Author: Techub News Compilation

Recently, the latest holding document (13F) of Leopold Aschenbrenner, the highly watched young AI investor on Wall Street, has caused a huge sensation. This 24-year-old, who turned $250 million into approximately $14 billion (nominal value) in less than two years through his "Situational Awareness Fund," has altered his long-standing image as a "major bull" and established an unprecedented short position portfolio. The Bankless channel offers an in-depth interpretation, revealing this "god-level" trader's latest judgments and bets on the future direction of the AI market.

$8 Billion Short Turn: From Chips to Infrastructure

Leopold's latest holding report indicates that his fund has established ashort position totaling approximately $8 billion, which is unprecedented in the fund's history. The size of the short side has even surpassed the nominal value of the long side for the first time. This move is shocking because his core investment thesis was based on a 64-page "Situational Awareness" article, whose core argument was that computational power (FLOPs) would increase by orders of magnitude over the next decade, leading to a bullish stance on semiconductors. Now, this $8 billion short is essentially a significant revision or tactical deviation from his original thesis.

His main short targets focus on top companies in the semiconductor supply chain:

• NVIDIA: Established about $1.9 billion in short risk exposure (calculated based on NVIDIA's weighting in the VanEck semiconductor ETF) through put options and other methods. This is his largest single short target.
• VanEck Semiconductor ETF (SMH): Approximately $2 billion in short position. The largest holding of this ETF is NVIDIA (about 20%).
• Broadcom: This company, which provides key infrastructure for OpenAI's "Stargate" project, has also become a short target.
• AMD, Intel, Applied Materials, Corning, etc.: New short positions have emerged along the entire supply chain from chip design and manufacturing to optical glass.

It is particularly noteworthy that Intel was once one of the stocks that earned him the most profit, but has now been turned into a short. This clearly indicates that his bearish stance is not aimed at individual companies but rather issues a warning about the valuation of the entire semiconductor design segment.

Analysis suggests that Leopold's turn may be based on two points: First, he believes that semiconductor trading has become overcrowded, with valuations having already reflected an ideal future that does not exist (for example, the SMH ETF has risen 66% this year), facing short-term downward pressure and volatility; second, he may have discovered a certain "break point" in his original thesis but has not publicly stated it.

Bulls Hold Steady and Increase Positions: Power, Memory, and "New Cloud" Data Centers

While establishing a large-scale short, Leopold has not broadly turned bearish on AI. His long positions clearly point to new directions: power (energy), memory, and data center infrastructure.

He has maintained and increased investments in the following areas:

• Data Centers/"New Cloud" (Neocloud): CoreWeave, retained as a long-term core holding. This company provides GPU computing power to top AI labs like Meta and Anthropic, representing the "delivery layer" of computing power. Additionally, he has increased positions in CleanSpark, Riot Platforms, Applied Digital, etc. Many of these companies are related to the transformation of Bitcoin mining, having ready access to power and data center infrastructure, and can quickly transition to AI computing services by swapping out chips. It is estimated that U.S. Bitcoin miners will bring online approximately 30 gigawatts (GW) of interconnected power capacity this year, matching the total announced by Microsoft, Google, Amazon, and Meta.
• Power Energy: Bloom Energy still holds over $1 billion in positions (despite having taken profits of about $1 billion). The company provides portable gas turbines that can be quickly deployed to supplement data centers, directly addressing the greatest constraint faced by current AI data centers—insufficient grid capacity.
• Memory: He doubled down on SanDisk. This company specializes in NAND flash memory, providing temporary memory storage for AI models, and is critical hardware for AI inference. Memory prices have increased by an average of 300%-500% over the past nine months, and capacity is allocated until the end of 2027.

Leopold's new narrative can be summarized as: the bottleneck has shifted from chips (transistors) to electrons (power). The issue is no longer a lack of chips, but a lack of large-scale infrastructure and power to deploy these chips. Therefore, he is shorting what he believes to be overvalued and overcrowded chip design layers (e.g., NVIDIA, Broadcom, AMD) while going long on physical infrastructure layers (power, data centers, memory) that are difficult to commoditize and have structural advantages.

Bi-directional Trading and Uncertainty: Hedge or Directional Short?

Another characteristic of Leopold's holdings is the bi-directional trade. For instance, he has simultaneously established both put and call options on companies like Micron, constituting what is known as a "collar trade." This indicates that in certain areas, he is not purely bearish or bullish, but rather acknowledging the uncertainty of market direction and earning premiums or hedging risks through options combinations.

This raises a core question: is he turning bearish on the overall market, or is he hedging for massive bullish profits? Given that his short scale has exceeded his long positions, this appears to be more of a directional bearish bet on the market. He may anticipate the AI market (especially the semiconductor sector) will decline, while at the same time, sectors such as power and memory will continue to rise. This is a sort of "pseudo-hedge," betting against most of the market while betting on specific bottleneck segments.

Analysis points out that this strategy cleverly leverages “new cloud” companies like CoreWeave. These companies possess GPUs (thus if semiconductor stocks decline, their stock prices could also be under pressure), but more importantly, they have the permission and capability to access existing grid infrastructure. Therefore, regardless of how chip stocks perform, as long as power demand persists, these companies can benefit from Leopold's favorable view on "power trades."

Potential Risks and Market Discrepancies

Leopold's gamble is not without risks. The market has shown some signs contrary to his judgments:

• Timing Risk: AMD, which he shorted, rose by 74% within a month of his establishing a position. This implies that he may have shorted at a very expensive time.
• NVIDIA’s "Moat": This represents the main challenge facing his largest short position. NVIDIA not only has hardware advantages but crucially, its CUDA software ecosystem has formed a strong platform lock-in effect. Many AI companies (such as xAI) have their entire data centers built on NVIDIA GPUs, making the migration cost extremely high. Meanwhile, NVIDIA's profit margin remains around 80%, with strong demand (even old H100 chip rental prices are now higher than two years ago). The upcoming earnings report on May 28, if it shows robust guidance, could put pressure on his short position.
• Lag in Holding Information: The 13F document reflects holdings as of March 31. Over the past month, he may have adjusted these positions. In fact, his fund's net value has increased from $5.5 billion three months ago to $14 billion now, indicating he is still making money overall.

Additionally, retail investors should differentiate their strategies from professional traders. Polymarket’s prediction market shows that the market does not believe an AI bubble will burst soon (with only a 24% probability of bursting by the end of this year) and there is still a 93% probability that NVIDIA will maintain the world's largest market value by the end of May. This serves as a reminder to ordinary investors to not blindly follow the trends based on a holding report, especially not to engage in high-frequency short-term trading imitation.

Insights for Investors: Focus on the Physical World and Energy

Setting aside the exact positions of Leopold to follow closely, this conversation reveals the key themes for the next phase of AI investments:

The extreme importance of energy and physical infrastructure. The software world iterates rapidly, but the construction of the physical world—manufacturing, construction, energy acquisition, and building permits—is slow and complex. Any company with unique advantages or quasi-monopolistic positions in energy acquisition or physical manufacturing could have enduring structural advantages.

In contrast, competition at the chip design level is intensifying. Besides NVIDIA, Amazon and Google are developing their own chips (like TPU and Trainium), and companies such as Cerebras with new architectures are continuously emerging. Although profit margins remain high, competition may gradually erode some profit margins.

For ordinary investors, the strategy could be: maintain a long-term optimism toward AI but shift attention from highly valued chip design companies to "hard bottleneck" areas such as energy, data center infrastructure, and memory. These areas have more rigid demand and are less likely to be rapidly commoditized.

Leopold's shift marks a profound narrative change in AI investments: from pursuing companies that manufacture the "brain" (chips) to those that provide the "blood" (power) and "memory" (storage) for these "brains." This is not only a battle at the trading level but also a deep insight into how the industrialization of AI processes delves into the physical constraints. The upcoming earnings season (NVIDIA, AMD, etc.) and the progress of infrastructure companies will be a crucial test for this new narrative.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。