§ cohort  ·  Landscape  ·  Foundry · GaN · Cooling · Power-infra
Filed 2026-05-04  ·  from competitor.md fan-outs · refinement-log Findings 1, 5, 6, 14

Four maps. One thesis per map.

The cohort sits inside four distinct competitive structures: foundry leadership (3-player), GaN four-way race (corrected from prior three-way), cooling/thermal three-player (post-M&A), and power-infra electrical six-player. Each map is annotated with cohort positions and the strategic trajectory.

§ 01

Foundry leadership · process node × yield.

TSM · INTC · Samsung — 3-player

Cost per transistor flat since N7 and rising at N2/A16. Density now comes from advanced packaging (CoWoS, SoIC, EMIB, Foveros). "TSMC's 'process lead' is now significantly an 'advanced packaging lead.'" INTC 14A binary determines whether Intel rejoins the race or is permanently sidelined. Samsung yield problems since SF3.

Trailing-edge N5 / N4 N3 / N2 A16 / 14A A10 / High-NA 100% yield struggling YIELD PROCESS NODE LEADERSHIP → TSM TSMC · 90% leading-edge CoWoS-L · A16 H2 2026 Samsung Foundry SF2P yield improving · Tesla / Qualcomm pursuit INTC 18A external book press-release-grade 14A is binary · 2027 target SMIC N7 with multi-patterning · DUV-only GF UMC High-NA EUV adoption → 14A "if it works" Mature · disciplined Leadership zone Trailing-edge Leading-edge struggle Foundry leadership · process node × yield
TSMC: ~90% leading-edge logic share · A16 H2 2026 (first node with backside power) · CoWoS-L is the binding AI shipment chokepoint. INTC: 14A is binary — "if it works, Intel is back. If it doesn't, permanently sidelined." Samsung: SF3 yield problems persist; pursuing Tesla/Qualcomm externally. SMIC: parallel stack at N7 with DUV multi-patterning.
§ 02

GaN four-way race · device share × strategy.

CORRECTED · Innoscience #1 was missing from prior synthesis
⚠ Correction · refinement-log Finding 1  ·  The prior synthesis's "three-way GaN race" frame (Infineon scale / TI vertical / Navitas density) was incomplete. Innoscience runs 8" GaN-on-Si IDM at scale, has ~30% global share (#1), and is the sole Chinese partner on NVIDIA's 800VDC list.

Four-way structure: Innoscience #1 (~30% share, IDM cost-floor) · NVTS #2 (~17%, density bet, GaN-IC patent estate) · POWI (~17%) · EPC (~12–15%) · IFX (~10%, scale, 300mm 2025) · TXN (~5–7%, system-BOM bet not device-share). The April 2 2026 ITC FD in Infineon v. Innoscience is the dated GaN-IP regime catalyst.

5% 10% 17% 25% 30%+ discrete FET system BOM STRATEGY DEVICE SHARE → ~30% Innoscience 8" GaN-on-Si IDM NVIDIA 800VDC partner cost-floor risk ~17% NVTS density bet · GaN-IC patents single-stage 800V→6V ~17% POWI ~12% EPC ~10% IFX scale · 300mm 2025 5–7% TXN system-BOM bet NVIDIA GTC silicon-anchor if Innoscience moves up to GaN-IC, NVTS moat erodes stays at L8b–L8c+L7+L6 · system-BOM GaN four-way race · device share × strategic positioning April 2 2026 ITC FD = dated GaN-IP regime catalyst cohort long competitor watchlist private (Innoscience)
Source · refinement-log Finding 1 (C-NVTS-1 fan-out · Innoscience #1 omission corrected) + Finding 5 (C-TXN-1 · TXN's bet is system-BOM not device-share). Disk size proportional to share. The risk: Innoscience moving up from discrete commodity GaN to integrated GaN-IC products at 8" wafer economics — the NVTS kill scenario.
§ 03

Cooling / thermal · three-player consolidation by 2027.

VRT incumbent · Boyd-via-ETN · Ecolab/CoolIT · post-March 2026

Until March 2026, VRT held unchallenged CDU market leadership at ~22% share. Two simultaneous M&A events changed the structure permanently: ETN closed Boyd Thermal ($9.5B) and Ecolab acquired CoolIT ($4.75B). VRT's cooling moat compressed from wide to narrow, entering the highest-demand window of the AI-DC cycle.

CDU market structure · pre vs post March 2026 Pre-March 2026 VRT incumbent VRT ~22% CDU fragmented field · Asetek · Boyd · CoolIT · etc M&A · Mar 2026 ETN buys Boyd $9.5B Ecolab buys CoolIT $4.75B By 2027 · 3-player race moat narrowed VRT incumbent ETN/Boyd $1.7B rev · 80% DC Ecolab/ CoolIT Three well-capitalized competitors entering highest-demand window of AI-DC cycle Monitoring trigger 5+ ppt VRT CDU share loss over 2 consecutive quarters → trim to sub-1% NAV Liquid cooling penetration: 33% of new AI-DC racks (2025) → 50–70% (2030). Physics-mandatory above 30kW/rack. 800V architecture mandates 100% liquid cooling at 45°C inlets.
Source · refinement-log Findings 6 + 14. ETN closed Boyd Thermal $9.5B / $1.7B revenue at 80%+ DC mix March 2026. Ecolab acquired CoolIT $4.75B March 2026. The Uptime Institute reports cooling systems accounted for 13% of datacenter failures in 2024 — failure-consequence reality drives incumbent pricing power.
§ 04

Power infrastructure · electrical six-player.

ETN · VRT · Schneider · ABB · Siemens · Delta — diversification × AI-DC concentration

The chip-to-grid electrical layer at G1–G3. Two cohort longs (ETN, VRT) versus four watchlist competitors. The single most important relative-positioning observation: VRT is a pure-play (~80% AI-DC) at 53.4× EV/EBITDA · Schneider is a diversified industrial at ~22× EV/EBITDA — the multiple gap is 143% in VRT's favor and is the entire valuation question for the cohort.

5% 15% 30% 55% 80% 55× EV/EBITDA 35× 22× 12× MULTIPLE % AI-DC REVENUE MIX → VRT 53.4× EV/EBITDA · ~80% AI-DC 91% premium to ETN ETN 27.9× · 16–19% AI-DC → 22–26% post-Mobility spin Schneider ~22× · ~24% AI-DC 2028–2030 real impact ABB ~21× Siemens ~18× Delta overtook Foxconn Jan 2026 ETN post-Mobility spin Q1 2027 Schneider 800V ramp 2028–2030 VRT vs Schneider: 143% multiple premium Power-infra electrical · valuation × AI-DC concentration
Source · VRT thesis §6 valuation comparison; ETN thesis §7 multi-cycle cushion; portfolio-summary §4 Pair 3 (VRT-Schneider relative-value consideration). VRT 53.4× EV/EBITDA = 91% premium to ETN, 143% premium to Schneider. The valuation gap is the right gap for the risk-profile gap (VRT pure-play vs Schneider diversified) — but it leaves no margin for execution slip.
§ 05

Read across the four maps.

cohort positioning · synthesis

Foundry leadership is the cleanest divergence in the cohort: TSM owns ~90% leading-edge logic; INTC is the binary 14A bet at the opposite corner. The TSM long / INTC short pair is the cleanest structural-foundry expression — purely process-node, not Section-232-mediated.

GaN four-way race is structurally different: NVTS bets on density-and-IC-architecture lock-in; TXN bets on system-BOM capture (without needing device-share win). The April 2 2026 ITC FD is the dated catalyst for the IP regime. Innoscience moving up from discrete to GaN-IC products is the NVTS kill scenario.

Cooling consolidation is the cohort's most time-sensitive competitive monitoring requirement. Three well-capitalized competitors entering the highest-demand window of the AI-DC cycle. VRT's cooling-margin expansion is the load-bearing support for the 53.4× EV/EBITDA multiple — and the moat just narrowed.

Power-infra electrical is the cohort's most direct relative-value choice: VRT pure-play premium vs Schneider diversified discount. The 143% multiple premium implies AI-DC sustained at $500B+/year through 2027. SU.PA on the bench is the cohort's "AI-DC infrastructure with lower multiple risk" expression.