Skip to content

P1-C6 · 5 Roles of the Supply Chain + 60 Ticker Map

Core One-Liner

Mapping the supply chain = seeing how money flows. With C1-C5 as foundation, this map is no longer isolated tickers, it's cause and effect.

AI Industry Knowledge — History → Technology → Supply Chain → Business → Application → Geopolitics

P1-C6 (Part 1, Chapter 6). After this chapter, you can place any AI stock into one of the 5 supply chain roles and explain why it's there (historical + technical reason), not just memorize.

1. The Problem: You Look at a Supply Chain Map with 60 Tickers and Still Can't Recognize Them All

You've already learned history / technology / hardware in C1-C5. Now look at the supply chain:

  • ASML / TSM / SK Hynix / NVDA / SMCI / VRT / CEG / MSFT / OpenAI ...
  • 60+ tickers: which are upstream, which are downstream, which are "selling shovels," which are "panning for gold"?

Memorization is useless. What you need is a 5-role map + where each ticker fits + why it's there (derived from C1-C5).

2. The Solution: 5-Role Framework

💡 Each ticker is clickable → go to its Multi-Source Profile to see supply chain coordinates / upstream-downstream / key data.

Role What They Do Representative Tickers C1-C5 Explanation Why
Upstream Shovels for shovel sellers (equipment/materials) ASML · AMAT · LRCX · SNPS · CDNS · SK Hynix · MU · Samsung Physical bottlenecks → who owns that link (C5 §3.2 HBM)
Midstream Shovel sellers (accelerators/network/optics) NVDA · AMD · AVGO · COHR · LITE · ANET · MRVL NVDA's 20-year build (C3) + network bottleneck (C5 §3.3)
Downstream Data centers / Cloud (shovel buyers) MSFT · GOOGL · AMZN · ORCL · CRWV · NBIS Hyperscaler vs Neocloud (C5 §3.4 + Application C8)
Customer AI labs (compute users) OpenAI · Anthropic · xAI · Mistral · DeepSeek Positioning (C3 OAI vs Google) + Application (C8)
Support Power / Cooling / Real Estate CEG · VST · VRT · ETN · HUBB · GEV · EQIX · DLR Energy bottleneck (C5 §3.4)

Key insight: In the 5 AI industry roles, the further upstream, the deeper the moat (equipment/material monopoly), the further downstream, the shallower the moat (cloud/SaaS high substitutability). This directly relates to your Layer 2 valuation (detailed in C7).

3. How It Works: Simplified Supply Chain Diagram + Key Dependencies

graph LR
    %% Upstream
    ASML[ASML EUV] --> TSM[TSM Wafers]
    AMAT[AMAT/LRCX Equipment] --> TSM
    SNPS[SNPS/CDNS EDA] --> NVDA[NVDA Design]
    SKHynix[SK Hynix HBM] --> NVDA
    MU[Micron HBM] --> NVDA
    TSM --> NVDA

    %% Midstream
    NVDA --> Hyperscaler[MSFT/GOOGL/AMZN]
    NVDA --> Neocloud[CRWV/NBIS/ORCL]
    COHR[COHR/LITE Optical Modules] --> NVDA
    ANET[ANET Networking] --> Hyperscaler

    %% Downstream + Customer
    Hyperscaler --> OpenAI[OpenAI]
    Hyperscaler --> Anthropic[Anthropic]
    Neocloud --> OpenAI

    %% Support
    CEG[CEG/VST Power] --> Hyperscaler
    VRT[VRT/ETN Electrical Liquid Cooling] --> Hyperscaler
    VRT --> Neocloud

    classDef upstream fill:#fef3c7,stroke:#d97706
    classDef midstream fill:#dbeafe,stroke:#2563eb
    classDef downstream fill:#dcfce7,stroke:#16a34a
    classDef customer fill:#fee2e2,stroke:#dc2626
    classDef support fill:#fce7f3,stroke:#db2777

    class ASML,AMAT,SNPS,TSM,SKHynix,MU upstream
    class NVDA,COHR,ANET midstream
    class Hyperscaler,Neocloud downstream
    class OpenAI,Anthropic customer
    class CEG,VRT support

5 Transmission Rules (you'll use these repeatedly in your thesis):

  1. NVIDIA is not isolated — it depends on TSM foundry + SK Hynix HBM + COHR optical modules + SNPS/CDNS design software. Any link breaks → NVDA is affected. (e.g., Samsung strike → HBM shortage → NVDA H200 shipments constrained)
  2. AI labs are not direct customers — OpenAI buys compute from MSFT Azure, MSFT then buys GPUs from NVDA. So "hyperscaler capex" in NVDA's earnings is key, not "OpenAI revenue."
  3. Neocloud (CRWV / NBIS / ORCL OCI) differs from Hyperscaler — they only do AI compute, not general cloud. Highly concentrated customers (CRWV 60%+ MSFT, ORCL 54% OpenAI). This is a valuation premium but also a RISK.
  4. Power is the real bottleneck — AI data centers consume enormous electricity. Not just hardware, but also CEG (nuclear PPA) / VST / GEV (gas turbines) / VRT (liquid cooling) — all "shovel sellers."
  5. Supply chain upstream-downstream transmits — learning to track upstream + downstream = a complete thesis.

4. vs C5 What You Already Know

Dimension C5 Gives You C6 Adds
Hardware physics ✓ Doesn't map to companies
Company map ✗ 5 roles + 60 tickers + causality (why in that role)
Investment meaning Know bottlenecks Know which company corresponds to which bottleneck — see news "HBM shortage" and immediately know SK Hynix / Micron benefit, NVDA shipment cap

C5 = physics. C6 = company mapping. Without C6, you see news but don't know which stock to focus on.

5. Try It: Pick 1 Edge, Answer "Why Them"

Task (15 minutes): Pick 1 of the 3 edges below, use C1-C5 knowledge to answer "Why not someone else":

Edge Question
NVDA ← SK Hynix HBM (70%+ share) Why not Samsung? (Hint: qualification difficulty + strike + yield)
TSM ← ASML EUV (100% share) Why not Canon / Nikon? (Hint: 25 years R&D + physical optics difficulty)
MSFT Azure ← OpenAI (exclusive through 2024) Why not Google? (Hint: C3 startup vs incumbent mindset)

Self-check (3 items met → proceed to P1-C7):

  • You can explain why ASML is the ultimate moat (single company, physical bottleneck)
  • You can predict "If Samsung HBM3e suddenly passes NVDA qualification" which 3 stocks move (SK Hynix - / Micron - / Samsung +)
  • You can say in one sentence the moat source for each supply chain link (upstream = physics, midstream = ecosystem, downstream = scale)

6. What's Next

You've mapped the supply chain. But how much each link earns varies hugely — ASML 50% gross margin, AMD 12%. Who is the true king, who is a passerby?

→ P1-C7 · Business Model + Value Capture 5-dimension scoring for each link — moat source + profit margin + switching cost.

7. Deep Dive (optional): Atlas 1643 LLM-mined Articles / Cross-Border Dependency / China Domestic Substitution

Click to expand deep supply chain version

Atlas (1643 LLM-mined articles): The edu_site now has supply_chain_atlas.md — from SemiAnalysis / Stratechery / The Information's 1643 articles, LLM extracted 269 edges with ≥2 citations. 10x more detailed than the manual 30-edge version. Advanced users, check this.

Cross-border dependency: NVDA → TSM (Taiwan) → ASML (Netherlands) — every link crosses borders. Any shock to Taiwan Strait / China-EU relations → NVDA shipments affected. This is a macro overlay your thesis cannot ignore.

China domestic substitution (2023+): After H100 banned from China, Huawei Ascend 910C + SMIC 7nm + YMTC HBM emerged. China market splits in two (domestic Huawei / foreign NVDA). NVDA loses ~25% China revenue, but other markets compensate.

→ This "market segmentation" is the geopolitics (C9) main thread. Any ticker with China exposure in your thesis must consider this.

New roles emerging (2026+): - AI Real Estate: EQIX / DLR / IRM (data center REITs) - AI Legal/Compliance: Harvey / Lex Machina (vertical SaaS) - AI RegTech: SAS / Palantir (compliance models) - AI Data Labeling: Scale AI / Surge (RLHF data)

These new ecosystem layers are still forming, investing is in an early window.