Aschenbrenner just shorted $8.5B of chips: the Situational Awareness Q1 2026 13F pivot

Situational Awareness LP — Leopold Aschenbrenner's AI-supercycle hedge fund — filed its Q1 2026 13F-HR with the SEC on May 18, 2026 (accession 0002045724-26-000008, period of report 2026-03-31). The filing documents the largest single-quarter tactical pivot we've seen in any AI-thematic 13F: 42 positions, $13.68B in notional value, with $8.46B (62% of the book) expressed as puts.

The author of "Situational Awareness: The Decade Ahead" — the canonical AGI-by-2027-2028 essay that argued the entire industrial base would bend to feed AI compute — just placed the largest semiconductor short ever disclosed in an AI-thematic hedge-fund book.

Live data: /portfolios/situational-awareness.

The pivot: Q4 2025 → Q1 2026 in one chart

| | Q4 2025 | Q1 2026 | Δ | | --- | --- | --- | --- | | Total notional | $5.52B | $13.68B | +$8.16B (+148%) | | Position count | 29 | 42 | +13 | | Long cash | $5.32B (96%) | $3.86B (28%) | -$1.46B | | Calls (long delta) | $0.19B (3%) | $1.36B (10%) | +$1.17B | | Puts (short delta) | $0.01B (under 1%) | $8.46B (62%) | +$8.45B |

Q4 2025 was 100% bullish (one $9M INFY put as the only short). Q1 2026 introduces a chip-short stack larger than the entire Q4 long book.

The long side was not abandoned — Bloom Energy is still the anchor ($879M, held flat), SanDisk tripled, the nine BTC-miner-to-HPC names all survive (and most got bigger). What changed is the addition of an $8.5B short structure on top.

The $8.5B short stack — what's actually being shorted

The Q1 2026 puts, ranked by notional. Excluding the MU and TSM straddles (covered in the next section), the directional chip-short stack is ~$7.4B:

| Rank | Ticker | Position | Notional | | --- | --- | --- | --- | | 1 | $SMH | Put | $2.04B | | 2 | $NVDA | Put | $1.57B | | 3 | $ORCL | Put | $1.07B | | 4 | $AVGO | Put | $1.01B | | 5 | $AMD | Put | $969M | | 8 | $MU | Put (straddle leg) | $584M | | 10 | $TSM | Put (straddle leg) | $535M | | 11 | $ASML | Put | $494M | | 18 | $INTC | Put | $159M | | 28 | $GLW | Put | $21M | | 37 | $INFY | Put | $7M |

Read the structure, not just the line items. The single largest position in the entire book is the SMH put at $2.04B — shorting the VanEck Semiconductor ETF basket itself, the cleanest expression of "AI is real, the chips are over-priced." SMH holds NVDA, TSM, AVGO, AMD, ASML, MU, INTC, LRCX, KLAC, AMAT — every chip Aschenbrenner is also shorting individually. The SMH short is a leveraged overlay on top of the single-name short stack.

The five names short at over $900M each — NVDA, ORCL, AVGO, AMD, plus the SMH basket — are not random. They are the entire short list of "chip names that have re-rated hardest on AI capex announcements over the last 18 months." This is a pure factor-short: the AI-capex-priced-into-the-chip factor.

ORCL is the surprise. Oracle is the only non-Mag-7 enterprise-software name to repeatedly rerate on AI capex (OCI cloud + Stargate exposure + AI-database positioning). At $1.07B short, the ORCL position is the third-largest in the book and the only software short. It maps onto our Hyperscalers failed bubble — empirically, ORCL co-moves with SPY beta, not with a coherent AI-compute cluster.

Cross-reference with the QuantAbundancia bubble taxonomy:

• Semiconductors (NVDA, AMD): short $2.5B combined
• Semi Equipment (ASML, AVGO): short $1.5B
• Memory · HBM (MU straddle leg): short $584M
• Hyperscalers (ORCL): short $1.07B
• SMH basket: short $2.04B (overlay on the above)

Every validated AI-bottleneck bubble that has rerated hard is shorted at scale. Notably absent from the short list: the Datacenter Power bubble (no shorts on VRT / GEV / VST / CEG / TLNE) and the Networking · Optical bubble (no shorts on ANET / CSCO — though COHR and LITE were closed from the long side).

The straddles: MU and TSM are NOT directional shorts

Two positions look like big shorts on the put screen but are actually paired:

| Ticker | Put | Call | Net structure | | --- | --- | --- | --- | | $MU | $584M (1.73M shares) | $422M (1.25M shares) | Roughly delta-flat straddle | | $TSM | $535M (1.58M shares) | $355M (1.05M shares) | Slight put-skew straddle |

Aschenbrenner is long volatility on memory and foundry, not directionally short. The implied view: MU prints big either way as the HBM3e → HBM4 transition reprices the NAND/DRAM cycle, and TSM prints big either way as N2 / N2P pricing + geopolitical premium resolves. The market sees both as low-vol AI-supercycle compounders; the straddle says it isn't.

These are the most interesting structures in the filing. The straddle on Micron specifically rhymes with the SanDisk triple-up on the long side: if you're tripling SNDK long and straddling MU, you have a strong opinion about NAND vs. DRAM cycle timing — bullish on NAND inflection, agnostic on DRAM mark-to-market.

The picks-and-shovels longs — same thesis, bigger book

The $3.86B cash-long book reads almost identically to Q4 2025: AI infrastructure where the bottleneck is not the chip:

| Rank | Ticker | Notional | Q4 → Q1 | | --- | --- | --- | --- | | 6 | $BE | $879M | flat (was $876M) | | 7 | $SNDK | $724M | +190% (was $250M) | | 9 | $CRWV | $556M | +27% (was $437M) | | 13 | $IREN | $401M | +22% (was $329M) | | 14 | $CORZ | $389M | -7% (was $419M, activist 13D at 9.4%) | | 17 | $APLD | $320M | +15% (was $278M) | | 19 | $RIOT | $142M | +82% (was $78M) | | 21 | $CLSK | $104M | +513% (was $17M) |

Plus the call overlays: SNDK $389M call + BE $55M call + CRWV $141M call — adding ~$585M of long-delta on top.

SanDisk tripled

The single biggest long-side conviction call this quarter. SNDK went from a $250M holding (rank 9 in Q4) to a $1.11B aggregate exposure ($724M long + $389M calls) — 8.1% of the entire book, the second-highest-conviction position after Bloom Energy. Post-WDC-spinoff NAND pure-play. Combined with the MU straddle, the read is clear: Aschenbrenner thinks NAND is mispriced relative to where the data-center NAND cycle is heading.

Bitcoin miners pivoting to AI datacenters

The full basket: $CORZ ($389M, activist), $IREN ($401M), $APLD ($320M), $RIOT ($142M), $CLSK ($104M), $BITF ($39M), $BTDR ($30M), $WYFI ($21M), $HIVE ($6M, new). Combined: ~$1.45B across 9 names, all expressing the same thesis: the stranded-power megawatt portfolios that the Bitcoin miners built over the last cycle are the cheapest substrate to convert into AI HPC datacenter capacity. CleanSpark and Riot got the biggest percentage adds.

CORZ remains the only position that triggered a Schedule 13D — Aschenbrenner went from undisclosed to 5.8% in Aug 2025, amended to 9.4% in Oct 2025, and held at the 9.4% line through Q1 2026 (26M shares = the line in the 13F).

On-site AI power

$BE ($879M anchor + $55M calls), $SEI ($62M Solaris Energy Infrastructure), $BW ($20M Babcock & Wilcox / advanced nuclear), $PSIX ($26M Power Solutions), $TE ($44M T1 Energy / battery storage, NEW), $PUMP ($13M ProPetro / pressure pumping). Combined: ~$1.10B on the on-site-AI-power leg. The natural-gas-feedstock names (EQT, LBRT, KRC) all closed — the power-thesis tilt rotated from upstream gas toward fuel-cells + batteries + small modular reactors.

The new names

Three new positions this quarter, all sub-$50M starters:

• $SHAZ — SharonAI Holdings ($18M long). IPO'd on NASDAQ in February 2026; pure-play AI cloud-GPU operator. Signed a $1.25B five-year deal with ESDS Software Solutions in April 2026. The book's youngest holding by listing date.
• $TE — T1 Energy ($44M long). Solar/battery formerly known as FreyrBattery; grid-scale storage for datacenter power smoothing.
• $HIVE — HIVE Digital Technologies ($6M long). Canadian BTC miner with disclosed HPC pivot. The smallest of the nine-miner basket.

What closed

The Q1 2026 exits reveal as much thesis as the additions:

| Closed | Q4 2025 size | Reading | | --- | --- | --- | | $LITE Lumentum | $479M long | Optical leg fully exited (also closed COHR) | | $COHR Coherent | $89M long | Optical leg fully exited; replaced by a GLW put | | $CIFR Cipher Mining | $155M long | Miner basket trimmed to 9 names | | $HUT Hut 8 | $40M long | Miner basket trimmed | | $EQT EQT | $133M long + $38M calls | Natural-gas-feedstock leg exit | | $KRC Kilroy Realty | $50M long | REIT-as-data-center-land thesis exited | | $LBRT Liberty Energy | $10M long | Pressure-pumping consolidation around PUMP | | $TSEM Tower Semi | $85M long | Trailing-edge foundry exit |

The optical exits are loud. Q4 had $568M aggregate optical exposure (LITE + COHR); Q1 has a $21M short on Corning (GLW). The optical leg of the AI-supercycle thesis has fully flipped from net long $568M to net short $21M in one quarter.

The Intel position also reversed completely: Q4 had a $747M INTC call (the foundry-turnaround / 18A bet, rank 3 in Q4); Q1 has a $159M put + a $9M residual long. Either Aschenbrenner gave up on the IDM-comeback thesis, or this is a put-spread overlay against a much larger off-13F long.

What it means: the bottleneck has moved off the chip

The cleanest reading of the Q1 filing: Aschenbrenner is not bearish on AI. He's bearish on how AI is priced into the chip-stack.

The thesis from "Situational Awareness" was that compute would 100×, the industrial base would bend to deliver it, and the bottleneck names would compound. That thesis is intact in Q1 2026 — the book is still long Bloom Energy at $879M, still long ~$1.5B of bitcoin-miner-to-HPC conversions, still long $1.1B of SanDisk, still long CoreWeave at $700M aggregate. What's new is the read that the market is now mis-pricing where the bottleneck is.

The chips are no longer the bottleneck. The data-center power is. The land near substations is. The fuel cells, the batteries, the small modular reactors. The NAND for the hyperscale training corpus and inference cache. The ex-Bitcoin-mining shells with permitted megawatts. None of those have a 90% NVDA-style margin profile and an obvious chip-shaped story to tell on CNBC. So the market under-prices them, while the chips have absorbed all the AGI-imminence premium that the rest of the stack deserved.

The $8.5B short stack is the trade against the second half of that — the part where the market has put all the AI-margin premium into the wrong place.

It is, structurally, the same trade that the residualized correlation work surfaced two weeks ago: the Semiconductors and Hyperscaler names are no longer a coherent thematic cluster — they are SPY-beta. The capital flow inside them has stopped behaving like a bubble, which means the AI-supercycle premium baked into their multiples is at risk of being marked down. The names that are still clustering by thesis — Memory · HBM, Datacenter Power, the miner-pivot basket — are the long side of the Aschenbrenner Q1 2026 book.

How to read this against your own positioning

If you're long the AI supercycle:

• If your long book is concentrated in $NVDA / $AVGO / $AMD / $ASML / $ORCL alone — your positioning is on the wrong side of the largest AI-thematic short book disclosed this quarter. Acknowledge the trade Aschenbrenner is making before deciding whether to size around it or trade against it.
• If your long book mirrors the Aschenbrenner long side (BE, SNDK, the miner basket, CRWV) — you have an aligned book with the smartest dedicated AI-cycle fund.
• If you're running an AI-basket short — the $8.5B short stack is the playbook: hit the basket (SMH) plus the single names that rerated hardest. The size of the position is the most public conviction signal we've seen on the chip-short side.

The full filing data — every position, weight, share count, rank, with rationale — is live at /portfolios/situational-awareness. The Q4 2025 → Q1 2026 history is visible on the same page (the fund's filings table runs back to Q4 2024).

---

Disclosure: QuantAbundancia tracks fund portfolios for editorial purposes only. The Aschenbrenner fund and its principals have no relationship with QuantAbundancia. None of the above is investment advice; the 13F documents publicly-disclosed long-side positions only — short and derivative books that aren't in the form of long-leg puts are not reported and may materially change the read of the book.