§ cohort  ·  Architecture  ·  Two layers, held independently
Filed 2026-05-04  ·  from portfolio-summary §2 + §4 + refinement-log Findings 12, 16

Two layers, held independently.

The cohort holds chip-layer names (NVTS / TXN at L8b–L8c) and grid/system-layer names (ETN / VRT at G1–G4) as independent positions, not as a chain. Box-builders are vendor-agnostic on power-semi supplier by commercial design. ETN orders do not mechanically translate into NVTS or TXN P&L.

§ 01

Why we hold two layers, not one chain.

refinement-log Finding 12 · ETN customer-analyst confirmed

Per the C-COHERENCE pass: NVIDIA GTC March 2026 named VRT, ETN, and Schneider all at the systems tier, and TI exclusively at the silicon-anchor tier. No box-builder names a power-semi partner by commercial design. Eaton's customer-analyst confirmed: ETN purchases IGBTs/GaN/SiC on a competitively sourced basis and does not name power-semi suppliers. The chip-to-grid frame justifies why these names are researched together, not because their P&Ls flow mechanically through each other.

Independent positions · separate bets on the same buildout Grid / system layer · G1–G4 Captures $700K–$2.8M /MW · vendor-agnostic on power-semi supplier · Boyd-via-ETN closes cooling gap ETN COMPLETION · 1.75% VRT COMPLETION · 1.25% Held as completion tier · combined cap ≤ 1.5–2.0× TXN Chip layer · L8b–L8c Captures $50K–$150K /MW · margin per dollar · TXN system-BOM bet at L8b–L8c+L7+L6 TXN CORE+HEDGE · 2.5% NVTS PROBE · 0.65% TXN-NVTS pair finalized at 1:3 in TXN's favor · pair breaks at TXN $210 NOT a chain × Box-builders vendor-agnostic on power-semi A strong ETN order book does not automatically translate into NVTS or TXN revenue. Two independent bets on the same infrastructure buildout.
Source · refinement-log Finding 12 (C-VRT-1 fan-out · NVIDIA GTC March 2026 + ETN customer-analyst confirmation) + Finding 16 (ETN+VRT combined cap mandate). Cohort architecture roles per portfolio-summary §2.
§ 02

The ETN+VRT cap rule, visualized.

non-negotiable · binding at full size

The cohort's largest residual risk is AI-capex single-factor concentration. ETN (16–19% AI-DC mix, growing toward 22–26% post-Mobility spin) and VRT (65–80% AI-DC mix) are both AI-capex amplifiers. Holding both alongside NVDA and AVGO means a sustained hyperscaler capex pause hits the cohort from the top (accelerator revenue) and the bottom (infrastructure revenue) simultaneously. The cap rule is the portfolio-construction mechanism that prevents the AI-capex amplification from becoming a single-factor bet wearing two tickers.

Combined ETN + VRT ≤ 1.5–2.0× TXN NAV position portfolio-summary.md §2 · binding constraint · refinement-log Finding 16 Base sizing · today TXN 2.5% ETN 1.75% VRT 1.25% SUM 3.00% Ratio = 1.2× CLEARS · within ceiling Full size at correct entries TXN 3.0% ETN 2.5% VRT 2.0% SUM 4.5% Ratio = 1.5× EXACTLY AT CEILING · do not exceed Hypothetical · undisciplined TXN 3.0% ETN 3.0% VRT 2.5% SUM 5.5% Ratio = 1.83× BREACHES · single-factor amplification In Scenario A (AI capex pause), the cap rule's value: At base sizing (3.0% combined): ETN+VRT contribution to cohort drawdown −1.02% NAV. At full-size (4.5% combined): −1.55% NAV. Cap rule saves ~0.53% NAV. In Scenario C (rates higher-for-longer), VRT's 38% drawdown alone benefits from the cap by ~0.47% NAV. → The cap rule is most valuable in Scenarios A and C. It is working as designed.
Source · portfolio-summary.md §2 (NAV math + cap-rule check); scenario-stress.md §3 (per-scenario quantification of cap-rule value). Net long exposure at base sizing: ~15.25% NAV.
§ 03

Pair structures.

3 explicit · 2 portfolio-construction · 1 not recommended

Pair 1 · TXN long / NVTS long — 1:3 in TXN's favor.

Long · core

2.5% NAV · vertical-integration anchor at L8b+L8c. System-BOM bet captures the 1,500+ analog SKUs around the headline GaN socket. ~5–7% discrete GaN device share. Dividend yield 2.0%. Lower-beta, higher-conviction expression.

VS
Long · probe

0.75% NAV · density bet at L8b. Single-stage 800V-to-6V GaN-IC architecture. ~17% global GaN share, #2. Higher-beta, asymmetric upside on Kyber design-in conversion. Pair breaks at TXN $210.

Both names express the GaN four-way race at different layers of the bet. TXN bets on system-BOM vertical integration; NVTS bets on density-architecture lock-in at the discrete socket. The pair captures both the conservative-and-broad and speculative-and-concentrated expressions of the same thesis without doubling down on either. Finalized · 1:3 ratio · pair breaks at TXN $210.

Pair 2 · ETN long / VRT long — concentration trade, not hedge.

Long · completion

1.75% NAV (buy on weakness; full size below $360) · 16–19% AI-DC mix growing to 22–26% post-Mobility spin. Utility T&D + industrial + aerospace floor. 27.9× EV/EBITDA. Lower bear-case drawdown (−27 to −36%).

+
Long · completion

1.25% NAV (probe at $330) · 65–80% AI-DC mix · 32% ROIC · 2.9× book-to-bill. 53.4× EV/EBITDA = 91% premium to ETN — the entire problem. Bear case −42 to −54% on multiple compression alone.

Per macro analyst (refinement-log): "ETN + VRT is a concentration trade, not a hedge. Both are AI-DC longs." Combined cap binding at ≤1.5–2.0× TXN NAV. ETN gets the larger allocation (lower bear case, multi-cycle cushion); VRT gets the probe (pure-play upside option for the Kyber cycle).

Pair 3 · WOLF short / ON long — SiC structural break + selective long.

Short · medium

1.0–1.5% NAV gross short · substrate share collapse 60% → 34% over 36 months. May–July 2026 window: Q3 FY26 + $575M refi + CHIPS PMT resolution stack. Cohort's only clean cross-scenario hedge.

VS
Long · bench
ON

~1.5–2.0× WOLF gross · onsemi (not yet deep-dived) · Qorvo SiC JFET acquisition; AI-DC revenue >$250M in 2025. Strengthening competitive position in same end-market.

Same end-demand (EV + AI-DC), opposite competitive trajectory. Captures the "structural transitions don't guarantee equal outcomes" theme. Approximated at 1:1.5 gross WOLF:ON until ON deep-dive completes. NVTS is NOT the right WOLF pair leg (different layer: GaN vs SiC).

Pair 4 · INTC short / TXN long — foundry leadership + embedded analog.

Short · probe

0.5–0.75% NAV · structurally-impaired-IDM betting on 14A binary · −$48B cumulative FCF · ROIC ~0.7% vs WACC 12.7% · foundry external book press-release-grade

VS
Long · core+hedge

2.5% NAV · vertical-integration-with-CHIPS-subsidy thesis · ~$6.6B PV from CHIPS + §48D + state subsidies · 95%+ internal manufacturing by 2030 · no foundry ambiguity

US fab buildout outcomes — opposite directions. Note · Section 232 is ambiguous for this pair (a punitive tariff on advanced-node imports hurts TSM/NVDA but supports INTC's domestic-foundry narrative). The TSM/INTC pair (the prior portfolio summary's structural-foundry expression) is cleaner because it is purely process-node, not tariff-mediated.

Pair 5 · TSM long / INTC short — the foundry leadership pair (structural).

The cleanest structural-divergence expression in the cohort. TSM 90% leading-edge logic share + CoWoS-L chokepoint vs INTC −$48B cumulative FCF on 14A binary. Purely process-node · not Section-232-mediated. Sized: TSM 4.5% NAV / INTC −0.65% NAV gross short (asymmetric in TSM's favor reflects asymmetric conviction).

§ 04

Layer coverage · what's owned, what's a gap.

portfolio-summary §3 · 2 explicit gaps

Cohort coverage by chip-to-grid layer. Two explicit gaps identified by the B1 watchlist refresh: L4 (memory) closes with SK Hynix; L1 (litho/WFE) closes with ASML. Both are the highest-priority next deep-dive candidates per the B1 sequencing recommendation.

LayerLayer nameCohort name(s)Coverage
G0Power generation / grid— out of scope · tracked via Crucible privates (Shatterdome · Oklo · Bloom)By design
G1Site / utility interfaceETN (primary) · VRT (UPS/ATS subset)Covered
G2DC backboneVRT · ETN (overlapping)Covered
G2TCooling / thermalVRT (incumbent ~22% CDU) · ETN/Boyd · Ecolab/CoolIT (bench)3-way
G3800V transition layerETN · VRT · VRT H2 2026 launch is primary catalystCovered
G4Rack / row powerVRT · connector layer (TE · APH) uncoveredPartial
L8aSiC (HV front end)WOLF short · ON paired-long on benchShort-only
L8bGaN (high-density conversion)NVTS (density) · TXN (system-BOM vertical) · Innoscience #1 + IFX uncoveredTwo expressions
L8cBoard-level powerTXN · MPWR uncovered (bench)Partial
L8dMOCVD equipment— AIXTRON · Veeco watch-tierBy design
L7Networking / opticsNVDA (NVLink · IB) · AVGO (Tomahawk)Covered
L6Compute / acceleratorsNVDA · AVGO · INTC (short)Covered
L5Advanced packagingTSM (CoWoS-L) · INTC (EMIB · Foveros · short)Covered
L4Memory (HBM · DRAM · NAND)— None — SK Hynix on bench (next add)GAP
L3Foundry / processTSM (90% leading-edge) · INTC (foundry pivot · short)Covered
L2EDA— Cadence · Synopsys watch-tier · no deep-dive plannedBy design
L1WFE / lithography— None — ASML on bench · top single completion candidateGAP
L0Materials / gases— Shin-Etsu · SUMCO · SUPRA (Crucible) watch-tierBy design

Cohort net direction: 7 longs / 2 shorts. Layer coverage on the long side genuinely broad (G1 through L7). The two gaps (L4 memory, L1 WFE) are the highest-priority next deep-dive candidates — both already conviction-5 / +2-sentiment names sitting on the bench. The cohort does not hedge Taiwan-tail through the shorts (INTC and WOLF hedge foundry-leadership and SiC structural-share respectively); Taiwan-tail is materially closed instead by the TXN/ETN/VRT inclusion (low-to-zero Taiwan production exposure).

§ 05

What this architecture does and does not protect against.

portfolio-summary §5 · 3 risks · honest assessment
Risk 01 · Largest residual ~9.5%

NAV in names where AI-capex pause is the dominant bear-case mechanism (NVDA 4.0% · AVGO 2.5% · ETN 1.75% · VRT 1.25%). At −35% drawdown on this group: −3.3% NAV. The ETN+VRT cap rule is the containment mechanism.

Risk 02 · Cooling 3-player 22% → ?

VRT CDU share (~22% pre-M&A) contested by ETN/Boyd ($1.7B closed Mar 2026) and Ecolab/CoolIT ($4.75B). Quarterly bookings monitor required. 5+ ppt VRT loss over 2 quarters = trim to sub-1% NAV.

Risk 03 · Calendar mismatch 128–144wk

GOES (ETN's gating) and HFO refrigerants (VRT's gating) gate calendar conversion of orders to revenue independent of demand. Order disappointment driven by chokepoint-limited production, not demand, is "add on weakness" not exit.

Secondary · Mexico tariff $150–250M

25% Section 232 scenario adds annual COGS to ETN with only partial pass-through. VRT's Mexicali also exposed. Cohort risk-matrix gap not in the prior Taiwan/China framing.

Secondary · MOFCOM TXN 3–6%

FY26 revenue at meaningful adverse-outcome risk on Sept 13 2026 final determination. Largest near-term asymmetric binary in any cohort long. Hold TXN at 1/2–2/3 size through Q3 2026.

Secondary · VRT EMEA −20%

Q1 2026 YoY in second-largest market — first geographic crack in the AI-DC order story. Two more negative quarters confirm structural Schneider share loss; one quarter is timing.