experiencing the test of a major sell-off. We mentioned Bill Ackman earlier, achieving 30x returns and surviving 30 years in the market, which is a different story altogether.

If he can indeed sustain this growth, learn when to hit the sell button, manage risk, employ hedging to protect his positions, it becomes even scarier. We are already starting to see a glimpse of this ability. The $9 billion short position was not a straightforward $9 billion cash short, but achieved through options and leverage, not a direct naked short. In any case, this 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 personally like to buy, which one would it be?

My own answer would be energy stocks. I have always been bullish on energy because even if AI demand slows down, energy remains a global necessity, and this demand will only rise.

Even without considering AI, we need more energy, more power. Companies like Bloom Energy, which can enhance power generation and distribution, are the most exciting to me because they resemble a hedge. A single trend that will continue to rise in any scenario is our demand for energy, electricity, and power. These are the assets I am most willing to be long on in the long term.

Ejaaz Ahamadeen: My answer is a bit of a cheat. What I'm most interested in following are those companies that align with Jensen's investment style while intersecting with Leopold's logic. The closest target I'm tracking right now is Marvell. Although it's not a company Leopold publicly holds, it aligns well with his bets on fiber optics and power, and Jensen has already invested a solid $1.5 billion.

I've noticed a trend that whenever Jensen invests in a company through NVIDIA, whether it's Intel, CoreWeave, or others, it tends to keep rising. So my current position is mostly here. I also hold some CoreWeave myself because both Jensen and Leopold are extremely bullish on it.

Josh Kale: Marvell has already surged by 270% in the past 6 months. This may really be a good rule of thumb: for people like Jensen, or even influential figures like Trump, if they publicly say to buy a certain stock, many times you should really take a serious look.

It has been proven many times in the past that such signals often have a lot of room for fulfillment. Whether it's Intel or Marvell, these cases demonstrate that, on the one hand, they do understand what they are talking about, and on the other hand, they also have the ability to influence the outcomes of these companies. So, this rally is really crazy.

I hope it continues. From what we see now, it is also very likely to continue. At least for now, we are still biased towards the long side, optimistic, and will continue to make judgments based on the changes every day.

Josh Kale: Regarding Leopold's portfolio update, is there anything else you would like to add?

Ejaaz Ahamadeen: I would actually like to hear what those who are skeptical about this have to say. If you have listened to our analysis just now and feel that we are completely wrong, or if there is any misunderstanding, feel free to point it out directly.

Yesterday, I stared at the news of NVIDIA's $25 billion financing for a long time, intending to find fault. However, looking at it from a financial logic perspective, this matter does make sense.

Why not take this almost risk-free cheap money? Borrowing money from others to expand clearly makes more sense than selling your own equity because this way, you can retain more of the future earnings.