Question: Aschenbrenner research dossier — essay-mined lanes, the pair-trade put book, 13F track-record calibration, and cross-manager convergence Verdict: open

Dossier hub. Built 2026-06-05 on top of the newly-archived primary source (Situational Awareness full text). Five parts. Part 1 is the standalone scorecard (linked, not duplicated); Parts 2–5 are new. Price truth throughout = the scan summaries, latest 2026-06-03/04; no prose math; producer-side (candidates, not watchlist mutations; no self-promote).

What we're asking

Now that the 2024 essay is primary research and the essay→tape scorecard is done, go deeper on the man whose 13F is our ai-power-bottleneck spine:

1. (Part 2) What's in the essay we never turned into trades?
2. (Part 3) The put book — "he's shorting now." Hedge or directional short?
3. (Part 4) How much should we trust his moves — a real calibration.
4. (Part 5) Is the long-bottleneck/short-megacap barbell a lone bet or consensus?

Executive summary

Aschenbrenner runs a barbell: long the AI picks-and-shovels (BE/SNDK/CRWV/miners, ~77% of equity book in the top 5), short/hedged against the consensus AI mega-caps ($8.46B put notional on NVDA/ORCL/AVGO/AMD/ASML/MU/TSM/INTC + the SMH ETF). The long leg is the essay's thesis and it has paid enormously; the short leg is the essay's thesis taken to its bearish edge and it has been wrong on every name so far. Three new findings sharpen the picture:

• The long leg is now a forming consensus; the short leg is near-lone. Other tracked managers are piling into SNDK/CRWV (his bottlenecks) but are long the very mega-caps he's putting. Brad Gerstner/Altimeter is the clean mirror image — long both ends. (Part 5)
• His direction is excellent, his exit timing is poor. Every name he fully exited in Q1 2026 (LITE/COHR/CIFR/HUT/TSEM) kept ripping +39% to +142%. Weight his entries, discount his exits. (Part 4)
• The essay still has two genuinely un-traded lanes: confidential-compute / secure-datacenter hardware (Ch. IIIb) and government-compute / federal-AI integrators (Ch. IV — "The Project"). Both are real coverage gaps. (Part 2)

---

Part 1 — Essay → tape scorecard (standalone, linked)

The essay's predictions vs what the tape did, and the longs/exits/puts read, live in: 2026-06-05-aschenbrenner-essay-vs-13f-scorecard.md. One-line: power-as-bottleneck vindicated (BE +1863% since the essay); his long book worked (7/8 strong-up); his exits were early; his new put book is underwater on the tape. Read it for the per-name tables.

---

Part 2 — Essay chapters we never traded → new lanes

Chapter IIIa ("Racing to the Trillion-Dollar Cluster") is already our ai-power lane. Mining the rest of the essay against our 41 watchlists + active perspectives surfaced two high-value coverage gaps and three marginal ones.

Ranked gaps:

1. Confidential-compute / secure-datacenter hardware (IIIb) — HIGH, cleanest gap. The essay's load-bearing security claim is hardware (confidential compute, hardware encryption, air-gapped secure datacenters, hardware roots-of-trust) — a different supply chain than our SaaS-security cybersec list. Beneficiaries: enclave silicon (NVDA/AMD/INTC — already held) + memory-security IP (RMBS, only in spacex-s1-supply-chain today) + secure-element vendors. → filed as a research task (scope the pure-plays via research-trace). → RESOLVED 2026-06-17 (confidential-compute-...-coverage-gap): no watchlist — substrate already covered, remainder is a thin-exposure basket; only RMBS + LAES are real pure-plays.
2. Government-compute / federal-AI integrators (IV) — HIGH. "The Project" = USG-orchestrated AGI on a defense-contractor model. PLTR is tracked only as generic AI-infra exposure; LDOS/SAIC/BAH/CACI/ACN-federal are uncovered. Slow, policy-contingent (2027/28), so watch-and-seed, not urgent. → filed as a watchlist-decision + research task. → RESOLVED 2026-06-17 (government-compute-...-coverage-gap): watchlist WARRANTED (gated). URGENCY CORRECTION — the "2027/28, not urgent" call was wrong: the catalyst is live in 2026 ($32B DoD AI ceiling, $800M lab awards, Leidos passed SAIC).
3. Nuclear/SMR + uranium completeness (IV) — MEDIUM. SMR/OKLO/CEG/VST covered; CCJ/LEU/BWXT (fuel-cycle + SMR-builder) uncovered. Lower urgency — power demand is already well-owned. → RESOLVED 2026-06-17 (verification only): already closed by the nuclear-fuel-cycle lane promoted 2026-06-06 — one day after this dossier. The nuclear-fuel-cycle watchlist (13 names) covers CCJ/LEU/BWXT + all 4 layers (mining CCJ/UEC/UUUU/LEU/ASPI, builders BWXT/OKLO/SMR/NNE, utilities CEG/VST/TLN/DUK). Deferred decision is moot. NXE/DNN (dev-stage miners) left uncovered deliberately — spec, off the upstream-monopoly + cash-flow-utility picks.
4. AGI-2027 ⟂ Taiwan-2027 convergence (IIId) — MEDIUM, framing only. Constituent trades all trackable; value is the named scenario as an eyes-and-ears trigger. → noted as a perspective-framing candidate, not new work.
5. Inference-specific emphasis (II) — LOW. A sharpening of existing perspectives; one-line note on next ai-power-bottleneck refresh. No standalone work.

(Full per-chapter read + the watchlist/perspective files checked: research-thread notes, this session. Prices code-computed from data/summaries/*.json, bar 2026-06-03.)

---

Part 3 — The pair-trade / put book ("he's shorting now")

Structure (Q1 2026 13F, R19 corrected read): three layers — equity longs $3.86B (bullish), calls $1.36B (bullish leverage), puts $8.46B underlying notional (bearish/hedge). Put-to-equity ratio 2.19×. This was new in Q1: his Q4 2025 book was essentially un-hedged long. The puts are all on consensus mega-caps + the SMH semi ETF — none on his own cohort. (Source: published/research-notes/2026-05-20-r19-aschenbrenner-q1-corrected-all-action.md, raw EDGAR salp13fq1xml.xml, as of 2026-03-31.)

Hedge or directional short? The honest answer: structurally a pair trade, but P&L is unscoreable from a 13F. Arguments:

• For "real short, not tail-hedge": the magnitude. $8.46B notional is 2.19× the equity book; even delta-discounting OTM puts (δ 0.2–0.4) leaves $1.7–3.4B delta-adjusted — comparable to the whole equity book. A tail-hedge sleeve is usually a few percent, not 200%+. R19 calls it "not a tail-hedge sleeve… a pair-trade structure."
• For "could still be cheap insurance": a 13F discloses no strike, expiry, or premium. If these are deep-OTM far-dated puts, the actual capital outlay (and thus the loss when they fail) is small even at $8.46B notional. We genuinely cannot tell tail-hedge from directional short without Greeks — and a 13F never provides them. (Feasibility gate: Greek/P&L modeling is NOT possible from this data. We stop at "structure + tape direction," and do not manufacture a P&L number.)

The cleanest tell — INTC. Q4 2025 he held an INTC call $747M (bullish); Q1 2026 he flipped to an INTC put $159M (bearish) — a complete directional reversal — then INTC ran +147% ~3m (scan summaries, bar 2026-06-04). A clean, badly-timed flip.

Has the short leg worked? No — every put underlying is up since entry, most near 52-week highs (3-month return since his Q1 entry quarter, per the scan summaries): SMH +60%, NVDA +17%, AVGO +51%, AMD +168%, MU +154%, TSM +23%, ASML +24%, INTC +147%, GLW +39%, ORCL +52% (the one name −33% off its high). See the scorecard table for the full grid.

The implicit thesis this encodes: "the mega-caps are over-valued relative to the infrastructure they ultimately depend on — so I'm long the bottlenecks and short the consensus." That is the essay's own logic pushed to its bearish edge. It is also the R19-flagged "AI mega-cap mean-reversion" candidate perspective — still unspawned. This dossier is more evidence for scoping it, not a decision to open it. The tape has rejected it for two quarters; the open question (Part beneath) is whether he holds.

---

Part 4 — 13F track-record calibration

Data reality: only ~2 usable quarters (Q4 2025 + Q1 2026 normalized) — one Q-over-Q diff. The tracking profile's hit-rate fields are literally all zeros; the 30/60/90d window hasn't matured. A real multi-year hit-rate is NOT computable yet — do not fake one.

What we CAN say:

• Direction/entries: excellent (regime-confounded). The paper mirror shows ~23 of 25 closed positions positive — but most mirror exits fired on our mechanical +20%/−15% rules, not his trades, and the window is a roaring AI-infra/memory tape. So this measures "his disclosed longs went up after disclosure," not a cross-cycle edge. Ledger-realized standouts: SNDK +356%, BE +194%, the miner/memory cohort broadly positive.
• Exit timing: poor (this is the decision-relevant finding). Every name he fully exited in Q1 kept rising — LITE +39%, COHR +55%, HUT +142%, TSEM +117%, CIFR +64% (off a −20.9% realized loss) (3-month return, per the scan summaries). EQT is the lone good exit (−11%, the gas pure-play — consistent with the "gas-as-fuel → compute-as-bottleneck" thesis evolution).
• Concentration: extreme. Top 5 = ~77% of equity book (BE 22.85% + SNDK 18.84% + CRWV 14.46% + IREN 10.43% + CORZ 10.12%), per holdings.normalized.json.

Suggested signal-weight: High on entries/direction, near-zero on exit timing, and always net the put overlay. When he initiates or concentrates (the SNDK double, the miner-pivot longs), use it as positioning confirmation for ai-power-bottleneck / memory-supercycle. When he sells, do NOT mirror — the existing guardrail ("we don't sell because Aschenbrenner did") is now empirically confirmed by the early-exit price truth. And don't read his concentrated longs as an unhedged all-in vote — the $8.46B put overlay means his net stance is more contested than a long-only mirror shows.

Feasibility caveats: no real hit-rate (window immature); the ~92% mirror win-rate is regime-confounded with no drawdown in-sample; mirror exits are mostly our rules not his; the put book is unscoreable; 13F is 45–95d stale and long-US-equity-only; hold-time distribution (profile prior: 270d typical) is unverified on one diff.

---

Part 5 — Cross-manager convergence

Verdict: forming consensus on the LONG bottleneck leg; near-LONE on the SHORT mega-cap leg. Splitting the barbell and checking who else is positioned each way:

• Clearest agreement: SNDK (memory bottleneck) — Aschenbrenner + Tepper + Druck. CRWV runner-up (him + Gerstner + Baker).
• Clearest disagreement: the mega-cap puts. Brad Gerstner / Altimeter is the mirror image — long CRWV (agrees on the bottleneck) and long NVDA/AVGO/TSM (opposes the short). He's long both ends of the barbell Aschenbrenner is pair-trading.

Caveats: 13F is long-US-equity-only — "everyone else is long NVDA" means long on the disclosed equity; we can't see whether they hold undisclosed puts. Altimeter & Atreides come from the living-profile pipeline (Q1-dated) and have not had the put/call blast-radius audit — treat as disclosed-long-equity. Data-integrity flag: aschenbrenner-13f.md frontmatter still lists him "bull AMD/AVGO" — a stale artifact of the pre-fix importer that read his puts as longs (R19 corrected this); do NOT trust that profile's mega-cap stances. (Sources: published/research-notes/2026-05-20-r19-q1-13f-blast-radius-audit.md, .../r19-aschenbrenner-q1-corrected-all-action.md, .../r18-q1-13f-cross-manager-diff.md, .../r23c-13f-all-action-ledger-validation.md; profiles altimeter-13f.md, atreides-13f.md.)

---

Verdict + reasoning

Verdict: open (living dossier). The synthesis:

• Essay core thesis: vindicated (Part 1) — power-as-bottleneck built the lane and paid.
• His edge is direction, not timing (Part 4) — weight entries, discount exits.
• The barbell's long leg is becoming consensus; the short leg is his alone (Part 5) — Gerstner is the clean opposite. That makes the put book the single most contrarian thing he's doing, and so far the tape says he's wrong (Part 3).
• Two un-traded essay lanes are real (Part 2) — confidential-compute hardware and federal-AI integrators — filed as research tasks below.

All of this stays producer-side: positioning + tape observation, not buy/sell calls, no key_ticker mutations. We are not trading off a 65-day-stale options snapshot.

What's next / open questions

• Q2 2026 13F (~2026-08-15) is the hinge: did he hold / add / cover the megacap puts after they ran against him? Cover → "tactical hedge, closed"; add-into-loss → conviction. This is the single most important resolve, and only that filing can do it.
• Confidential-compute hardware (Part 2 gap 1) — scope the pure-plays (filed).
• Federal-AI integrators (Part 2 gap 2) — watchlist-decision on LDOS/SAIC/BAH/CACI (filed).
• TSEM exit attribution — R19 names a $85M TSEM exit; the normalized Q4 data shows an unmapped CUSIP at $84.9M. Price-truth conclusion (it ran +117%) is unaffected, but verify the TSEM tag before quoting it as a clean exit elsewhere.
• Profile re-derivation — the Aschenbrenner 13F tracking profile needs re-deriving post importer-fix (the options-parsing bug is already filed as an engine task); a warning annotation is added to the profile in the meantime.

Price-truth note

All %/price figures are code-computed from the scan summaries, latest 2026-06-03/04; 2y≈essay window, ~3m≈since Q1-2026 close. 13F dollar magnitudes are filing-reported underlying notional as of 2026-03-31, not live-price claims. Ledger P&L is paper-mirror-realized, not prose math. No returns/RSI/P&L were computed in prose anywhere in this dossier.