Read · February 20, 2026 · 8 min read

Wave 1: The 'Stacking' Peak

Strategy: Invest in the CREATION of the chip.

Closelooknet AI Buildout Series Wave 1 · Stacking Peak February 2026

Wave 1: The 'Stacking' Peak

Strategy: Invest in the CREATION of the chip.

Thomas Look, Alpha Capo · closelook.io

Wave 1 Thesis

The dominant constraint in 2026/27 is physical manufacturing and memory assembly. The 16-layer HBM4 stack is the primary bottleneck — and the companies solving it see peak revenue correlation before NVIDIA even ships.

Rubin's architecture requires 288GB+ of HBM4 per GPU, packed into 16-layer stacks on 3nm silicon, cooled by 120kW liquid systems per NVL72 rack. Every one of those specifications is a physics problem. And physics problems create investable bottlenecks that get paid two quarters before NVIDIA books revenue.

This is the era of physical hurdles — where the constraint isn't demand for GPUs but the ability to physically manufacture, bond, stack, inspect, and cool them. The companies in this wave are leading indicators. Their order books fill before NVIDIA's earnings calls. Their capex cycles predict NVIDIA's revenue trajectory. If you're timing the AI infrastructure cycle, this is where the clock starts.

Leading Indicator Alert

Tier III manufacturing equipment companies — SUSS MicroTec, Tokyo Electron, Camtek — get paid 2 quarters before NVIDIA ships. Their revenue is the earliest investable signal of NVIDIA's production ramp.

The Shadow Portfolio · First Draft

The Full Shadow Portfolio

Below is the complete 12-company first-draft portfolio ranked by correlation coefficient. Wave 1 beneficiaries are highlighted — these are the stocks with peak revenue realization in the 2026/27 window.

Company	Tier	Correlation	Time Zone	Wave 1
TSMC	I	0.98	Linear (2026–2030)	✦
SK Hynix	I	0.94	2026/27 Peak	✦
Astera Labs	II	0.92	2026/27 Peak	✦
Vertiv	III	0.91	2026–28 High	✦
SUSS MicroTec	III	0.90	2026 Leading	✦✦
ARM Holdings	I	0.90	2026/27 Peak	✦
Broadcom	II	0.88	Linear
Tokyo Electron	III	0.88	2026/27 High	✦
Micron	I	0.88	2028/29 Peak
Camtek	III	0.87	Linear	✦
Advantest	III	0.86	2028/29 High
Marvell	II	0.86	2028/29 Peak
Supermicro	I	0.85	2026/27 Peak	✦
Modine	III	0.84	2028–30 High
Credo Technology	II	0.84	Linear
Fabrinet	II	0.83	2028/29 Peak
Samsung	I	0.82	2030 High
Amphenol	II	0.81	Linear
Dell Technologies	I	0.80	2028–30 High
Corning	II	0.78	2030 High

The table above captures the full 20-company ecosystem. Note the expanded list from the original 12-company first draft — including Supermicro, Modine, Credo, Fabrinet, Samsung, Amphenol, Dell, and Corning from the presentation's Tier II and III breakdowns. Each has a specific role in the physical supply chain, a measurable correlation to NVIDIA's data center revenue, and a time zone indicating when their contribution peaks.

Wave 1 Deep Dives · Highest Correlation

Tier I: Intelligence & Compute Core

The monopoly manufacturers and design architects. These companies sit at the foundation of every GPU — without them, nothing gets built.

TSMC

TSM

0.98

Correlation

Tier I

Classification

Linear · 2026–2030

Role Monopoly Manufacturer

Why Every Rubin (3nm) and Feynman (2nm) chip is a TSMC wafer. There is no alternative supplier at these process nodes.

Metric CoWoS (Chip-on-Wafer-on-Substrate) advanced packaging capacity is the binding constraint. Track CoWoS capacity announcements for lead signals.

Type Type A — Linear Tax. Revenue scales 1:1 with every GPU shipped across all three waves. The highest-conviction, lowest-timing-risk position in the ecosystem.

SK Hynix

000660.KS

0.94

Correlation

Tier I

Classification

2026/27 Peak

Role HBM4 Market Leader

Why Primary yield leader for 16-layer HBM4 stacks. Rubin requires 288GB+ per GPU. HBM4 constitutes approximately 40% of rack BOM — by far the largest single component cost.

Type Type B — Step-Function. Revenue spikes during the Rubin transition as HBM4 demand exceeds all prior memory generations. Peak correlation in this cycle, before Micron scales as the release valve in Wave 2.

ARM Holdings

ARM

0.90

Correlation

Tier I

Classification

2026/27 Peak

Role Architecture Royalty

Why The Vera CPU inside every NVIDIA Superchip is ARM-based. Every unit shipped pays a royalty. This is the purest per-unit "tax" in the ecosystem.

Type Type A — Linear Tax. Royalty revenue scales with volume. Peak correlation during initial Rubin ramp when per-unit royalty rates are set for the architecture generation.

Tier I: Systems & Integrators

Supermicro

SMCI

0.85

Correlation

Tier I

Classification

2026/27 Peak

Role Speed-to-Rack

Why Specializes in liquid-cooled NVL72 cabinets. Captures the initial "Factory" build-out phase when hyperscalers race to deploy Rubin clusters.

Type Type B — Step-Function. Revenue spikes with each new rack architecture, then normalizes as Dell captures the sovereign and enterprise tail.

Tier II: Signal & Fabric

Astera Labs

ALAB

0.92

Correlation

Tier II

Classification

2026/27 Peak

Role Retimers — PCIe 6.0 / CXL

Why Essential for signal integrity in Rubin racks. As data rates increase with PCIe 6.0, signal degradation over copper traces becomes a hard physics problem. Astera's retimers restore signal fidelity at each hop — no retimer, no functional rack.

Type Type B — Step-Function. Each new PCIe generation and rack architecture requires updated retimer silicon. Astera rides the Rubin ramp, then faces transition risk at Feynman if photonics disintermediates electrical retiming.

Tier III: Manufacturing Equipment

These are the earliest leading indicators in the entire ecosystem. Manufacturing equipment gets ordered and paid for two quarters before NVIDIA ships product. If you're watching for inflection points, this is where the signal appears first.

SUSS MicroTec

SMHN.DE

0.90

Correlation

Tier III

Classification

2026 Leading Indicator ⚡

Role Temporary Bonding Equipment

Why The 'Glue' allowing HBM4 wafers to be thinned and stacked into 16-layer configurations. Without temporary bonding, you cannot manufacture HBM4 at scale. Period.

Signal Highest leading indicator in the ecosystem. SUSS order books fill before SK Hynix ramps HBM4 production, which happens before NVIDIA ships Rubin. This is the earliest investable signal.

Risk Highest volatility in the portfolio. Small-cap German stock with concentrated revenue exposure. The correlation is real but the position sizing must reflect the risk.

Tokyo Electron

8035.T

0.88

Correlation

Tier III

Classification

2026/27 High

Role 3D Stacking / Etching Tools

Why Builds the physical 16-layer stacks. TEL's etch and deposition tools create the through-silicon vias (TSVs) and bonding surfaces that make HBM4 stacking possible.

Type Type B — Step-Function. Revenue correlation peaks during the Rubin manufacturing ramp. More diversified revenue base than SUSS MicroTec, providing a lower-volatility way to play the same thesis.

Camtek

CAMT

0.87

Correlation

Tier III

Classification

Linear

Role Optical Inspection

Why The 'Yield Tax' on complex 3D chiplets. As stacking layers increase, inspection requirements grow exponentially. Every additional layer must be verified before the next is bonded — a defective layer at position 8 of 16 destroys the entire stack.

Type Type A — Linear Tax. Inspection is required across all architectural generations. Revenue correlation remains steady from Rubin through Feynman, making this a through-cycle hold.

Tier III: Thermal & Cooling

Vertiv

VRT

0.91

Correlation

Tier III

Classification

2026–2028 High

Role Liquid Cooling — Coolant Distribution Units (CDUs)

Why Liquid cooling is now mandatory, not optional. The 120kW per rack heat load of NVL72 systems cannot be managed by air cooling. Vertiv's CDUs are the thermal backbone of every Rubin deployment.

Type Type B — Step-Function. The air-to-liquid transition is the step change. Once deployed, Vertiv shifts to a maintenance and expansion revenue model. Peak correlation spans both Wave 1 and early Wave 2.

Wave 1 Summary

The Wave 1 Playbook

Top Wave 1 Correlations

1. SUSS MicroTec — Bonding Equipment (earliest signal, highest volatility)

2. Tokyo Electron — Etching & Stacking Tools

3. SK Hynix — HBM4 Delivery (40% of rack BOM)

4. Astera Labs — Internal Rack Signal Integrity

The Wave 1 portfolio concentrates on companies solving the "Stacking Peak" — the physical challenge of manufacturing 16-layer HBM4 memory and assembling it into functional Rubin racks. The leading indicators (SUSS MicroTec, Tokyo Electron, Camtek) see revenue two quarters ahead of NVIDIA. The primary beneficiary (SK Hynix) captures 40% of every rack's bill of materials. The signal integrity layer (Astera Labs) ensures the rack functions once assembled. And the thermal solution (Vertiv) ensures it doesn't melt.

Through-cycle holds — TSMC, ARM, Broadcom, and Camtek — are Type A "tax" companies whose revenue correlates across all three waves. These form the portfolio's stable core while the Wave 1 step-function plays generate the concentrated alpha.

The 'NVIDIA Multiplier' defines the real investment opportunity. Do not fight the physics.

Next in this series: Wave 2 — The Photonics Peak (2028/29) covers the shift from making chips to connecting them, when Marvell, Fabrinet, Advantest, and Micron see peak correlation as light replaces copper.

This is investment research, not financial advice. Correlation scores are author estimates based on supply chain analysis and BOM modeling. Past performance does not predict future returns. The author may hold positions in securities discussed.