• Type: thesis-shift (subtype: confirms; quarterly-refresh)
• event_id: 2026-07-01-ai-power-bottleneck-q1-2026-13f-refresh
• Source: Aschenbrenner Situational Awareness Q1 2026 13F-HR (accession 0002045724-26-000008, period 2026-03-31), filed ~2026-05-18. Machine diff vs Q4 2025 (0002045724-26-000002) at diff-vs-2025-12-31. Repost pointer: inputs/tweets/2026-05-18-stockmktnewz-aschenbrenner-q1-2026-13f.md.
• Method note: pct = pct-of-common-equity-book (common stock only). 13F option values are the notional market value of the underlying shares, not premium; a held put is a long-put (bearish/protective) position — 13Fs cannot report short options — but the filing does not disclose strike, expiry, or spread structure, and cannot distinguish an outright short-thesis from a hedge against a long held in another vehicle.

Price truth: current price/RSI/trend from summaries (ai-power.json, semis.json, spacex-s1-supply-chain.json, focus.json, robotics.json), as-of 2026-07-01 close.

(a) Common-equity book Q4 2025 -> Q1 2026: doubled down on the core, shed optical + gas-supply, reshuffled miners

The conviction (common-stock) book got more concentrated in the two legs this perspective owns — onsite/generation and neocloud/BTC-miner-pivot — while non-core legs were cut.

Anchor + core adds:

• BE 22.87% -> 22.85% (pct flat, still #1). But shares TRIMMED 10.08M -> 6.49M (-35.6%); value held (~$876M -> $879M) only because BE roughly doubled over the quarter. Read: trimmed into strength, retained anchor status.
• CRWV 11.41% -> 14.46% (+3.05pp); +1.08M sh (+17.7%). Genuine add.
• IREN 8.58% -> 10.43% (+1.85pp); +3.0M sh (+34.5%). Genuine add.
• APLD 7.26% -> 8.32% (+1.06pp); +2.14M sh (+18.9%). Genuine add.
• CORZ 10.94% -> 10.12% (-0.82pp); -2.75M sh (-9.6%). Slight trim, stays top-4.
• RIOT 2.04% -> 3.70% (+1.66pp); +5.33M sh (+86.5%).
• CLSK 0.43% -> 2.72% (+2.28pp); +10.6M sh (+648%). New conviction in CleanSpark.
• BITF 0.42% -> 1.01%; ~+13M sh. BTDR 0.52% -> 0.77%; +1.65M sh. HIVE NEW 0.17%.
• SNDK 6.54% -> 18.84% (+12.3pp) — the largest pct move in the book, but almost entirely price, not buying (only +86k sh, +8.2%). SNDK ripped (now ~$2,039, +193% vs SMA200, strong-up). This is a memory-supercycle read carried inside the 13F, not a fresh ai-power conviction signal.

Exits / cuts (capital leaving non-core legs):

• LITE 12.50% -> 0 (FULL EXIT of a top-3 optical position). COHR 2.32% -> 0. The optical/transceiver leg was fully shed — a notable divergence from sibling optical-supercycle.
• EQT 3.48% -> 0 (common) and EQT call ($37.5M) exited. The gas-supply/fuel leg — this perspective's persistent laggard (EQT strong-down, -4.8% 30D on current tape) — was cut. SEI 2.24% -> 1.62% and LBRT 0.27% -> 0 also trimmed/exited. Fuel-supply de-emphasis is internally consistent with the README's standing "EQT is the fade" read.
• CIFR 4.04% -> 0 and HUT 1.03% -> 0 exited (miner reshuffle — capital rotated into CLSK/RIOT/BITF/HIVE, not out of the miner leg). KRC (Kilroy REIT) 1.30% -> 0, Tower Semiconductor exited.

New non-miner common: TE (T1 Energy) NEW 1.14% ($43.9M) — this is the same TE the separate line-162 watchlist-add row covers. AMD NEW 0.52% ($20.2M) — see (b). SHAZ (SharonAI Holdings) NEW 0.47% ($18.1M). Small NEW commons in NVDA/TSM/ASML/MU/GLW/INTC, all <0.2% — these read as the equity legs paired with the options overlay below, not standalone longs.

(b) The AMD "position" is a PUT, not a CPU-side bet — task premise corrected + the real headline: a ~$8.5B semis put overlay

The backlog row filed AMD as "small absolute but structurally significant — CPU-side bet at the workload-shift inflection." The filing does not support that. AMD common is $20.2M (99,138 sh, de minimis 0.52%); alongside it is a NEW AMD put of $969M notional (4.76M sh underlying). Net AMD exposure in this filing reads as hedge/bearish, not a bullish CPU-workload-shift long. cpu-shortage-supply-chain should NOT cite this 13F as AMD-bull corroboration.

AMD is one node in the actual Q1 headline: a large NEW long-put overlay across the GPU / logic / HBM / foundry / lithography complex —

~$8.5B notional of long puts against exactly the "GPUs + the silicon that makes them" layer. On the call side he added MU ($422M), SNDK ($389M), TSM ($355M) calls, cut the CRWV call ($774M -> $141M), and fully exited the INTC call ($747M -> 0). The two-sided options book (puts on the complex + calls on memory names) argues this is relative-value / volatility positioning rather than a clean directional short — but the tilt is unmistakable: the manager whose Q4 13F seeded this perspective is now, directly via options, expressing the flip side of "own the electrons, not the GPUs" — hedging/fading the hardware layer while keeping the conviction longs in power-generation + neocloud/miners.

Effect on thesis

HARDENS the core, no posture change, no publish. The down-stack "electrons over GPUs" frame that the 2026-06-22 tape entry inferred from price action is now expressed directly by the originating manager: (1) common book concentrated further into generation (BE) + neocloud/miners (CRWV/IREN/APLD + CLSK/RIOT/BITF/HIVE), (2) optical + gas-supply legs shed, (3) a large put overlay fading the GPU/memory/foundry complex. This is corroboration of an existing posture, not a new signal — the perspective already owns every long name here, so nothing to promote.

Delayed-filing + trend-hold caveat (do NOT chase). The 13F is a 2026-03-31 snapshot filed ~mid-May — a ~3-month-lagged read. On the current tape the highest-beta names he ADDED to have de-rated hard: CRWV RSI 36.7 / down / -47.9% off high; IREN RSI 35.9 / down / -42.5% off high; CLSK RSI 39.1 / -25% 30D; APLD RSI 38.7 / -23% 30D; RIOT RSI 43.2 / -12.6% 30D. The generation/equipment leg held: BE RSI 54.7 strong-up +12.3% 30D; GEV RSI 62.2 strong-up, -2.4% off high. Per the user's trend-hold lens, the filing reinforces that the generation/equipment leg (BE/GEV) is the cleaner expression — both by the filing's own staying-power (BE anchor, trimmed-into-strength) and by which leg the current tape is rewarding. The miner/neocloud adds are a lagged conviction datapoint on names now mid-drawdown, not an entry trigger.

Housekeeping this refresh closes:

• The README "Q1 2026 13F (NEW) — filing window late-July 2026" catalyst (Catalysts-to-watch) is now stale — that filing is public and imported; re-point the note to the Q2 2026 13F window (~mid-Aug 2026, 45-day rule).
• Resolves the line-164 "$SHAZ real or phantom?" question: SHAZ is real (SharonAI Holdings Inc, US CUSIP 778920306, common stock, $18.1M new). It stays untracked per that row's existing not-doing/low-value disposition (fresh de-SPAC, tiny, no trend) — noting the fact, not re-proposing an add.

Per-ticker

BE anchor (trim-into-strength, still #1). CRWV / IREN / APLD / RIOT / CLSK / BITF / HIVE ^ conviction adds (but lagged snapshot; current tape de-rated — no entry). CORZ / SEI slight trim. LITE / COHR EXIT (optical leg shed — flag for optical-supercycle). EQT EXIT / LBRT EXIT (gas-supply leg shed — consistent with EQT being the standing laggard). CIFR / HUT EXIT (miner reshuffle, not de-risk). GEV neutral-positive (not in the 13F beyond BE call; current tape strongest leg). AMD = put overlay, NOT a CPU-bull tell (reframes task premise). NVDA / AVGO / ORCL / TSM / ASML / MU / SMH = NEW put overlay (down-stack fade expression; ai-capex-digestion relevant). SNDK pct-of-book jump is price, not buying (memory-supercycle).

README refresh banner (paste into README "Current Data" section, above the 2026-06-26 banner)

🔄 2026-07-01 Q1 2026 13F quarterly refresh (Aschenbrenner) — HARDENS the down-stack frame, no posture change: The originating 13F now expresses "electrons over GPUs" directly. Long common book concentrated further into generation + neocloud/miners (added CRWV +17.7%, IREN +34.5%, APLD +18.9%, RIOT +86.5%, CLSK +648% shares; BE trimmed-into-strength but still #1 at 22.9%), while the optical leg was fully exited (LITE 12.5%->0, COHR->0) and the gas-supply leg was cut (EQT->0, SEI/LBRT trimmed). New this quarter: a ~$8.5B notional put overlay across the GPU/logic/HBM/foundry/litho complex (SMH/NVDA/ORCL/AVGO/AMD/MU/TSM/ASML) — a direct fade of the hardware layer alongside the conviction power longs. Correction: AMD is a $969M put, not the bullish CPU-side bet the backlog row assumed ($20.2M common is de minimis). Delayed 2026-03-31 snapshot — the miner/neocloud names he added are now deeply de-rated (CRWV RSI 37/-48% off high, IREN RSI 36/-43% off high, CLSK/APLD RSI ~39); the generation/equipment leg holds (BE RSI 55 strong-up, GEV RSI 62). Conviction confirmation, NOT an entry signal. See log.md 2026-07-01 entry.

---

🔄 2026-07-05 operating book: Ratified into PRIMARY 2 — the power complex (this + ai-power-delivery + nuclear-fuel-cycle) now operates as ONE book thesis, anchored here. Breaks/exit-tape section added to README. Spec: operating-book.