AI Industry Chain Panorama¶

📊 Read This Page Before Atlas

This page is an introductory overview (60+ tickers × hand-curated 30 key edges). After reading, proceed to the advanced 🆕 Industry Chain Atlas — LLM-mined from 1643 SemiAnalysis + Stratechery articles, yielding 269 edges with ≥2 citations, including specific what flows + citation counts.

The AI industry is like a pyramid: at the top are AI labs like OpenAI/Anthropic, below them cloud providers offering compute, compute relies on NVIDIA GPUs, GPUs are manufactured by TSMC, TSMC uses ASML lithography machines, and ASML uses various lasers and components...

The diagram below shows the relationships among the 60+ tickers we track (edges with importance ≥ 0.7):

Simplified Industry Chain Diagram¶

graph LR
    %% Upstream
    ASML[ASML Lithography] --> TSM[TSM/Samsung Foundry]
    AMAT[AMAT/LRCX Equipment] --> TSM
    SNPS[SNPS/CDNS EDA] --> NVDA[NVDA/AMD/AVGO Chip Design]
    %% Memory
    SKHynix[SK Hynix HBM] --> NVDA
    MU[Micron HBM] --> NVDA
    %% Manufacturing
    TSM --> NVDA
    %% Network/Optical
    COHR[COHR/LITE Optical Modules] --> NVDA
    ANET[ANET Network Switches] --> Hyperscaler
    %% Cloud
    NVDA --> Hyperscaler[MSFT/GOOGL/AMZN Cloud]
    NVDA --> CRWV[CRWV/NBIS Neocloud]
    %% AI Labs
    Hyperscaler --> OpenAI[OpenAI]
    Hyperscaler --> Anthropic[Anthropic]
    CRWV --> OpenAI
    %% Supporting
    CEG[CEG/VST Power] --> Hyperscaler
    VRT[VRT/ETN/HUBB Electrical Infrastructure] --> Hyperscaler
    classDef upstream fill:#fef3c7,stroke:#d97706
    classDef midstream fill:#dbeafe,stroke:#2563eb
    classDef downstream fill:#dcfce7,stroke:#16a34a
    classDef support fill:#fce7f3,stroke:#db2777
    class ASML,AMAT,SNPS,TSM,SKHynix,MU upstream
    class NVDA,COHR,ANET midstream
    class Hyperscaler,CRWV,OpenAI,Anthropic downstream
    class CEG,VRT support

Layer Details (Relationships with importance ≥ 0.7)¶

Upstream (Equipment/Materials/Wafer/Memory)¶

31 relationships

Upstream	Relationship	Downstream	Importance	Description
SNPS	EDA Design Tools	NVDA	0.9	Synopsys EDA tools for all NVDA chip design.
CDNS	EDA Design Tools	NVDA	0.9	Cadence EDA tools for NVDA + most chip designers.
SNPS	EDA Design Tools	AMD	0.8	Synopsys for AMD.
CDNS	EDA Design Tools	AMD	0.8	Cadence for AMD.
SNPS	EDA Design Tools	AVGO	0.8	Synopsys for Broadcom.
CDNS	EDA Design Tools	AVGO	0.8	Cadence for Broadcom.
TSM	Wafer Foundry	NVDA	1.0	TSMC manufactures all NVDA datacenter GPUs (Hopper, Blackwell, Vera Rubin) on 4N/3N/2N nodes. NVDA i
TSM	Wafer Foundry	AAPL	1.0	TSMC manufactures all Apple A/M-series silicon.
TSM	Wafer Foundry	AMD	0.9	TSMC manufactures AMD EPYC + MI300 datacenter chips.
TSM	Wafer Foundry	AVGO	0.9	TSMC manufactures AVGO Google TPU + custom silicon.
TSM	Wafer Foundry	MRVL	0.8	TSMC manufactures Marvell DPU + custom AI silicon.
TSM	Wafer Foundry	QCOM	0.8	TSMC manufactures Qualcomm Snapdragon.
0981.HK	Wafer Foundry	BIDU	0.8	SMIC manufactures Baidu Kunlun AI chips (China domestic).
TSM	Wafer Foundry	META	0.7	TSMC manufactures Meta MTIA custom AI silicon.
TSM	Wafer Foundry	MSFT	0.7	TSMC manufactures Microsoft Maia 100 AI accelerator.
TSM	Wafer Foundry	AMZN	0.7	TSMC manufactures AWS Trainium2 + Inferentia (some are also Samsung).
ASML	Lithography / Equipment	TSM	1.0	ASML is sole supplier of EUV lithography to TSMC. Bottleneck for 3nm/2nm ramp.
ASML	Lithography / Equipment	005930.KS	0.9	ASML EUV to Samsung Foundry.
ASML	Lithography / Equipment	INTC	0.8	ASML EUV to Intel Foundry.
ASML	Lithography / Equipment	MU	0.8	ASML for DRAM/HBM lithography.
ASML	Lithography / Equipment	000660.KS	0.8	ASML for SK Hynix DRAM/HBM.
AMAT	Lithography / Equipment	TSM	0.7	Applied Materials supplies deposition + etch tools to all foundries.
AMAT	Lithography / Equipment	MU	0.7	AMAT tools to Micron DRAM/NAND.
LRCX	Lithography / Equipment	TSM	0.7	Lam Research supplies etch + deposition to all foundries.
LRCX	Lithography / Equipment	MU	0.7	Lam Research to Micron.
000660.KS	High Bandwidth Memory HBM	NVDA	1.0	SK Hynix is dominant HBM3e supplier to NVDA H100/H200/B100. ~60% market share.
000660.KS	High Bandwidth Memory HBM	AMD	0.9	SK Hynix HBM3e supplier to AMD MI300.
MU	High Bandwidth Memory HBM	NVDA	0.8	Micron ramped HBM3e in 2024, gaining share with NVDA Blackwell.
000660.KS	High Bandwidth Memory HBM	AVGO	0.8	SK Hynix HBM supplier to Google TPU (via AVGO).
005930.KS	High Bandwidth Memory HBM	NVDA	0.7	Samsung Electronics HBM3e gradually qualified by NVDA (slower than Hynix).
MU	High Bandwidth Memory HBM	AMD	0.7	Micron HBM3e supplier to AMD MI300 (smaller share).

Midstream (Design/Network/Optical Modules)¶

9 relationships

Upstream	Relationship	Downstream	Importance	Description
ANET	Network Switches	META	0.9	Arista is Meta's primary network supplier.
AVGO	Network Switches	GOOGL	0.9	Broadcom is Google's primary custom networking + TPU silicon partner.
ANET	Network Switches	MSFT	0.8	Arista Networks switches for hyperscaler AI fabrics.
MRVL	Network Switches	AMZN	0.8	Marvell DPU + custom silicon for AWS Trainium2.
AVGO	Network Switches	META	0.7	Broadcom custom silicon (MTIA) to Meta.
COHR	Optical Modules / Optical Communication	NVDA	0.9	Coherent supplies optical transceivers for NVDA Spectrum-X networking. NVDA $2B strategic equity inv
LITE	Optical Modules / Optical Communication	NVDA	0.8	Lumentum supplies optical chips. NVDA $2B strategic investment.
AAOI	Optical Modules / Optical Communication	NVDA	0.7	Applied Optoelectronics supplies optical transceivers.
300308.SZ	Optical Modules / Optical Communication	NVDA	0.7	Innolight is dominant Chinese optical transceiver supplier.

Downstream (Cloud/AI Labs/Customers)¶

12 relationships

Upstream	Relationship	Downstream	Importance	Description
MSFT	Cloud Customer Relationship	OPENAI	1.0	Microsoft Azure is OpenAI's primary compute (since 2019, exclusive until 2024).
AMZN	Cloud Customer Relationship	ANTH	0.9	AWS provides primary compute to Anthropic ($100B+ deal).
MSFT	Cloud Customer Relationship	ANTH	0.7	Microsoft Azure provides compute to Anthropic (after 2024-Q4 multi-year deal).
GOOGL	Cloud Customer Relationship	ANTH	0.7	Google Cloud also provides compute to Anthropic (multi-cloud).
NVDA	GPU → AI Labs	MSFT	0.9	NVDA GPUs power Microsoft Azure AI + OpenAI compute.
NVDA	GPU → AI Labs	CRWV	0.9	CoreWeave is largest NVDA GPU customer outside hyperscalers ($100B backlog).
NVDA	GPU → AI Labs	META	0.9	Meta is one of NVDA's largest customers (Llama 3 training on 16K H100).
NVDA	GPU → AI Labs	NBIS	0.9	Nebius (formerly Yandex) NVDA GPU customer, alternative neocloud.
NVDA	GPU → AI Labs	ORCL	0.8	Oracle Cloud massive NVDA GPU buildout for OpenAI compute (Stargate).
NVDA	GPU → AI Labs	AMZN	0.8	AWS offers NVDA GPUs (alongside Trainium2 alternative).
NVDA	GPU → AI Labs	ANTH	0.8	Anthropic uses NVDA GPUs for Claude training.
NVDA	GPU → AI Labs	GOOGL	0.7	NVDA GPUs in Google Cloud (alongside TPU).

Supporting (Data Centers/Power/Cooling)¶

13 relationships

Upstream	Relationship	Downstream	Importance	Description
VRT	Data Center Infrastructure	CRWV	0.9	Vertiv to CoreWeave (neocloud).
VRT	Data Center Infrastructure	MSFT	0.8	Vertiv power + cooling infrastructure for MSFT datacenters.
VRT	Data Center Infrastructure	GOOGL	0.8	Vertiv to Google datacenters.
VRT	Data Center Infrastructure	AMZN	0.8	Vertiv to AWS datacenters.
VRT	Data Center Infrastructure	META	0.8	Vertiv to Meta datacenters.
VRT	Data Center Infrastructure	NBIS	0.8	Vertiv to Nebius.
ETN	Data Center Infrastructure	MSFT	0.8	Eaton power distribution for hyperscaler datacenters.
ETN	Data Center Infrastructure	GOOGL	0.8	Eaton for Google.
ETN	Data Center Infrastructure	AMZN	0.8	Eaton for AWS.
HUBB	Data Center Infrastructure	MSFT	0.7	Hubbell electrical for hyperscalers.
CEG	Power	MSFT	0.8	Constellation Energy 20-year PPA with Microsoft for nuclear power (Three Mile Island reopening). Cri
VST	Power	MSFT	0.7	Vistra power supply to hyperscalers.
CEG	Power	AMZN	0.7	Constellation power.

Key Takeaways (5 Things)¶

NVIDIA is not isolated — It depends on TSMC for foundry, SK Hynix for HBM memory, Coherent/Lumentum for optical modules, and Synopsys/Cadence for design software. Any disruption in one link affects NVDA.
AI labs are not direct customers — OpenAI buys compute from Microsoft Azure, and Microsoft then buys GPUs from NVIDIA. So in NVDA's earnings, "hyperscaler capex" is key, not "OpenAI revenue."
CoreWeave is a "Neocloud" — Unlike hyperscalers, it only rents AI compute (no general cloud), with highly concentrated customers (MSFT accounts for 60%+), and has an equity relationship with NVIDIA ($36B equity).
Power is the real bottleneck — AI data centers consume enormous electricity, involving not just hardware but also "picks and shovels" like Constellation Energy (nuclear PPA), Vistra, Eaton, and Vertiv.
The industry chain transmits upstream and downstream — Samsung strike → HBM shortage → NVDA shipments affected → customers queue. Learn to track upstream + downstream for a complete view on any single stock.