The 6-Layer Model: How Closelook Maps the AI Supply Chain
Closelook organizes the entire AI infrastructure supply chain into six functional layers: Silicon (lithography, fabrication, packaging), Memory (HBM, DDR5, storage), Networking (interconnects, switches, optical), Compute (GPUs, custom ASICs, CPUs), Data Platforms (cloud infrastructure, databases, observability), and End Markets (enterprise AI, consumer AI, robotics). Each layer has distinct competitive dynamics, bottleneck risks, and investment characteristics. The 6-Layer Model is the structural backbone of the Functional Index — every constituent is classified by its primary layer. (Note: in the 2026-06-30 reconstitution the Rubin index expanded to an eight-layer supply-chain framework, adding an AI Factory and a Physical AI layer; this six-layer model remains a useful conceptual primer, and later structural work may supersede it.)
Why Layers Matter for Investors
Most investors think of "AI stocks" as a single category. But the AI supply chain is a deep, layered structure where each layer has different cycle dynamics, margin profiles, and bottleneck risks. Silicon (Layer 1) is capital-intensive with long lead times. Memory (Layer 2) is highly cyclical with pricing power swings. Compute (Layer 4) is where NVIDIA dominates but custom ASICs are emerging. Understanding which layer you're investing in determines your risk profile.
The Functional Index tracks all six layers independently. When Layer 1 (Silicon) outperforms Layer 4 (Compute), it signals that the supply chain is building ahead of demand — a healthy leading indicator. When Layer 4 outperforms alone, it suggests demand concentration without broad supply chain confirmation.
The Six Layers
Layer 1 — Silicon: ASML, TSMC, Applied Materials, Lam Research, BESI. Lithography, fabrication, packaging. The physical foundation. Longest lead times, highest capital intensity.
Layer 2 — Memory: Micron, SK Hynix, Samsung Memory. HBM, DDR5, storage. Most cyclical layer. HBM pricing is the key signal for AI demand health.
Layer 3 — Networking: Broadcom, Arista, Marvell, Coherent. Interconnects, switches, optical. The connective tissue between compute nodes. Increasingly bottlenecked as cluster sizes grow.
Layer 4 — Compute: NVIDIA, AMD, Intel, Google TPU, AWS Trainium. GPUs, custom ASICs, inference accelerators. The most visible layer but not the only one that matters.
Layer 5 — Data Platforms: Snowflake, Datadog, Confluent, Elastic. Cloud infrastructure, databases, observability. The software layer agents and AI applications run on.
Layer 6 — End Markets: Tesla, Apple, Meta, enterprise deployers. Where AI meets the real world. Demand-side indicator for the entire stack.