AI Supply Chain
Physical infrastructure powering artificial intelligence
Reference data from industry roadmaps — updated periodically
AI Supply Chain Stress Index (ASCSI)
Concentration Risk Scorecard
| Layer | Dominant Supplier | Share | HHI | Risk |
|---|---|---|---|---|
| EUV Lithography | ASML | 100% | 10,000 | |
| Advanced Packaging | TSM, ASX | ~95% | 9,025 | |
| Fabrication | TSM, INTC | ~90% | 8,100 | |
| EDA & Design | SNPS, CDNS, ARM | ~55% | 5,000 | |
| Server / Rack Integration | NVDA, AMD, SMCI, DELL | ~70% | 5,000 | |
| Memory / HBM | MU | ~50% | 4,100 |
HHI (Herfindahl-Hirschman Index): >2500 = highly concentrated. Single points of failure highlighted.Run cascade simulation →
AI Hardware Stack
Electronic design automation tools and chip architecture IP. Synopsys and Cadence duopoly; Arm licenses cores used in most mobile and growing DC silicon.
Leading-edge semiconductor manufacturing. TSMC controls ~90% of advanced node (<7nm) production. Intel Foundry and Samsung compete but trail in yield and volume.
2.5D/3D chip packaging including TSMC CoWoS and InFO. CoWoS is THE bottleneck for AI accelerators — every H100/B200 requires a CoWoS interposer.
High Bandwidth Memory stacking for AI GPUs. SK Hynix leads HBM3e supply (~50% share), Micron ramping, Samsung catching up. Capacity constrained through 2025.
High-speed optical transceivers, switches, and interconnects for GPU clusters. 800G→1.6T transition driving demand for pluggable and CPO solutions.
GPU server assembly, liquid cooling, and rack-scale systems. Supermicro, Dell, and hyperscaler ODMs compete on density and thermal management.
Physical data center construction and power delivery. Hyperscalers (MSFT, GOOGL, AMZN, META) building at unprecedented scale; power availability is the binding constraint.
Data Centers
Hyperscaler capex, GPU shipments, and power demand
View Data Centers →
Supply Chains
17 commodity chains with cascade simulation
View Supply Chains →
Rare Earths
Critical minerals for magnets, electronics, and defense
View Rare Earths →
Methodology & Data Sources
Supply chain layer structure, bottleneck assessments, lead times, and capacity utilization estimates are reference data compiled from industry reports (SIA, SEMI, public filings). GPU specs from NVIDIA datasheets. TSMC node data from public earnings calls and roadmap presentations.
Live overlays: Equity prices (Yahoo Finance, 5min via cron/equities), FRED macro series (daily via cron/fred), Disruptions (AI-classified from news, every 4h). Trade flow data from UN Comtrade.