error

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

Original Video Source: Limitless Podcast
Hosts: Josh, EJ

Original Article Translation: Deep Tide TechFlow

Editor's Note

Wall Street's most well-known AI long investor, Leopold Aschenbrenner (former OpenAI researcher and founder of Situational Awareness Fund, turned $250 million into $13.7 billion in 2 years), has just filed his latest 13F with the SEC, revealing the most surprising signal of all — he has established a massive $8 billion short position across the entire semiconductor supply chain, including NVIDIA, AMD, Broadcom, ASML, and Micron, which is 40 times the size of his fund's total assets 18 months ago and the first time the fund's short exposure has exceeded its long exposure.

His new thesis can be summarized in one sentence: the bottleneck of AI investment is shifting from the chip (design layer) to power and memory (infrastructure layer).

On the long side, he continues to heavily invest in CoreWeave, Bloom Energy, and other data center and clean energy newcomers, while also adding new positions in SanDisk, CleanSpark, Riot Platforms, Applied Digital, IREN, and other Bitcoin mining companies with grid integration capabilities. This is a key document for understanding the 2026 AI investment narrative shift.

Key Quotes

From Bullish on AI to Shorting the Semiconductor Industry

· "This is the first time in the fund's history that the short position has exceeded the long position, with an $8 billion short exposure, which is 40 times the fund's net value 18 months ago. This is not hedging; this is a directional bet."

· "If it were merely to hedge realized gains, you would see small hedging positions to offset long positions on the books. But when the put size is larger than the long, that is a bet on the market going down."

Core Argument: The bottleneck has shifted from silicon to electronics

·"The bottleneck has shifted from the chip to electricity. The chip is actually sufficient; the problem is where to plug these chips in. Anthropic would rather cooperate with competitor SpaceX to obtain computing power, not because there is a shortage of chips, but because there is no corresponding infrastructure to deploy them at scale."

·"Leopold understands data centers and GPUs better than anyone in this market; no one has spent more time studying this. So he is more aware than anyone else of where the next bottleneck is, which he believes is in power and energy."

·"I don't think he is really bearish on GPUs; he just thinks that in the short term, this is an overly congested trade, and his money is better spent on power and memory for higher returns."

About Neocloud and Power: A Trade that Benefits Both Sides

·"CoreWeave-like Neoclouds have something that NVIDIA doesn't, which is the right to access power. Betting on Neoclouds is not just because they can run GPUs; any well-capitalized data center can do that. More importantly, they have access permits and capacity to the existing power grid infrastructure."

·"This is a win-win trade. Even if the semiconductor sector declines and the GPU valuation held by these companies drops, they can still benefit from the power premium because of their power capacity."

·"U.S. Bitcoin miners will add about 30 GW of power capacity this year. For reference, this is roughly equivalent to the sum of the power plans disclosed by Microsoft, Google, Amazon, and Meta for their data centers. They already have power, land, and facilities; they just need to swap out the miners for AI accelerator cards."

Memory and Infrastructure Layer

·"Memory prices are skyrocketing. In the past 9 months, the average price of major memory manufacturers has risen by 300% to 500%; if you look at capacity, almost all schedules are already booked until the end of 2027."

·"SanDisk has surged by about 40,000% in the past year. In theory, this should be one of the most crowded trades, but Leopold remains bullish. This is because SanDisk's core product is NAND flash memory, crucial for AI model memory and context recall."

NVIDIA: The Biggest Short, but Possibly the Biggest Miscalculation

·"Leopold has approximately a $1.9 billion short position on NVIDIA (including an indirect short on the VanEck Semiconductor ETF SMH, where NVIDIA has the highest weight, approximately 20%)."

· "NVIDIA's moat may be stronger than imagined. CUDA is a software lock-in platform, and once you've built on it, you don't want to leave because reconstructing custom infrastructure for new chips is very complex."

· "NVIDIA's old GPUs from 6 to 8 years ago now have higher second-hand rental prices than two years ago, and contracts are being signed one year in advance."

Retail-Oriented Practical Advice

· "If you are a retail investor who entered this market after seeing Leopold's 13F, please be conservative. This is not the time to go all-in on a single stock. Over the past two years, the rise of the S&P 500 has mainly come from the Mag 7. Funds have flowed all the way down to the companies we are discussing. This may be an overcrowded trade, so be cautious."

· "Two things I am personally bullish on in the long term: first is energy, as everyone is short on power; second is the manufacturing and construction capabilities in the physical world. Whoever has any near-monopoly advantage in manufacturing, building factories, and obtaining grid access permits is a target for long-term investment with a sustainable moat."

· "NVIDIA will release its earnings report on May 20. If their guidance for the next quarter is above $78 billion, then this batch of put options is highly likely to backfire."

Excerpts from the Conversation:

Ace of AI Longs Goes Short

Josh: Wall Street's most renowned AI long has just called out a 'top' for the entire AI market. Leopold Aschenbrenner, the 24-year-old former OpenAI researcher who was fired and then founded his own fund, has turned $250 million into $13.7 billion in less than 2 years in the 'Stock Market's Son' style, and he's making another move, but his latest portfolio is not what you think.

He has turned bearish on the entire stock market, established an $8 billion short position against the biggest companies in the AI industry, including NVIDIA, AMD, Broadcom, and the entire semiconductor supply chain. But it's not all doom and gloom. He also revealed where his next big AI investment is going: into power and memory. He is doubling down on data centers and three brand-new companies. We will break it down one by one, but first, let's talk about the biggest shift.

EJ: The world's most valuable company, the face of the AI revolution, and the company that has made countless investors rich, NVIDIA, is now in the crosshairs. This is Leopold's largest short position at the moment, but you wouldn't see it at first glance from the documents because the top position in his short portfolio is listed as the VanEck Semiconductor ETF (code SMH), with NVIDIA itself coming in a close second.

Currently, he has a $1.5 billion bearish put option exposure directly to NVIDIA. For those unfamiliar with put options, let me explain. A put essentially gives Leopold the right, but not the obligation, to sell the underlying asset at an agreed-upon price. What he purchased is the right to sell at a higher price if NVIDIA's stock price falls below a certain level.

Josh: SMH is his top short position, with a scale of approximately $2 billion. I looked into the holdings, and the largest single position is indeed NVIDIA, with about a 20% weight. So, combining the top two short positions, he is effectively short around $1.9 billion of NVIDIA. For those who firmly believe NVIDIA will only rise, this may come as a blow. But Leopold clearly thinks otherwise.

In addition, Broadcom, Oracle, AMD, Micron, ASML, Intel, and Corning, all of these are new short positions. Keep in mind that Intel was his claim to fame, the biggest gainer in his fund's history, and now he is shorting Intel. Broadcom is a key builder of the OpenAI Project Stargate (a large-scale data center initiative with OpenAI and SoftBank), and shorting Broadcom is essentially a bet against OpenAI and Stargate.

Corning is a fiber optic glass company, and he has also established a large short position. So, overall, there is an $8 billion short exposure, equivalent to 40 times his fund's total assets 18 months ago. This is an extremely aggressive bet.

EJ: Very aggressive. You see, his entire fund's thesis is based on that 64-page paper "Situational Awareness," and the core bet is on semiconductors, specifically on the idea that compute FLOPs (floating-point operations) will increase by multiple orders of magnitude over the next decade.

This $8 billion short position he now holds is actually a bet against that thesis. So, this can only suggest two possibilities: either he believes the current trades are overly crowded, leading to short-term volatility and downward pressure, or there is a flaw in some part of his core thesis, although he hasn't publicly disclosed which part.

Long Side: Neocloud, Power, and Bitcoin Mining Corps

EJ: He is not entirely bearish. If you look at the long side of the chart, he still holds massive stock positions in many types of companies and has also purchased some call options. Let's start with CoreWeave. He has maintained his position, and CoreWeave has always been one of his largest investments in data centers or Neocloud since the fund's inception. He has placed various bets on CoreWeave, including through private investments or acquiring Core Scientific (a company that helps CoreWeave operate Bitcoin mining/data centers).

Neocloud is essentially a new cloud provider that purchases and builds GPU clusters to lease to leading AI labs. CoreWeave has already signed multi-billion-dollar contracts with Meta, Anthropic, and others.

Next up is Bloom Energy, Leopold's largest new position last quarter. Bloom Energy primarily produces portable gas turbines that can be air-shipped directly to power any data center. One of the current major bottlenecks in AI data centers is having plenty of GPUs but not enough power from the grid to feed them, hence the need for such supplemental energy solutions.

Leopold hasn't exited this position but has trimmed it by $1 billion. This is understandable, as his position grew from around $800 million to about $2.5 billion in the past 3 months. Taking some profits off the table at the top makes sense, and he still holds slightly over $1 billion in Bloom Energy.

Further down, he increased his holdings in a batch of Bitcoin mining companies, including CleanSpark, Riot Platforms, Applied Digital, and IREN. If these names sound familiar, it's because they operate in the same Neocloud space as CoreWeave. Therefore, he is going all-in on data centers and Neocloud.

He observed that Anthropic and OpenAI continue to release new models, and the compute scaling law keeps expanding, indicating that GPUs are still in demand. However, the current bottleneck is "delivery." These companies have addressed the delivery issue, making GPU manufacturers themselves (such as NVIDIA, Broadcom, etc.) his short targets.

Josh: This is an emerging narrative trade where funds are shifting from semiconductors towards infrastructure, power, data centers, and memory. He is doubling down on the trend we saw last quarter but is simultaneously opening short positions, betting against companies he believes will underperform.

A quick reminder, 13F is a snapshot based on the previous quarter's trades, covering January 1 to March 31. Leopold has been almost always right, and his fund size has grown from $220 million to $13.7 billion on the books.

However, there could be areas where he might have missed the mark. For instance, his short on AMD, which has surged 74% in the past month. He may have picked the most expensive time in the AI sector rotation to go short. Is this a timing issue or a thesis issue? Another example is ASML; as far as I know, ASML is the only company in the world that makes photolithography machines, enjoying a 100% monopoly. Yet, he is shorting it. Hence, his thesis clearly leans towards memory, power, and infrastructure rather than semiconductors themselves.

Argument Breakdown: What is Leopold Really Betting On?

EJ: Instead of focusing on Leopold's long and short positions separately, I think it's better to discuss the underlying arguments behind each of his positions. These positions are very aggressive; an $8 billion short is not a small number, and on the long side, Neocloud and the power companies are relatively unknown to many.

Are they really promising investments? My view is that he is engaging in a directional cross-trade, shorting silicon and going long on electricity. He believes that GPU design firms like NVIDIA, Broadcom, and the manufacturing end represented by Taiwan Semiconductor Manufacturing Company (TSMC) are crowded trades. I don't think he is truly "bearish"; I think he believes their current valuations are too high.

In turn, he has a heavy allocation to electricity because he understands more than anyone else where the next bottleneck for data centers and GPUs lies, which he has identified as being in power and energy. He does not think there is currently enough power, or a good enough way to deliver electricity to GPUs.

He is also heavily invested in memory. SanDisk has risen by about 40,000% in the past year, which should theoretically be the most crowded trade, but he remains optimistic. SanDisk's core product is NAND flash memory, which is crucial for AI models to store temporary memory to recall previous interactions, precisely what SanDisk provides. So, I don't think he has a deep bearish view on GPUs themselves; he just thinks they are short-term overcrowded, and he will get a higher return by investing in electricity and memory.

Josh: This makes me start to question whether he is bearish on the entire market. This is the first time in the fund's history that the short position size has exceeded the long position size. For a fund that has always been "long-only" and "bullish," this is a very clear shift. Initially, when I was evaluating, I was thinking, maybe it's just hedging. He has made too much profit in the past and wants to lock in gains and protect against downside.

But if it's just hedging, the position size should be significantly smaller than the longs to "offset," rather than such a directional bet. Last quarter he did have some hedging positions, but the proportion was very small and not directional. This quarter, the total put size has exceeded the longs, indicating a directional bet on a market downturn.

So, he has found himself in a very peculiar position. He seems to believe that the AI market as a whole will fall, but even so, memory, infrastructure, and energy will continue to rise. This is his bet.

EJ: What you're actually talking about is uncertainty. He himself is not entirely clear about the outcome, as can be seen from one detail: he matched a batch of put positions with corresponding call positions. This kind of structure is called a collar trade in hedge funds. If you're not sure whether the market will rise or fall, you hedge both sides and make money from the price difference of the premiums on both sides.

He applied this strategy to four companies, with the largest one being Micron. If he truly believed in the growth of a memory player like SanDisk, he shouldn't have taken a bearish position on Micron since Micron is the largest memory representative in the United States. Leopold, being a pure-play US stock investor, has made the most money in his fund from long positions in Intel, Bloom Energy, and NVIDIA, among other US stock targets.

Therefore, his short position on Micron doesn't seem to be driven by a specific argument but rather seems like a "pairs trade." He was unsure about the direction, so he hedged his position. He believes the market is overly crowded, but he still sees a positive long-term outlook. This is actually a clever move.

Four Core Arguments

Josh: I've condensed this new argument into four points. First, the bottleneck has shifted from chips to electricity. We all know that we have enough chips; the problem is there's nowhere to plug them in. Look at the recent partnership announcement between SpaceX and Anthropic. Anthropic is so thirsty for computational power that they are willing to collaborate with competitors to access this power. This is not a chip shortage; it's a lack of infrastructure to scale up these chips.

Second, chip valuations are based on a world that no longer exists. SMH has risen by 66% since the beginning of the year, while Intel has surged by 200%. The entire market is pricing in the semiconductor sector using the logic of "every semiconductor company equally benefits from AI demand." However, Leopold is taking a contrarian view; he believes there are winners and losers, and the previous winners will continue to win. He plans to keep benefiting from this trend.

EJ: I was just looking at these long positions and had a thought. These Neocloud companies could benefit from the full set of arguments in his new portfolio. When the semiconductor sector declines, theoretically, their stock prices will also fall because they hold GPUs.

However, companies like CoreWeave have something that NVIDIA doesn't, which is access to electricity. He invested in these Neocloud companies not because they can run GPUs – any well-funded data center can do that – but because they have permission and capacity to access existing power grid infrastructure. So, through one company, he is expressing both a pro-electricity and anti-semiconductor thesis, killing two birds with one stone.

Josh: Third, he also dropped a hint on "where to access electricity," namely, Bitcoin mining companies.

Last quarter, we briefly mentioned this, and now he has increased his position in this area. This year, US Bitcoin miners are expected to bring online about 30 GW of power capacity. For context, this is approximately equivalent to the total data center power planning announced by Microsoft, Google, Amazon, and Meta combined. They already have a significant amount of critical infrastructure in place, such as power, data centers, and scalable buildings. They just need to swap out the mining rigs for AI accelerator cards. This is a perspective that not many have delved into – the pivot of Bitcoin to AI is purely a case of "follow the money."

EJ: Lastly, what he doubled down on was "physical infrastructure." He doesn't believe this layer will be commoditized, but he thinks the semiconductor "design layer" is overly crowded. Just a reminder, NVIDIA itself does not manufacture chips; they are a design company that sends the blueprints to TSMC in Taiwan for manufacturing. The same goes for Broadcom. Intel and AMD both do CPU/GPU design, and they were all targeted by Leopold this time, as they all design chips but do not manufacture them themselves.

Intel and AMD's plan is to manufacture on their own, but currently, they don't have the necessary fabs and infrastructure. So his logic is that the chip design space is overly crowded, and the hardware infrastructure layer is where the money flows, with electricity being the most critical substrate in this layer.

Where It Could Go Wrong: NVIDIA's Moat

Josh: Let's also discuss where this trade could go wrong. As mentioned earlier, AMD surged by 74% in a month, which he shorted, definitely a setback. His total short position in NVIDIA is about $1.9 billion, and one possible breaking point is that NVIDIA's moat is stronger than he anticipated.

He bet on NVIDIA being "commoditized," believing that Google's TPU, Amazon's Trainium, and other custom chips would slowly erode NVIDIA's monopoly. But reality may not be so, as seen from purchase orders and an 80% gross margin, orders are still flooding towards NVIDIA. Behind this is CUDA, a highly customized software stack with a high barrier to entry.

Those who have built infrastructure in this ecosystem do not want to migrate again because customizing a full set of infrastructure for new chips is very complex. Whether this is true, whether Leopold is wrong, we don't know.

Anthropic is taking a "not-so-strongly-locked-in" approach, cooperating with Amazon on Trainium, partnering with Google on TPU, and also using NVIDIA. However, Anthropic's xAI Colossus data center is almost entirely equipped with NVIDIA GPUs, directly maxing out the latest Blackwell architecture (NVIDIA's latest AI acceleration architecture) and fully betting on CUDA.

So there may be an argument where one outperforms the other. In any case, NVIDIA is the world's most valuable company, so its collapse would not be a small matter.

Even more exaggerated is that NVIDIA GPUs shipped 6 to 8 years ago are now priced higher in the second-hand leasing market than two years ago, and contracts are even signed a year in advance, meaning some are willing to rent old GPUs at a higher price than new machines from that year.

EJ: Leopold's style reminds me of Michael Burry. We talked a few months ago about his high-profile short position near the peak of Nvidia's stock price, which ended up burning him badly. I hope Leopold doesn't suffer the same fate.

Adding a few more potential risks or cognitive blind spots, the Situational Awareness Fund is a hedge fund, not a VC fund. It's rare for a hedge fund to be so aggressively "long AI" as he is. All the 13F positions we discussed today are quarterly snapshots; he must file once every three months. As of this moment today, he could have completely reversed these positions.

Another issue is when did he establish these puts? It was likely at the beginning of the year when the fund could have been vulnerable. Of course, the opposite is also true - the fund grew from $5.5 billion to $14 billion within three months, so he made money.

The key point is that these puts and calls are leveraged. With an $8 billion notional exposure, he may have only put in around $1 billion, while also having to pay premiums and fees, making it a short-term trade.

So, I must emphasize, he may have exited some of these trades. If you watch this and think, "Oh no, I need to overhaul my entire portfolio," remember, your trading style is not his. You are not day trading or high-frequency trading; you are a long-term holder, which is a completely different paradigm.

What Should Retail Investors Do: Polymarket Data and Personal Views

Josh: There is some data on Polymarket that can support the idea that things may not be as bad as they seem because retail investors are not playing the same game as Leopold. If you believe the AI bubble is about to burst, this is what this 13F implies.

However, according to Polymarket, as of December 31st, the probability of the AI bubble bursting is only 24%. I also looked at another market, and according to Polymarket, the probability that Nvidia will remain the world's most valuable company in the few weeks remaining this month is still 93%. This is evidence that the volatility may not be as severe as implied by his 13F. Again, this is last quarter's news; the tide has turned, and we don't know how he's been trading these past months. But indeed, things are not that bad; he has simply changed his strategy. EJ, as a retail investor, how would you adjust?

EJ: Two answers. If you are a beginner and are thinking of trading based on Leopold's 13F, please be cautious.

This is not the time to bet on a single stock, and I have never advised doing so. Leopold is cautious for a reason. In the past two years, the market has risen by several hundred percentage points on average, which is a huge increase for a normal stock market. The rise of the S&P 500 is mainly driven by the Mag 7, whose money has flowed down through AI to all the companies we just discussed. He may just be saying that this is an overcrowded trade, so please be cautious.

But Josh, I always have a long limit, and right now, what I am most bullish on is power and energy. Leopold and I agree on the line of Bloom Energy and data centers.

One thing I learned from this research that got me particularly excited is that the top Neocloud investment simultaneously expressed a dual view of power + air semiconductor. They, along with Anthropic and Meta, have signed contracts worth billions of dollars, and even if semiconductors fall, they still have power capacity. This is a trade I might place a bet on.

However, I have reservations about his shorting of Corning and other fiber optic bottlenecks. NVIDIA just signed a multi-billion-dollar cooperation deal with Corning, but he is shorting it. He is picking which bottleneck is worth betting on, and I agree with his choice of power, but I'm not sure if he's right about fiber optics. Josh, what do you think?

Josh: For me personally, the two most important things in AI investments are energy and the "movement of atoms in the physical world." The physical world is hard, complex, and far slower than the software world. Any company with a near-monopoly advantage in manufacturing, construction, and obtaining grid access permits has a huge structural advantage.

Second is energy. Everyone lacks electricity, but no one wants to be the "bad guy" — building data centers next to cities, taking electricity from ordinary households, raising electricity prices. Everyone wants cheap, easy, fast, efficient electricity production and an abundant energy supply. Any company with a near-monopoly in these two areas is worth investing in because it is enduring.

As for the chip layer, the 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 are still very high at the moment, there is room for them to be squeezed back.

The next few key checkpoints to watch are NVIDIA's earnings report on May 20. If the guidance for the next quarter is above $78 billion, these put options will likely be hit hard; AMD's Analyst Day in 2026; and Bloom Energy's key deployment milestones. These are checkpoints that can be used to cross-check Leopold's positions. However, at the thematic level, energy and infrastructure are the directions that will not deviate.

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. NVIDIA goes down, all stocks go down. That's how I see it.

A $1.9 billion short on NVIDIA, I'm a bit puzzled. With such a high gross margin, everyone is looking for Blackwells. We've just received the earliest Blackwell model, the first one called Mythos. NVIDIA's infrastructure stack and software hold immense value.

It's unilaterally upward, it's the world's most valuable company. Not continuing to bet on the winner sounds like a loser's strategy. But then again, we'll keep tracking. We'll stay updated and share with everyone the latest developments at the forefront of AI investment every day.

Original Article Link

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia