Filed 2026-05-04 · from watchlist.md (B1 refresh) + portfolio-summary §2

Nine in core. Ten on the bench. The book by tier.

Three tiers: Core (deep-dive complete · 9 names · 7L 2S) · Bench (high-quality non-deep-dive worth tracking · 10 longs) · Watch (privates · Crucible · restricted). Bench-shorts honestly empty — only WOLF clears the structural-break bar.

§ 01

Core longs · 7 names.

deep-dive complete · ranked by conviction × layer uniqueness

TSM

TSMC — the chokepoint of chokepoints. L3 + L5 · ~90% leading-edge logic share · CoWoS-L is the binding AI shipment chokepoint · A16 H2 2026 first node with backside power · Taiwan-tail uninsurable

4.5%

NVDA

NVIDIA — reference architecture; Kyber 800V is the next pull. L6 + L7 · CUDA moat migrated from software to rack-as-product · ~60% TSMC CoWoS-L · ~70% SK Hynix HBM4 · 40% reserved for entry

4.0%

AVGO

Broadcom — captures margin on every hyperscaler ASIC. L6 + L7 · TPUv7 · Trainium3 · MTIA · Maia 2 · OpenAI 2027 chip · Tomahawk scale-out · the cleanest custom-silicon counter-leverage long

2.5%

TXN

Texas Instruments — primary cohort Taiwan-tail hedge + system-BOM bet. L8b + L8c · ~10–15% Taiwan exposure · NVIDIA GTC silicon-anchor partner · MOFCOM Sept 13 binary · 2.0% dividend yield floor · ~5–7% discrete GaN device share

2.5%

ETN

Eaton — buy on weakness; Boyd Thermal closes cooling gap. G1 + G2 + G2T + G3 · $19.6B record backlog · Boyd Thermal ($9.5B closed Mar 2026) · Mobility spin Q1 2027 · GOES gating backlog conversion · 16–19% AI-DC mix

1.75%

VRT

Vertiv — best business, worst valuation. Probe only. G1–G4 · purest AI-DC pure-play (~80% mix) · 32% ROIC · 2.9× book-to-bill · $15B backlog · 53.4× EV/EBITDA = 91% premium to ETN — the entire problem

1.25%

NVTS

Navitas — call-option-sized probe on density-bet GaN. L8b · single-stage 800V-to-6V GaN-IC architecture · ~17% global GaN share · April 2 2026 ITC FD catalyst · TSMC GaN exit July 2027 = triple foundry transition risk

0.65%

§ 02

Core shorts · 2 names.

deep-dive complete · honest assessment

WOLF

Wolfspeed — the cohort's load-bearing cautionary tale. L8a · substrate share collapse 60% → 34% in 36 months · May–July 2026 window: Q3 FY26 + $575M refi + CHIPS PMT · cohort's only clean cross-scenario hedge

−1.25%

INTC

Intel — foundry binary; 14A is back-from-the-dead or permanently sidelined. L3 + L5 + L6 · −$48B cumulative FCF · ROIC ~0.7% vs WACC 12.7% · federal-equity tail caps asymmetry to ~1:1.3 · pair preferred (TXN long or TSM long)

−0.65%

§ 03

Pair structures within the book.

3 explicit · finalized

L · CORE+HEDGE

TXN

2.5% NAV · system-BOM at L8b+L8c · vertical integration · 95%+ internal mfg by 2030 · 2.0% dividend

1:3

L · PROBE

NVTS

0.75% NAV · density bet at L8b · single-stage 800V→6V · GaN-IC patent estate · pair breaks at TXN $210

Pair 1 · TXN long / NVTS long sized 1:3 in TXN's favor. Both are GaN four-way race expressions at different layers of the bet. Finalized · not subject to re-sizing without thesis event.

L · COMPLETION

ETN

1.75% NAV (buy on weakness) · 16–19% AI-DC · multi-cycle cushion (utility T&D + aerospace) · lower bear case −27% to −36%

L · COMPLETION

VRT

1.25% NAV (probe) · ~80% AI-DC pure-play · 53.4× EV/EBITDA · bear case −42% to −54% on multiple alone

Pair 2 · ETN+VRT concentration trade · NOT a hedge. Combined cap binding: ≤1.5–2.0× TXN NAV. Both move directionally together on AI-capex. ETN gets larger allocation; VRT gets probe.

S · MEDIUM

WOLF

1.25% NAV gross short · substrate share 60% → 34% · refi stack May–July 2026 · cross-scenario winner in A, B, C

L · BENCH

~1.5–2.0× WOLF gross · onsemi (not yet deep-dived) · Qorvo SiC JFET · >$250M AI-DC revenue 2025

Pair 3 · WOLF short / ON long · same end-demand, opposite competitive trajectory. Approximated at 1:1.5 gross WOLF:ON until ON deep-dive completes. NVTS is NOT the right pair leg (different layer).

§ 04

Bench longs · 10 names worth tracking.

non-deep-dive · ranked for next-cycle promotion

#	Tk	Layer	Why on bench / catalyst	Sent.
01	ASML	L1 · EUV	"The only irreplaceable company in chips." Conviction 5; not promoted only because the analyst budget went to chip-to-grid. High-NA EUV adoption decisions at TSMC/Samsung 2026 are the live inflection. Top single completion candidate for next deep-dive cycle.	+2
02	000660.KS	L4 · HBM	SK Hynix · >50% HBM bits; HBM4 ramp into Rubin/MI450X/TPUv7. Memory bottleneck self-reinforces with HBM4 base die now consuming TSMC logic capacity. Next-cycle deep-dive candidate · closes L4 gap.	+2
03	KLAC	L1 · inspection	"Impossible to replace." Binding WFE constraint China cannot domesticate. Smaller than ASML but cleaner concentration; export-control tailwind.	+2
04	SU.PA	G1/G2/G3	Schneider Electric · record backlog · explicit 2028–2030 800V revenue ramp guide · ~22× EV/EBITDA = 60% better downside margin-of-safety vs VRT · cohort calendar-mismatch referent.	+2
05	2308.TW	G3/G4	Delta Electronics · overtook Foxconn by Taiwan market cap January 2026 · Rack-PSU consolidator · cleanest Taiwan-listed AI-server-PSU concentration.	+2
06	MPWR	L8c · VRM	Monolithic Power Systems · last-mm VRM at the GPU board · ~70% NVIDIA Vera Rubin VRM share · structurally adjacent to TXN's analog/embedded power layer · clean per-GPU-board content-growth thesis.	+1
07	IFX.DE	L8b · GaN	Infineon · the "scale" leg of the GaN four-way race · €2.5B DC revenue target by 2027 · 300mm GaN customer samples late 2025 · April 2 2026 ITC FD anchor.	+1
08	MU	L4 · HBM	Micron · HBM rule beneficiary; closing yield gap to SK Hynix. Memory layer hedge to SK Hynix on US-listing/CHIPS.	+1
09	AMAT	L1 · WFE	Applied Materials · Sculpta moves the EUV step count · broadest WFE exposure outside ASML.	+1
10	ON	L8a · SiC	onsemi · Qorvo SiC JFET acquisition · AI-DC revenue >$250M · cleanest paired long against WOLF short.	+1

§ 05

Bench shorts · honest assessment.

none clear the structural-break bar

The corpus carries 7 lean-short names (excluding WOLF in core). The honest assessment: none cleanly meet the WOLF "structural break" bar. Manufacturing shorts to balance the book would dilute conviction.

Tk	Honest assessment	Treat as
UMC	China 28nm oversupply pressure is real but cyclical, not a structural break. Trailing-edge oversupply is multiple-quarter pain, not the WOLF pattern.	Cyclical · low priority
PSMC	Same as UMC — trailing-edge oversupply, smaller and more concentrated.	Cyclical · low priority
VIS	Same as UMC — trailing-edge oversupply.	Cyclical · low priority
7731.T	Nikon — locked out of EUV. Structural share loss but already-priced and slow-burn.	Structural but tired
7751.T	Canon — same as Nikon. NIL is a sidebar, not a path back.	Structural but tired
Rapidus	Consortium · "Japan's prior leading-edge efforts have all failed." Not directly investable — tells on Toyota/Denso/SoftBank/MUFG/IBM funder set, not a stock.	Pass · not investable
STM	Watching for migration to −1. Most likely SiC structural-break name to follow WOLF. If WOLF deep-dive validates "this is the pattern, not the exception," STM migrates from neutral toward −1.	Watch for promotion

The cohort is structurally long-biased and the bench-short bar is appropriately empty. Manufacturing shorts to balance the book would dilute conviction. The right portfolio-construction tool is sizing discipline + Taiwan/TWD/KRW hedge structure, not synthetic shorts.

§ 06

Watch · privates / restricted / Crucible.

no public action — track only

Name	Status	Why on watch
OpenAI · Anthropic · xAI · DeepSeek	Private	Frontier-lab demand-side reference. Their commitments anchor multiple cohort theses (OpenAI/MI450X · Anthropic/TPUv7+Trainium3 · xAI/Colossus).
Innoscience	Private · added v3	The missing #1 GaN player at ~30% global share. 8" GaN-on-Si IDM at scale · NVIDIA 800VDC partner · sole Chinese vendor · cost-floor risk for NVTS · IFX patent litigation Apr 2 2026 ITC FD.
Ecolab/CoolIT	M&A · added v3	$4.75B CoolIT acquisition March 2026 · third well-capitalized CDU competitor · existing hyperscaler relationships · contests VRT cooling moat 2027–28.
Crucible portfolio · Supra · Elkhorn · Aravolta · Reliability Engine · Shatterdome · OrnnCx · Compute Index · Internet Backyard	Private	Crucible Capital is the user's investor-relevant adjacency. Supra (gallium recovery) · Elkhorn (hydrovaporization chiller, PUE 1.18 → 1.09) · Aravolta (GPU useful-life debate arbiter) · Shatterdome ($150–200/kW/yr BESS arbitrage).
Ayar Labs · Lightmatter · Celestial AI	Private	Co-packaged optics. Late-2020s structural impact at the chip package itself. Track for IPO.
Amperesand	Private	SST collaborations on 800V · private-side expression of the SST commercialization theme.
Groq · Cerebras · Tenstorrent · SiFive · Ampere · Crusoe Energy · KRAMBU · Lambda · CoreWeave	Private	Inference-chip and neocloud names · demand-side context for the unit-cost-of-intelligence theme.
Huawei · SMIC · CXMT · YMTC · Biren · TanKeBlue · SICC · SMEE	Restricted	China parallel stack. TanKeBlue/SICC are the WOLF share-takers — track for read-throughs to the SiC short thesis.

No GOOGL by orchestrator instruction: sentiment is TPU-driven but a position in Alphabet is a position in the entire ad business. AVGO is the cleaner expression of the TPU thesis.

§ 07

Top 5 for next deep-dive cycle.

recommended sequencing · after current cohort closes

ASML — long. The single highest-conviction non-deep-dive name. High-NA EUV adoption decisions at TSMC/Samsung in 2026 are the live catalyst. Closes the L1 lithography gap.
SK Hynix (000660.KS) — long. Memory bottleneck name and one of the user's three core bottleneck framings. HBM4 ramp into Rubin/MI450/TPUv7 is the live catalyst. Closes the L4 memory gap.
STM — short candidate (currently 0 sentiment, watching for migration). Most likely SiC structural-break name to follow WOLF. Deep-dive should test whether the "growth narrative is no longer as simple" framing migrates sentiment to −1.
MPWR — long. Last-mm voltage at the GPU board · structurally adjacent to TXN. Mention-count understates research surface.
Schneider Electric (SU.PA) — long. The 2028–2030 framing is the cohort's calendar-mismatch referent — Schneider deep-dive would calibrate VRT/ETN conversion-timing risk directly.