AlphaOS

What are the key dependencies in the AI chip supply chain?

AlphaOS investment intelligence · Research and education only — not investment advice · Updated Sep 27, 2026

About artificial-intelligence

Direct answer

The AI chip supply chain depends on a concentrated set of chokepoints spanning chip design, advanced fabrication, specialized packaging, and critical materials. NVIDIA dominates AI GPU design with approximately 80% data center GPU market share, but all leading-edge AI chips — including those from AMD, Intel, Google, and custom designs from AWS, Microsoft, and Meta — rely on TSMC for manufacturing at 3nm and 5nm nodes. TSMC controls roughly 90% of advanced node foundry capacity globally. Downstream, advanced packaging (CoWoS, HBM integration) creates additional bottlenecks, with TSMC and SK Hynix as primary suppliers. ASML holds a monopoly on EUV lithography machines required for sub-7nm fabrication, and rare earth materials including silicon carbide and specialty gases introduce further geopolitical and geographic concentration risks.

This week

Companies with new signals this week

METAgrowth ranks with an opportunity score of 90, driven primarily by growth signals.90MXLgrowth ranks with an opportunity score of 89, driven primarily by growth signals.89

2 of 24 shown

See the full board →

Key Takeaways

  • TSMC controls ~90% of advanced node (sub-7nm) chip fabrication capacity, making it the single most critical chokepoint in the entire AI chip supply chain
  • NVIDIA holds ~80% share of the data center GPU market; its H100/H200 and Blackwell chips are the de facto standard for AI model training and inference
  • ASML is the sole supplier of EUV lithography machines, which are required to manufacture leading-edge chips — no substitute exists, giving it irreplaceable leverage
  • SK Hynix and Micron supply High Bandwidth Memory (HBM3/HBM3E), which is stacked directly onto AI GPUs; HBM capacity constraints have repeatedly bottlenecked H100 supply
  • Advanced packaging — specifically TSMC's CoWoS (Chip-on-Wafer-on-Substrate) — has been a persistent supply constraint, limiting AI chip output independent of wafer availability
  • Geopolitical risk is acute: TSMC's primary fabs are in Taiwan, and U.S. export controls restrict advanced chip sales to China, affecting ~20-25% of NVIDIA's prior revenue base
  • Specialty chemical and gas suppliers (including Air Products, Shin-Etsu Chemical) and substrate manufacturers (Ibiden, Shinko Electric) represent less visible but critical upstream dependencies
  • Hyperscalers including Google (TPUs), Amazon (Trainium/Inferentia), and Microsoft (Maia) are developing custom silicon to reduce dependency on NVIDIA, though TSMC reliance remains

Evidence & Analysis

  • TSMC accounts for approximately 90% of global advanced node (sub-5nm) semiconductor production capacity as of 2024, with Samsung holding most of the remainder
  • NVIDIA's H100 GPU shortages in 2023 were partly caused by CoWoS advanced packaging capacity constraints at TSMC, not wafer supply alone
  • SK Hynix supplies over 50% of global HBM capacity; HBM3E lead times extended to 12-18 months at peak 2023-2024 AI demand
  • ASML shipped 374 EUV systems in 2023 with a current backlog extending into 2026; each machine costs approximately $150-200 million and takes over a year to build
  • U.S. Bureau of Industry and Security export controls introduced in October 2022 and expanded in October 2023 restricted NVIDIA's A100, H100, and equivalent chips from China — a market representing roughly $4 billion in annualized prior revenue
  • TSMC's Arizona fabs (Fab 21) are targeted for 2nm production by 2026 but represent a small fraction of total capacity versus Taiwan-based facilities

Key Companies

NVDA

NVIDIA Corporation

Primary AI chip designer — ~80% data center GPU share; H100, H200, and Blackwell architectures are the industry standard for AI training

TSM

Taiwan Semiconductor Manufacturing Company

Sole manufacturer of leading-edge AI chips; controls ~90% of sub-7nm foundry capacity; CoWoS packaging is a persistent bottleneck

ASML

ASML Holding N.V.

Global monopoly on EUV lithography machines; indispensable for fabricating all leading-edge AI chips at 7nm and below

000660.KS

SK Hynix

Leading supplier of HBM3/HBM3E memory stacked on NVIDIA GPUs; HBM capacity is a direct AI chip supply constraint

AMD

Advanced Micro Devices

Second-largest AI GPU designer; MI300X competes with NVIDIA H100; also relies exclusively on TSMC for advanced node manufacturing

Related stock research

Connected companies and research pages from the AlphaOS knowledge graph.

Priority stock research

High-intent stock intelligence pages — connected from AlphaOS research hubs.

Related stock research

Structured stock intelligence for companies connected to this research.

Related Questions

Generated by AlphaOS from the Knowledge Graph, earnings intelligence, and industry analysis. Content is for research and education only — not investment advice.