Organization & Compilation: Deep Tide TechFlow

Hosts: Josh, EJ

Original Title: The Best AI Investor Just Shorted the Entire Market

Podcast Source: Limitless (AI Investment Show)

Broadcast Date: May 19, 2026

Editor’s Introduction

Wall Street's most famous AI bull, Leopold Aschenbrenner (former OpenAI researcher, founder of Situational Awareness Fund, taking $250 million to $13.7 billion in 2 years), just submitted the latest 13F filing to the SEC, and an unexpected signal for the entire market has emerged—he has established a short position of up to $8 billion against the entire semiconductor supply chain including Nvidia, AMD, Broadcom, ASML, and Micron, which is 40 times the total assets of his fund 18 months ago, and is the first time since the fund's inception that the short exposure exceeds long exposure.

His new thesis can be condensed into one sentence: the bottleneck for AI investment is shifting from chips (design level) to power and memory (infrastructure level). On the bullish side, he continues to heavily invest in new forces in data centers and power like CoreWeave and Bloom Energy, and has added positions in bitcoin mining companies such as SanDisk, CleanSpark, Riot Platforms, Applied Digital, and IREN that have grid access capabilities. This is a key document for understanding the narrative switch in AI investment in 2026.

Key Quotes

From Bullish on AI to Shorting Semiconductors

• “This is the first time in the fund's history that the size of shorts exceeds the size of longs; an $8 billion short position is 40 times the net worth of the entire fund 18 months ago. This is not hedging; this is a directional bet.”
• “If it were just to hedge realized gains, you would see small-scale hedging positions to offset the long books. But when the total puts are already greater than the longs, that's a bet on the market going down.”

Core Argument: The Bottleneck Has Shifted from Silicon to Electronics

• “The bottleneck has shifted from chips to electronics. There are enough chips; the problem is where to plug them in. Anthropic would rather cooperate with competitor SpaceX to get computing power, not because there aren't enough chips, but because there is no corresponding infrastructure to deploy them on a large scale.”
• “Leopold understands data centers and GPUs better than anyone else in this market; he has researched this issue longer than anyone. So he knows better than anyone where the next bottleneck is, which he believes is power and energy.”
• “I don’t think he is really bearish on GPUs; he just feels that in the short term it is an overcrowded trade, and his money is better spent on power and memory.”

On Neocloud and Power: A Trade That Can Profit from Both Sides

• “Neoclouds like CoreWeave have access to power that Nvidia itself does not. Betting on Neoclouds is not just because they can run GPUs; any capitalized data center can do that; more importantly, they have access permits and capacity of the existing grid infrastructure.”
• “This is a win-win trade; even if the semiconductor sector declines and the valuations of the GPUs they own drop, they will still benefit from power premiums due to their control over capacity.”
• “US bitcoin miners will come online this year with about 30 GW of power capacity. For reference, this is roughly equivalent to the total power planning disclosed by Microsoft, Google, Amazon, and Meta for data centers. They already have power, land, and facilities, they just need to swap out their mining machines for AI acceleration cards.”

Memory and Infrastructure Layer

• “Memory prices are skyrocketing. Major memory manufacturers have seen average prices increase by 300% to 500% over the past 9 months; if you look at production capacity, almost all schedules are booked until the end of 2027.”
• “SanDisk has increased about 40,000% over the past year; logically this should be one of the most crowded trades, but Leopold remains bullish. This is because SanDisk’s core product is NAND flash memory, which AI models use for temporary storage to recall context.”

Nvidia: The Biggest Short and Possibly the Biggest Misjudgment

• “Leopold has a short position of about $1.9 billion on Nvidia (including indirect shorts on the VanEck Semiconductor ETF SMH, which has the highest weight in Nvidia, about 20%).”
• “Nvidia’s moat may be stronger than expected. CUDA is a software-locked platform, and those who have built on it don’t want to leave because remaking customized infrastructure for new chips is very complex.”
• “Old Nvidia GPUs shipped 6 to 8 years ago are now renting for higher prices than they were two years ago, and contracts have to be signed a year in advance.”

Practical Advice for Retail Investors

• “If you are a retail investor who entered this market after watching Leopold's 13F, please be conservative. This is not the time to go all-in on a single stock; the rise of the S&P 500 over the past two years has mainly come from the Mag 7, and capital has trickled down to all the companies we just discussed. This could be an overcrowded trade, so be cautious.”
• “Personally, I am optimistic about two things in the long run: the first is energy; everyone is short on power; the second is manufacturing and construction capacity in the physical world. Whoever has a near-monopoly advantage in manufacturing, building factories, and obtaining grid access permits is an investment worth making.”
• “Nvidia is releasing its earnings report on May 20. If they guide above $78 billion for the next quarter, this batch of put positions will likely be hit.”

The King of AI Bulls Turns Bearish

Josh: Wall Street's most famous AI bull has just called "top" for the entire AI market. Leopold Aschenbrenner, the 24-year-old former OpenAI researcher who founded his own fund after being laid off, has taken $250 million to $13.7 billion in under two years with the "American version of a son" and has now made a move, but his latest portfolio is not what you might expect. He has turned bearish on the entire stock market, establishing an $8 billion short position against the biggest companies in the AI industry, including Nvidia, AMD, Broadcom, and the entire semiconductor supply chain. But he is not entirely pessimistic; he has also revealed where the next biggest AI investment will likely be, which is in power and memory. He has doubled down on data centers and added three completely new companies. We will break this down one by one, but first, let's talk about the biggest changes.

EJ: The world's largest company by market capitalization, the representative of the AI revolution, and the provider of wealth for countless investors, Nvidia, is now in his sights. This is Leopold's largest short position, but you can't see it at a glance from the filing because his top position in shorts is listed as the VanEck Semiconductor ETF (ticker SMH), followed closely by Nvidia itself. Currently, he holds a put option exposure of $1.5 billion directly on Nvidia. For those unfamiliar with put options, a put essentially gives Leopold the right, but not the obligation, to sell the underlying asset at a predetermined price, which means he has bought the right to sell Nvidia at a higher price when the price drops below a certain level.

Josh: SMH is his top short position, with a scale of about $2 billion. I checked its holdings, and the largest single holding is indeed Nvidia, with about a 20% weight. So, adding the top two shorts, he effectively has a short position of about $1.9 billion on Nvidia. This must be a shock for those who firmly believe Nvidia will only rise. But Leopold clearly thinks otherwise.

In addition, Broadcom, Oracle, AMD, Micron, ASML, Intel, and Corning are all newly established shorts. It is worth noting that Intel was once his most famous trade, the one that made the most money in the history of his fund, and he is now shorting Intel. Broadcom is a major builder of OpenAI Project Stargate (the large-scale data center project in collaboration with OpenAI and SoftBank), so shorting Broadcom essentially means shorting OpenAI and Stargate. Corning is a fiber glass company, and he has also established a significant short position there. So overall, the $8 billion short position is equivalent to 40 times his fund's total assets 18 months ago. This is an extremely aggressive bet.

EJ: Very aggressive. It’s important to note that his entire fund's thesis is built upon the 64-page paper "Situational Awareness," with the core bet being that semiconductor compute FLOPs (floating-point operations) will grow across multiple orders of magnitude over the next decade. His current $8 billion short is, in fact, a bet against that thesis. So, this can only imply one of two things: either he believes the current trade is overcrowded and will face volatility and downward pressure in the short term; or he made a mistake in one part of his core thesis, though he hasn't publicly stated which one.

Bullish Side: Neocloud, Power, and Bitcoin Mining Companies

EJ: He is not completely bearish, however. If you look at the bullish positions on the right side of the chart, he still holds substantial stock positions in many types of companies while also buying some call options. Let’s look at CoreWeave; he is maintaining his position. CoreWeave has been one of his largest data center or Neocloud investments since the fund's inception. He has bet on CoreWeave in various ways, including through private investment or acquiring Core Scientific (a company that helps CoreWeave operate in bitcoin mining/data centers). Neoclouds can be simply described as cloud service providers that procure and build GPU clusters to rent them to the largest AI laboratories, and CoreWeave has signed multi-billion dollar contracts with Meta, Anthropic, and others.

Next is Bloom Energy, which is his largest new position from the last quarter. Bloom Energy primarily produces portable gas turbines that can be airlifted next to any data center to power it. One of the biggest bottlenecks for AI data centers right now is that there are a bunch of GPUs installed, but the power grid can't feed them adequately, so you need this supplementary energy solution. Leopold did not fully liquidate his position but reduced it by $1 billion. I can understand this; his position increased from about $800 million to around $2.5 billion in three months, so it makes sense to take some money off the table, and he still holds just over $1 billion in Bloom Energy.

Furthermore, he has increased positions in several bitcoin mining companies, including CleanSpark, Riot Platforms, Applied Digital, and IREN. If these names sound familiar, it's because they are in the same Neocloud lane as CoreWeave. So he is all in on data centers and Neocloud. He has observed that both Anthropic and OpenAI are continuously releasing new models and that the scaling law for computing power continues to expand, meaning GPUs are still needed, but the bottleneck right now is "delivery." These companies have solved the delivery issue, while the GPU manufacturers themselves (like Nvidia and Broadcom) are the ones he chose to short.

Josh: This is a new narrative trade forming, where funds are shifting from semiconductors themselves to infrastructure, power, data centers, and memory. He has doubled down on the direction we saw last quarter while simultaneously establishing shorts on companies he believes will not outperform.

It’s worth noting that a 13F is a snapshot; it reflects trades from the previous quarter, covering the period from January 1 to March 31. Leopold is almost always right; his fund size has grown from $220 million to the current $13.7 billion. But there are places he might have misjudged, like his short on AMD, which has risen 74% in the past month; he might have picked one of the most expensive moments in the rotation within the AI sector to short. Is this a timing issue or a thesis issue? Also, ASML is, to my knowledge, still the only company in the world capable of making lithography machines, holding a 100% monopoly, yet he has shorted it. So his thesis clearly tilts toward memory, power, and infrastructure rather than semiconductors themselves.

Dissecting the Thesis: What Exactly is Leopold Betting On?

EJ: Rather than looking separately at the bullish and bearish positions, it may be more effective to directly discuss the thesis behind each position. These positions are very aggressive; an $8 billion short is not a small number, and many of the bullish Neocloud and power companies are not well known. How good are they? In my view, he is engaging in a directional opposing trade, shorting silicon while going long on electricity. He thinks the GPU designers like Nvidia and Broadcom and manufacturing companies like TSMC are overcrowded trades. I don’t believe he is genuinely "bearish" on them; I think he believes their current valuations are too high. Conversely, he is heavily invested in power because he understands better than anyone where the next bottleneck in data centers and GPUs lies, which he identifies as power and energy. He does not believe there is enough energy or sufficient means to deliver power effectively to GPUs.

He is also heavily betting on memory. SanDisk increased about 40,000% over the past year, which should theoretically be the most crowded trade, but he remains optimistic. SanDisk’s core product is NAND flash memory, which AI models need for temporary storage to recall previous context, and this is exactly the type of storage that SanDisk provides. Therefore, I don’t think he is deeply bearish on GPUs; he just feels they are overcrowded in the short term, and money spent on power and memory yields higher returns.

Josh: This makes me start to suspect whether he is bearish on the entire market. This is the first time in the fund's history that the size of shorts exceeds the size of longs, which is a very clear shift for a fund that has always been "long only, only upward." Initially, I considered whether this is just hedging; he has made too much money in the past and wants to lock in profits and protect against downside. However, if it were simply hedging, the position sizes should clearly be smaller than longs to "offset" rather than representing such a directional bet. Last quarter he did have some hedging positions, but the ratio was small and not directional. This quarter, the total size of puts has exceeded longs, indicating a directional bet on a market decline.

So he finds himself in a very strange position; he seems to think the AI market overall will decline, but even so, memory, infrastructure, and energy will continue to rise. This is his bet.

EJ: What you mention is uncertainty. He himself may not be entirely clear on the outcome. This can be seen from a detail: he has paired some puts with calls. This structure in hedge funds is called a collar trade. If you are unsure whether the market will rise or fall, you hedge on both sides and profit from the difference in premiums between the two sides. He has employed this strategy on four companies, with the largest being on Micron. If he is genuinely bullish on memory players like SanDisk, then theoretically he should not be shorting Micron, as Micron is the largest memory representative in the US, and Leopold is dedicated to US stocks—his most profitable trades are long on Intel, Bloom Energy, and Nvidia, which are all US stocks. Thus, the short position on Micron seems less driven by a thesis and more like a "parity trade," as he does not know the direction, so he hedged it. He believes the market is overcrowded, yet remains optimistic in the long run. This is actually a smart approach.

Four Core Conclusions

Josh: I have condensed this new thesis into four points. First, the bottleneck has shifted from chips to electronics. We all know that there are enough chips; the problem is there is nowhere to plug them in. Look at the recent collaboration announced between SpaceX and Anthropic, where Anthropic is so desperate for compute power that they are willing to collaborate with competitors to get it. This isn't a shortage of chips; it's a lack of infrastructure to make these chips run at scale.

Second, chip valuations are priced based on a world that no longer exists. The SMH has risen 66% year-to-date, while Intel has risen 200%. The entire market is pricing the sector based on the logic that "every semiconductor company equally benefits from AI demand," but Leopold is betting against that, believing there are winners and losers, and that the early winners will continue to win—he intends to capture that return.

EJ: I just thought of something while looking at those bullish positions. These Neocloud companies can benefit from the entire premise of his new portfolio. When the semiconductor sector drops, their stock prices would theoretically also drop because they hold GPUs. However, CoreWeave and such companies have something that Nvidia does not: access to power capacity. He invests in these Neoclouds not because they can run GPUs—any capitalized data center can do that—but because they have access permits and capacity for the existing grid infrastructure. Therefore, he expresses both a pro-power and anti-semiconductor thesis through one company, effectively covering both bets.

Josh: Third, he has buried an Easter egg in the matter of "where can power be accessed," namely bitcoin mining companies. We briefly mentioned this last quarter, and this time he has continued to increase positions here. This year, American bitcoin miners are expected to come online with approximately 30 GW of power capacity; for comparison, this approximately equals the total power plans disclosed by Microsoft, Google, Amazon, and Meta for data centers. They already have significant key infrastructure, such as power, facilities, and scaled construction—they just need to swap out their mining machines for AI acceleration cards. This is a perspective I haven’t seen many people discuss, the pivot from bitcoin to AI, simply driven by "where the money is, I will go."

EJ: Finally, he is doubling down on "physical infrastructure." He does not think this layer will be commoditized, but he believes the "design layer" of semiconductors is overcrowded. Just to remind, Nvidia itself does not make chips; they are a design company that sends blueprints to TSMC for manufacturing. Similarly, Broadcom, Intel, and AMD all design CPU/GPUs; they are all being shorted by Leopold. They design chips but do not manufacture them. Intel and AMD's plans are to manufacture on their own, but currently lack the corresponding factories and infrastructure. Thus, his logic is that the chip design space is overcrowded, while the hardware infrastructure layer is where the money flows, and the foundational layer here is power.

Where It Might Go Wrong: Nvidia's Moat

Josh: We should also discuss where this trade might collapse. As mentioned earlier, AMD rose 74% in a month while he shorted it, which undoubtedly contradicts his position. He has a total short exposure of about $1.9 billion on Nvidia, and one possible downfall is that Nvidia's moat may be stronger than he thinks. He bets Nvidia will be "commoditized," believing that Google's TPU, Amazon's Trainium, and other custom chips will gradually erode Nvidia's monopoly, but that reality may not be the case. Looking at purchase orders and an 80% gross margin, orders continue to flock to Nvidia. The reason behind is CUDA, a highly customized and complex software stack. Those who have built infrastructure in this ecosystem do not want to migrate again, as customizing a full set of infrastructure for new chips is complicated. Whether this is true or whether Leopold is wrong, we do not know.

Anthropic is adopting a "less strongly locked" approach, collaborating with Amazon on Trainium and with Google on TPU, while also using Nvidia. However, xAI's Colossus data center is almost entirely filled with Nvidia GPUs, directly utilizing the latest Blackwell architecture (Nvidia's latest AI acceleration architecture) and fully betting on CUDA. So there may be one thesis that outperforms and another that underperforms. Regardless, Nvidia is the largest company by market capitalization, and its collapse is not a minor issue.

Even more astonishing: Nvidia GPUs from 6 to 8 years ago are now renting for higher prices than they were two years ago, and contracts need to be signed a year in advance, which means some are willing to rent old GPUs for higher prices than they were when new.

EJ: Leopold's style reminds me of Michael Burry; we discussed his high-profile short on Nvidia near its price peak a few months ago, which resulted in severe losses. I hope Leopold does not repeat that mistake. Additionally, let's point out a few potential risks or blind spots. The Situational Awareness Fund is a hedge fund, not a VC fund; it is rare for hedge funds to be as aggressive as "long only AI." All the 13F positions we discussed today are quarterly snapshots; he must report every three months. At the very moment we are speaking, it is entirely possible that he has flipped some of these positions.

Another question is, when did he build these puts? It’s likely it was in early January, when the fund may have faced setbacks. Of course, the counterexample is obvious; his fund size has grown from $5.5 billion to $14 billion in three months, which means he has made money. The key point is that these puts and calls are leveraged; behind the $8 billion nominal exposure, he may have actually invested only about $1 billion, along with paying premiums and fees, which makes this a short-term trade.

Thus, I must emphasize, he may have already exited some of these trades. If you watch this episode and think, "Oh no, I need to completely change my portfolio," please remember that your trading style is not the same as his; you are not doing short-term and high-frequency trades; you are a long-term holder, which is a completely different paradigm.

What Should Retail Investors Do: Polymarket Data and Personal Views

Josh: There are some data on Polymarket that support the idea that things are not as bad as they seem, as retail investors and Leopold are not playing the same game. If you believe the AI bubble is about to burst, this is a hint from this 13F. However, according to Polymarket, the probability of the AI bubble bursting by December 31 of this year is only 24%. I also looked at another market; according to Polymarket, the probability of Nvidia maintaining its status as the world's largest market cap company in the remaining weeks of this month is still 93%. This is evidence that suggests the volatility may not be as dramatic as implied by this 13F. Again, this is news from the last quarter; the tide has changed, and we do not know how he has traded in the past few months. But indeed, things are not that bad; he has simply switched strategies. EJ, how would you adjust as a retail investor?

EJ: Two answers to this. If you are a novice who just finished watching Leopold's 13F and want to trade based on that, please be conservative. This is not a time to bet on a single stock; I have never recommended doing that. Leopold is cautious for a reason; the market has seen average increases of several hundred percentage points over the past two years, which is extraordinarily high for a typical stock market. The rise in the S&P 500 has mainly come from the Mag 7, and their money has trickled down to all the companies we just discussed. He may simply be indicating that this is an overcrowded trade, please exercise caution.

But Josh, I always have a bullish limit; right now, my most bullish view is on power and energy. I agree with Leopold on Bloom Energy and data centers. I learned something particularly exciting from this research, which is that investing in leading Neoclouds simultaneously expresses the dual views of pro-power and anti-semiconductor; they have signed multi-billion dollar contracts with Anthropic and Meta, and even if semiconductors drop, they still hold power capacity. This is a trade I might bet on.

However, I have reservations about his short position on Corning and other fiber bottlenecks; Nvidia just signed a multi-billion dollar collaboration with Corning, and yet he shorts it. He is selecting which bottlenecks are worth betting on, I agree with his choice of power but am uncertain about his judgment on fiber. Josh, what do you think?

Josh: For me personally, the two most important aspects of AI investment are energy and "movement of atoms in the physical world." The physical world is hard, complex, and much slower than the software world. Any company with a near-monopoly in manufacturing, building, and obtaining grid access permits possesses enormous structural advantages.

The second point is energy; everyone is short on power, but no one wants to be the "bad guy"—building data centers next to cities and taking electricity away from ordinary households, driving up prices. Everyone wants cheap, easy, fast, and efficient energy production, along with abundant energy supplies. Any company that has near-monopoly advantages in these two areas is worth investing in because it is durable.

As for the chip layer, competition is fierce. Amazon's Trainium, Google's TPU, Cerebras (which just IPO'd last week with a brand-new architecture)… competition at this layer could compress profit margins. Although they remain very high now, there is room for them to be pushed back down.

The key checkpoints to watch next include Nvidia's earnings report on May 20; if their guidance for the next quarter exceeds $78 billion, this batch of puts will likely be adversely affected; AMD's Analyst Day in 2026; and key milestones in Bloom Energy's strategic deployment. These are all benchmarks that can be used to verify Leopold's positions. But in thematic terms, there will be no deviation from energy and infrastructure.

Conclusion: Think, Compare, and Do Not Follow Blindly

EJ: About a week and a half ago, we did an episode discussing how future AI investments would flow. We walked through the AI infrastructure stack from top to bottom: model laboratories, hyperscale cloud service providers (referring to companies like Mag 7 that build their own large-scale clouds), AI platforms, GPU/semiconductors. The conclusion is that money will flow from GPU/semiconductor sectors like Nvidia, AMD, and Broadcom down to memory and storage layers, power and infrastructure layers. And this matches the direction of Leopold’s 13F with both shorts and longs. We might have captured this ahead of time.

Importantly, AI is not a "one-to-one" trade. Of course, you can buy and hold Nvidia for the long term; in terms of direction, that may not be wrong over the next decade, but if you think that just putting money in one sector will secure you, then you are very mistaken. Money flows throughout the entire industry chain, and AI is like a car; it inhales fuel (capital) and consumes it across the entire infrastructure, ultimately expelling exhaust from the other end. We may currently be two-thirds along this supply chain.

This is not just speculation; there are factual data to back it up. Memory prices have increased by an average of 300% to 500% among all major manufacturers over the past nine months; their production capacity is essentially booked through to the end of 2027, almost a year and a half of orders are filled. Whether more supply will come up in the future and whether power can materialize out of nowhere is uncertain, but the directional bets Leopold made this time align with our checklist.

Josh: We will continue to track Cerebras, Leopold’s 13F, and the upcoming earnings reports; there’s a lot of news. Some memes are also quite funny, Nick Carter depicted Leopold as "I don't want to play with you anymore"—he is abandoning the AI industry. Another meme reads "The last glance before Intel investors panic sell"—he went long on Intel for tens of billions before turning around and hitting the sell button: "I don’t want it anymore."

EJ, before you leave, what do you want everyone to do? How should they adjust their portfolios based on Leopold?

EJ: My final feeling is that I am Leopold’s number one fan, but I do feel he might be wrong in certain areas. I’d like to ask the comments section to share which parts of his thesis you disagree with and why? I’m not speaking for Leopold, but I have my own reservations; I’m not sure he is fully aware of what he is doing. In fact, from the bilateral structure he has laid out this time, it shows he might also be uncertain; he is playing it safe.

Josh: If you had to pick just one thing he might be wrong about, what would it be?

EJ: Nvidia. I guess you would say the same?

Josh: Yes. If Nvidia drops, all stocks drop. That’s how I see it.

Shorting Nvidia at $1.9 billion has me a bit perplexed. With such high margins, everyone wants Blackwell. We have just received the first Blackwell model, called Mythos, and there is enormous value in Nvidia’s infrastructure stack and software. It is unidirectionally up and is the world’s largest company by market capitalization. Not continuing to bet on winners sounds like a loser's strategy. But then again, we will keep tracking this. We will provide updates and share the front-line developments of AI investment.

Thank you for watching. If you enjoyed this episode, please share it with friends, leave a comment, like it, and give us five stars. That’s it for this episode. None of this constitutes investment advice.