The investment narrative is shifting from AI chip designers to the infrastructure that powers them, with a massive $1.9 billion bearish bet against NVIDIA (NVDA) and a $2 billion short on the SMH ETF. Investors should consider rotating out of overcrowded semiconductor names like AMD, Broadcom (AVGO), and ASML in favor of energy and power access plays. Bloom Energy (BE) remains a high-conviction holding for its ability to provide off-grid power to data centers, while Applied Digital (APLD) and Iron Mountain (IRM) offer direct exposure to specialized AI cloud infrastructure. Bitcoin miners such as CleanSpark (CLSK) and Riot Platforms (RIOT) are emerging as "pivot plays" because they already own the power permits and data center space required for AI scaling. Finally, look to Western Digital (WDC) and Micron (MU) to capture the next AI bottleneck in memory and NAND flash storage, where supply is reportedly booked out through 2027.

Based on the latest 13F filings and the discussion regarding Leopold Aschenbrenner’s fund (Situational Awareness Fund), here are the investment insights extracted from the transcript.

NVIDIA (NVDA)

The podcast highlights a massive shift in sentiment from one of the most prominent AI bulls. Despite being the "poster child" of the AI revolution, NVIDIA is now the primary target of a significant short position.

• Short Exposure: Leopold has approximately $1.5 billion in short exposure via put options.
• Indirect Short: When combined with his short position in the SMH ETF (which holds 20% NVIDIA), his total bearish bet against the company reaches roughly $1.9 billion.
• The Thesis: The "design layer" of semiconductors is viewed as an overcrowded trade. There is a belief that NVIDIA’s monopoly may be chipped away by custom chips from Google (TPUs) and Amazon (Trainium).

Takeaways

• Bearish Sentiment: The fund is betting on short-term volatility and downward price pressure, suggesting the stock may be overvalued at current levels.
• Risk Factor: The "CUDA moat" (NVIDIA's software stack) remains a powerful retention tool. If NVIDIA guides above $78 billion in upcoming earnings, these short positions could face significant losses.

VanEck Semiconductor ETF (SMH)

This ETF is used as a broad tool to bet against the entire semiconductor supply chain.

• Short Position: This is the fund's largest overall short position at $2 billion.
• Broad Bearishness: By shorting this, the fund is expressing a negative outlook on AMD, Broadcom, ASML, and Intel.

Takeaways

• Sector Rotation: The discussion suggests a "narrative trade" shift away from the companies that design chips toward the infrastructure that powers them.

Power & Energy Infrastructure (Various)

The core of the new bullish thesis is that the AI bottleneck has moved from "chips to electrons."

• Bloom Energy (BE): A "fan favorite" that creates portable gas turbines for data centers. Although the fund trimmed $1 billion to take profits, it maintains a $1 billion+ position.
• The "Power Access" Advantage: The fund is betting on companies that have licenses and access to the existing energy grid, which is currently a major constraint for AI deployment.

Takeaways

• Investment Theme: Look for companies solving the energy crisis. As data centers scale, the ability to provide consistent, off-grid or supplemental power is becoming more valuable than the chips themselves.

NeoCloud & Data Centers (CoreWeave, Applied Digital)

The fund is doubling down on "NeoCloud" providers—specialized cloud companies that set up GPUs for AI labs.

• CoreWeave: A major private holding maintained by the fund. They provide infrastructure for Meta and Anthropic.
• Public Equities: The fund increased positions in Applied Digital (APLD) and Iron Mountain (IRM).
• Bitcoin Miners: CleanSpark (CLSK) and Riot Platforms (RIOT) are mentioned as key plays.

Takeaways

• The "Pivot" Play: Bitcoin miners are becoming AI infrastructure plays because they already possess the power permits and data center space.
• Actionable Insight: Investors should monitor Bitcoin miners that are pivoting their infrastructure to host AI compute.

Memory & Storage (SanDisk/Western Digital, Micron)

Memory is identified as the "next constraint" for AI models that need to store temporary data during conversations (NAND flash).

• SanDisk (Western Digital): The fund "doubled down" here, betting on the high demand for NAND flash memory.
• Micron (MU): Interestingly, the fund holds both put and call options (a "collar" or "straddle" trade) on Micron, indicating uncertainty about the immediate direction but a desire to capture volatility premiums.

Takeaways

• Bullish Theme: Memory prices have reportedly increased 300-500% in some sectors over the last nine months.
• Supply Constraints: Top memory manufacturers are reportedly booked out until 2027.

Summary of Short Positions (The "Bear Book")

The fund has taken an $8 billion total short position across the following names:

• AMD
• Broadcom (AVGO)
• Oracle (ORCL)
• Micron (MU) (as a hedge)
• ASML
• Intel (INTC) (notably, a former core long position for the fund)
• Corning (GLW) (Optical glass/fiber)

Takeaways

• Contrarian Move: Shorting ASML is particularly aggressive as they hold a monopoly on lithography machines.
• The "Stargate" Bet: By shorting Broadcom, the fund is effectively betting against the success or timeline of OpenAI’s "Project Stargate" supercomputer.