Leopold Aschenbrenner Portfolio Update May 2026: $8.5B in AI-Chip Puts Decoded

Leopold Aschenbrenner spent 2024 publishing a widely-read essay arguing AI superintelligence will arrive between 2027 and 2030. In May 2026 his fund filed a 13F disclosing $8.5 billion in put options against the very companies building it.

The disclosure landed on May 15, 2026 for the quarter ended March 31. The fund is Situational Awareness LP. The bearish positions cluster on NVIDIA, AMD, ASML, Broadcom, TSMC, Micron, Intel, Oracle, plus the VanEck Semiconductor ETF, totaling roughly 62% of the filing’s $13.68 billion reported value.

Key facts at a glance

• Filer: Situational Awareness LP (Leopold Aschenbrenner’s fund)
• Filing date: May 15, 2026
• Reporting period: Q1 2026 (quarter ended March 31, 2026)
• Total 13F value: $13.68 billion
• Information-table entries: 42
• AI-chip put options (nominal): $8.46 billion (~62% of filing)
• Largest put line: VanEck Semiconductor ETF (SMH) at $2.04 billion
• Largest single-name put: NVIDIA at $1.57 billion
• Source: SEC EDGAR Form 13F-HR

Who is Leopold Aschenbrenner

Leopold Aschenbrenner is a former OpenAI researcher who left the company in 2024 and founded Situational Awareness LP, an investment fund anchored to his thesis that artificial general intelligence and downstream “superintelligence” will arrive between 2027 and 2030. His June 2024 essay Situational Awareness: The Decade Ahead laid out the case. The fund’s positioning is read by markets as a barometer of how seriously a credentialed AI insider takes the equity-market pricing of AI exposure.

Two things make this 13F unusually informative. First, the fund is small and concentrated, so disclosed positions reflect real conviction rather than indexing. Second, the manager built his public reputation on calling AI progress correctly, so a bearish signal on AI-chip equities specifically (rather than AI demand broadly) carries weight.

What the Q1 2026 13F actually shows

The filing discloses 42 information-table entries. The bearish portion is the put-option overlay. Form 13F reports option holdings in terms of the underlying security count and dollar value, not the option premium paid, the strike, the expiry, or any paired long positions. So the disclosure is best read as evidence of a large bearish or hedging option overlay against parts of the AI-chip complex, not proof of a naked short book.

The put positions:

Three things stand out about the composition.

The SMH ETF put is the single largest line at $2.04 billion. That is a broad-basket bet, not a stock-picker’s view. The fund is wagering against the entire first-order chip complex, not just a handful of names.

The Oracle position is the only enterprise-software-and-cloud name in the put core, at $1.07 billion. Aschenbrenner does not put against Microsoft, Amazon, Google, or Meta. He puts against the one hyperscaler-adjacent company that bet hardest on becoming an AI infrastructure vendor. The implicit claim is that Oracle’s AI cloud story is over-promised.

The Micron line of $584 million is significant on its own. Memory has a cycle independent of accelerator demand, and Micron is the U.S. high-bandwidth-memory supplier. A separate bet on memory specifically suggests the manager thinks HBM pricing enthusiasm has run ahead of the credit cycle reset.

The Q4-to-Q1 evolution: a barbell, not a binary short

The Q4 2025 disclosure for the same fund looked very different. It read like an AI infrastructure long book: Bloom Energy, CoreWeave calls, Intel calls, Lumentum, Core Scientific, IREN, Cipher Mining, SanDisk, EQT. Power. Neoclouds. Storage. The names you would expect from someone who believes AI demand is real and durable.

Q1 2026 keeps that infrastructure long side roughly intact and adds the $8.5 billion bearish chip overlay on top. The pairing matters more than either side alone. The barbell reads as: AI demand keeps compounding, the build-out is real, but equity multiples on the first-order chip winners have run ahead of what 2027 and 2028 capex returns will support.

Read this way, the fund is not betting against artificial intelligence. It is betting that the public-equity market is pricing AI compute too narrowly through the same five or six accelerator names, while the actual second-order beneficiaries (power, grid, data-center infrastructure, networking, memory) are still re-rating.

Three industry camps interpreting the signal

Wall Street’s reaction to the filing splits roughly three ways. The disagreement is not whether AI demand is real. The disagreement is whether equity multiples correctly price the risks.

The consensus camp owns the tape. Hyperscaler capex guidance keeps revising up. NVIDIA’s fiscal-2026 results came in strong. Sovereign-AI deals add new demand. The Deloitte and McKinsey institutional outlooks for 2026 are positive on both semiconductors and data center build-out.

The skeptic camp is intellectually live but not dominant. Aschenbrenner’s put book is the most credentialed expression of it as of May 2026.

The bear camp, anchored by Michael Burry’s depreciation argument and Dan Niles’ macro-revision call, sits at the tail. Markets give it less weight, but its catalyst (a hyperscaler capex revision or a depreciation-policy disclosure shock) is the kind of event that triggers regime changes.

What this means for an AI-overweight portfolio

The wrong reaction to the filing is to mirror the disclosed puts. Three reasons.

First, a 13F is stale by roughly 45 days when it becomes public. By the time markets see the put book, the fund may have rolled, hedged, or closed parts of it. Copying disclosed positions is reading yesterday’s bet.

Second, the 13F shows the bearish side. It does not show paired longs, delta-hedging, short-vol structures, or anything else that might change the net economic exposure. The disclosed nominal is not the same as the disclosed conviction.

Third, shorting high-quality monopoly and oligopoly businesses is expensive and historically wrong over long horizons. Anyone who shorted NVIDIA at any point between 2022 and 2025 ended the period poorer for it. Even if the skeptic camp turns out right, the timing risk on the short side is asymmetric.

The right reaction is to use the signal as a stress test on portfolio construction, not as a trade instruction. Three concrete checks.

Concentration check. What fraction of the portfolio sits in first-order chip names (NVDA, AMD, AVGO, INTC, MU, ASML, TSM, ARM, plus the SOXX and SMH ETFs)? Above 15% of total assets is meaningful exposure to the exact scenario the skeptic camp is pricing. Above 25% is the level at which a 30% drawdown in the basket would deliver a 7-8% household-level hit, which matters even for growth-tolerant mandates.

Single-name check. Any single position above 10% of the portfolio is a prudent-limit breach regardless of camp. The skeptic camp does not have to be right for single-name concentration to cause asymmetric pain. The framework’s single-name flag is independent of the directional debate.

Tax-aware preparation. For households where trim is the right answer if the skeptic camp gains ground, the lead-time blocker is cost-basis data. Pulling tax lots from custodians is cheap insurance that does not commit you to a trade. It just makes the trade implementable when the moment comes.

What the 13F does not tell you

Five things the filing genuinely cannot answer, no matter how carefully you read it.

Net exposure. The 13F discloses long positions plus certain held options. Short equity positions are not disclosed. So a fund running paired pair-trades (long second-order, short first-order) looks identical on disclosure to a fund running an outright directional bet.

Option economics. The 13F reports option value in terms of underlying-security notional. It does not show strike, expiry, premium paid, delta, gamma, or whether the position is part of a spread structure. The disclosed $1.57 billion of NVIDIA puts could be deep out-of-the-money lottery tickets costing single-digit millions, or could be near-the-money positions with very different risk-reward.

Timing of acquisition. The filing is a quarter-end snapshot. Positions could have been opened on the first or last day of the quarter. Aggregate cost basis is not disclosed.

Subsequent activity. Between March 31, 2026 and May 15, 2026, the fund could have rolled, closed, or expanded any of these positions. Forty-five days is a long time in option markets, especially through earnings.

Conviction level. The 13F does not distinguish a sized-for-conviction core position from a risk-managed hedge. A $1 billion put line could be either.

Frequently asked questions

What did Leopold Aschenbrenner disclose in his May 2026 13F?

Situational Awareness LP disclosed $8.46 billion in AI-chip put options on May 15, 2026, for the quarter ended March 31, 2026. The puts cover NVIDIA, AMD, Broadcom, ASML, TSMC, Micron, Intel, Oracle, the VanEck Semiconductor ETF, plus smaller positions in Corning and Infosys. Total reported 13F value was $13.68 billion, so the bearish chip exposure represents about 62% of the filing.

Is Leopold Aschenbrenner short NVIDIA?

The filing discloses put options on NVIDIA with a reported nominal value of $1.568 billion against 8.992 million underlying shares. A 13F does not disclose short equity positions, only certain option holdings. The puts could be standalone bearish positions, paired hedges against undisclosed longs, or part of a structured trade. The disclosure shows bearish optionality on NVIDIA, not necessarily a directional naked short.

What is the largest position in the May 2026 13F?

The single largest line is a put option on the VanEck Semiconductor ETF (ticker SMH) with a reported value of $2.043 billion. This is a broad-basket bet against the entire first-order chip complex rather than a single-name view.

Why is Oracle in the put book but not other hyperscalers?

The Oracle put is $1.073 billion, the third-largest line in the filing. Microsoft, Amazon, Google parent Alphabet, and Meta are not in the put book. The implicit interpretation is that Aschenbrenner views Oracle’s specific AI-cloud positioning (Oracle Cloud Infrastructure as a competitive hyperscaler) as over-promised, while accepting the broader hyperscaler capex thesis on the others.

Does this mean Leopold is bearish on AI?

The pairing of a constructive Q4 2025 long book on AI infrastructure (power, neoclouds, fuel cells, uranium) with a Q1 2026 bearish overlay on first-order chip names suggests Aschenbrenner believes AI demand is real and durable, but that public-equity multiples on the chip leaders have run ahead of what 2027-2028 capex returns will support. This is a multiple-compression thesis, not a demand-collapse thesis.

Should I sell my NVIDIA position based on this filing?

No single hedge-fund disclosure should drive an individual trade decision. Three reasons: the 13F is stale by 45 days, the disclosure does not show paired longs that may offset the put book, and shorting high-quality monopoly businesses has been expensive over long horizons. The appropriate response is to check whether your overall AI-chip exposure is consistent with your mandate and time horizon, not to copy a disclosed position.

When is the next Situational Awareness 13F due?

The Q2 2026 13F covering the quarter ended June 30, 2026 is due 45 days after quarter end, so around August 15, 2026. That filing will show whether the put book has been rolled forward, expanded, or closed.

Sources and methodology

• Leopold Aschenbrenner, Situational Awareness: The Decade Ahead (June 2024 essay)
• SEC EDGAR, Form 13F-HR filing index, Situational Awareness LP, Q1 2026
• SEC EDGAR, Q1 2026 information table XML
• SEC EDGAR, Q4 2025 information table XML (for the Q4-vs-Q1 evolution)
• SEC, Form 13F FAQ (what 13F does and does not disclose)
• Deloitte, 2026 Global Semiconductor Industry Outlook
• McKinsey, The $7 trillion data center build-out
• NVIDIA, FY2026 financial results