§ cohort  ·  Hub  ·  9 names · 7L/2S · 15.25% NAV net long
Cohort in 30 seconds
Position book · base sizing
TkSide · Conv.% NAVRole
TSML · 54.5Core
NVDAL · 44.0Core
AVGOL · 42.5Core
TXNL · 32.5Core+hedge
ETNL · 31.75Completion
VRTL · 31.25Completion
NVTSL · 20.65Probe
WOLFS · 4−1.25Short
INTCS · 3−0.65Short
Net long+15.257L · 2S
Jump to
Filed 2026-05-04  ·  9 names covered  ·  168 in the ledger  ·  v3 C-COHERENCE

The rack is the new grid.

A 9-name book held as one continuous voltage system — from the substation, through the building, into the rack, onto the board, and into the die. Every layer is being redesigned simultaneously.

§ 01

The view, in one paragraph.

Cohort synthesis · v3

The cohort expanded from 5 pure-semi names to 9 chip-to-grid names because the dollar-weighted center of mass of the AI-DC build sits at G1–G4 (electrical infrastructure), not at L8b/L8c (power semis): ETN and VRT each capture $700K–$2.8M per MW of addressable content versus roughly 20–60× less for the power-semi BOM layer. Convictions anchor at the silicon tier (TSM at 5/5, NVDA and AVGO at 4/5) and progressively scale down toward the power-infrastructure completion tier (ETN, VRT at 3/5) and the probe tier (NVTS at 2/5). Two shorts (WOLF at 4/5, INTC at 3/5) hedge SiC substrate share collapse and foundry-leadership binary — neither is a Taiwan-tail hedge. The single most important post-expansion finding: Taiwan-tail concentration is materially reduced (TXN/ETN/VRT all carry low-to-zero Taiwan production exposure), but AI-capex single-factor concentration is materially amplified — which is why the ETN+VRT ≤ 1.5–2.0× TXN cap rule is binding and non-negotiable.

§ 02

The cohort, as one stack.

G0 → L0 · 9 names · $/MW addressable

Read top-to-bottom: from the power plant to the transistor. Each layer carries the cohort name(s) that own it, the dollars per MW captured at that layer, and the conviction tier. The asymmetry between G1–G4 and L8b/L8c is the structural argument for the scope expansion — the electrical layers capture an order of magnitude more dollar-volume per MW than the silicon they feed.

Voltage path ↓ DOWN-CONVERTING $/MW captured DOLLAR-VOLUME → Grid · Building · Rack G0 · Generation SMR · gas · BESS · HVDC — out of scope · tracked via Crucible privates G1 · Site interface switchgear · MV transformers · ATS ETN · VRT $500K–$2M /MW G2 · DC backbone UPS · BBU · busway · distribution VRT · ETN $600K–$1.3M /MW G2T · Cooling/thermal CDU · cold plates · immersion VRT · ETN/Boyd · CoolIT $300K–$700K /MW G3 · 800V transition SST · 800V PSU · supercaps ETN · VRT $200K–$500K /MW G4 · Rack power PDU · busbar · power shelves VRT · (Delta · TE · APH) $150K–$400K /MW ↑ Electrical layer captures the dollars · ↓ Silicon layer captures the margin · 20–60× per MW asymmetry Power semis L8a · SiC substrate 800V rectification · EV-class WOLF (short — share collapse) decaying L8b · GaN conversion 800V→48V · 800V→6V NVTS · TXN · (IFX · Innoscience) $50K–$150K /MW L8c · Board power VRM · BCM · eFuse · 48→1V TXN (MPWR · Vicor on bench) $30K–$80K /MW Chip stack · L7 → L0 L7+L6 · Compute · network GPU · ASIC · NVLink · Tomahawk NVDA · AVGO · (INTC short) $3M+ /MW (compute) L5+L3 · Foundry · packaging N3/N2/A16 · CoWoS-L · EMIB TSM · (INTC short) chokepoint L4 · L2 · L1 · L0 Memory · EDA · WFE · materials — gaps · ASML & SK Hynix highest-priority next adds ← DOLLAR-VOLUME CENTER OF MASS ← HIGHER MARGIN PER DOLLAR
Source · synthesis.md §2 + portfolio-summary.md §3 layer-coverage. $/MW figures from refinement-log Findings 8 + 11 (ETN $500K–$1.1M today → $1.0–2.0M Kyber era; VRT $600K–$1.3M today → $1.2–2.8M Kyber era).
§ 03

Top 5 catalysts on the calendar.

Next 12 months · ranked by impact
#DateEventNamesDirection
012026-04-02ITC FD · Infineon v. Innoscience (337-TA-1407). Single most important GaN-IP regime event in 2026. Base case ~70%: Innoscience prevails (cost-floor amplification). Surprise Infineon win shifts cohort GaN moat read materially positive.NVTS · TXN · IFXBinary
022026-09-13MOFCOM antidumping FD on TXN. Bernstein sizes ~11.4% revenue exposed; we estimate 3–6% at meaningful adverse risk. Largest near-term asymmetric downside in any cohort long. Catalyst-aware sizing: TXN is a Q1–Q3 long defended through Q4 binary.TXNBearish-tail
03Q2 2026Eaton Q2 earnings. First full Boyd quarter consolidated; H1 ramp headwind resolution; GOES disclosure; Mobility spin Form 10 timeline. Conviction-shaper.ETNBinary
04Q2 2026 (~Jul)Vertiv Q2 EMEA print. Second geographic data point after Q1 −20% YoY. Two more negative quarters confirm structural Schneider share loss in second-largest market.VRTBinary
052026-06EU F-Gas implementing act. Refrigerant transition timelines, R1234ze production requirements. Bearish for VRT cooling-ramp timeline (HFO chokepoint analog of GOES).VRTBearish
§ 04

Read the deck.

Cohort views
§ 05

The names.

9 deep-dive theses
"Power is no longer a supply problem. It has become a question of how to redesign the entire voltage stack from the power plant to the die."synthesis.md · §3.1