AI Supply Chain Physical Flow Visualization

Mapping the physical movement of materials, components, and systems across 6 layers and 12+ countries | April 2026

8-12
Border crossings per chip
15-20
Companies per GPU
4-6
Months wafer to inference
70%
AI trade value by air
180t
Weight of one ASML EUV
40
Containers per EUV machine

01 | Six-Layer Physical Flow

Click any layer to see detailed physical flows, companies, and freight characteristics.

L1

Raw Materials & IP

Wafers, chemicals, gases, rare earths, EDA

L2

Equipment

EUV lithography, etch, deposition, inspection

L3

Fabrication

Foundries, HBM memory, advanced packaging

L4

Chip Products

GPUs, ASICs, CPUs, AI servers

L5

Cloud Infra

Power, cooling, networking, cabling, DCs

L6

AI Platforms

Models, APIs, training, inference

L1 to L2: Raw Materials to Equipment Makers

OriginCompanyCargoDestinationMode
JapanShin-Etsu, SUMCO300mm silicon wafersASML (NL), Applied Materials (US)Air / Sea
JapanTOK, JSR, FujifilmEUV photoresistsTSMC (Taiwan), Samsung (Korea)Temp-controlled air
US/France/GermanyLinde, Air Liquide, Air ProductsSpecialty gases (xenon, neon)All major fabs globallyHazmat certified
US/AustraliaMP Materials, LynasRare earth mineralsMagnet / component mfrsBulk sea + air
Japan/TaiwanAjinomoto, Unimicron, IbidenABF substratesTSMC, ASE Group (Taiwan)Air freight

L2 to L3: Equipment to Fabs

OriginCompanyCargoDestinationMode
NetherlandsASMLEUV machines (180t, 40 containers each)TSMC Taiwan, TSMC Arizona, Samsung Korea, Intel ArizonaWhite-glove oversized
California, USApplied Materials, Lam Research, KLADeposition, etch, inspection tools + spare partsTSMC, Samsung, SK Hynix, Micron, Intel fabsAir freight + AOG express
JapanTokyo ElectronDeposition, cleaning toolsAll major fabsSea + air

L3 to L4: Fabs to Chip Designers

OriginCompanyCargoDestinationMode
TaiwanTSMCFinished wafers / packaged dies (B200, MI350X, TPU, Trainium)Nvidia, AMD, Broadcom, Google, Amazon (California/Seattle)High-value air freight
South KoreaSK HynixHBM3e memory stacksTSMC Taiwan (for CoWoS packaging)Temp-controlled air
South KoreaSamsung MemoryHBM3eTSMC, AMD, GoogleAir freight
US/JapanMicronHBM3eTSMC TaiwanAir freight
Taiwan/KoreaASE Group, AmkorFinal packaged AI chipsNvidia, AMD, BroadcomSecure air freight

L4 to L5: Chips and Servers to Data Centers

OriginCompanyCargoDestinationMode
CaliforniaNvidiaGPU modules (B200/GB200)Dell (TX), Supermicro (CA), HPE (TX), direct to hyperscalersSecure express
CaliforniaAMDMI350X accelerator modulesDell, HPE, Supermicro, Microsoft, MetaExpress air
Texas / CaliforniaDell, SupermicroAssembled AI server racks (multi-ton)AWS (Virginia, Oregon), Google (Iowa), Meta (Louisiana, TX)White-glove trucking

L5 Internal: Infrastructure Vendors to Data Centers

CategoryCompanyCargoOriginMode
PowerEaton, Schneider, Vertiv, ABBUPS systems, PDUs, switchgearUS, France, SwitzerlandHeavy freight / trucking
GeneratorsCaterpillar, CumminsOn-site generators (multi-MW)Illinois, Indiana (US)Oversized freight
CoolingVertiv, Schneider, Stulz, LiquidStackCDUs, liquid cooling, immersion tanksUS, Germany, NetherlandsHeavy freight
NetworkingArista, Cisco, Nvidia/MellanoxEthernet/InfiniBand switchesCalifornia, IsraelExpress air
CablingTE Connectivity, Amphenol, CorningDAC cables, fiber opticSwitzerland, Connecticut, New YorkStandard freight
OpticsCoherent, Lumentum, Ciena800G transceivers, coherent modulesPennsylvania, MarylandExpress air

L5 to L6: Cloud to AI Platforms

This flow is primarily digital (API calls, model weights over network). Physical flows are limited to initial GPU server deployments covered in L4 to L5. Key relationships: AWS serves OpenAI + Anthropic | Google Cloud runs Gemini on TPU pods | Microsoft Azure hosts OpenAI exclusively | Meta runs LLaMA on private infrastructure | CoreWeave provides overflow GPU capacity.

02 | End-to-End Journey of an Nvidia B200 GPU

16 steps from silicon sand to running GPT-5 inference.

1

Silicon Sand Extraction

Global sources

High-purity quartz sand refined for semiconductor use

2

300mm Wafer Manufacturing

Shin-Etsu, Gunma, Japan

Sand refined into ultra-pure silicon ingots, sliced into 300mm wafers

3

Wafers Shipped to Foundry

Japan to Taiwan (air freight, cleanroom packaging)

Temperature and vibration controlled transport to TSMC

4

EUV Photoresist Applied

TSMC Hsinchu, Taiwan (chemicals from JSR/TOK, Japan)

Light-sensitive chemical coating enables circuit patterning at 3nm/2nm

5

Circuits Printed via EUV Lithography

TSMC Hsinchu (using ASML EUV machine from Veldhoven, Netherlands)

Billions of transistors printed at nanometer scale

6

Etching and Deposition

TSMC (using Lam Research tools from Fremont, California)

Material layers added and removed to form 3D circuit structures

7

HBM3e Memory Arrives

SK Hynix, Icheon, South Korea to TSMC, Taiwan

High Bandwidth Memory stacks shipped for co-packaging with GPU die

8

CoWoS Advanced Packaging

TSMC Hsinchu, Taiwan

GPU die + HBM stacked together on silicon interposer (Chip-on-Wafer-on-Substrate)

9

Packaged Chip Shipped to Nvidia

Taiwan to Santa Clara, California (high-value air freight)

Completed B200 GPU modules shipped under tight security

10

GPU Module Assembly

Nvidia / contract manufacturer (Taiwan or California)

GPU mounted on PCB with power delivery and thermal solution

11

Modules Shipped to Server OEM

California to Dell (Round Rock, TX) or Supermicro (San Jose, CA)

8x GPU modules per server tray for DGX / HGX systems

12

AI Server Rack Assembled

Dell (Texas) or Supermicro (California)

Complete rack: GPUs + CPUs + networking + power + cooling integration

13

Server Rack Shipped to Data Center

Texas/California to AWS Ashburn, Virginia (white-glove trucking)

Multi-ton racks transported with vibration protection

14

Rack Installed with Power and Cooling

AWS Data Center, Ashburn, Virginia

Eaton UPS, Vertiv CDU liquid cooling, Arista Ethernet switch connected

15

Fiber and Optics Connected

Corning fiber + Coherent 800G transceivers installed

GPU cluster linked to storage and other racks at 800Gbps per link

16

Cluster Goes Live: AI Inference Running

OpenAI runs GPT-5 inference on this GPU

From silicon sand to answering user queries in ~4 to 6 months elapsed time

03 | Geographic Supply Chain Hubs

Japan

Shin-EtsuSUMCOTOKJSRTokyo ElectronAjinomotoIbidenFujifilm

Dominates L1 (wafers 55%, photoresists 70%) and L2 (TEL 15% equipment)

Taiwan

TSMCASE GroupGlobalWafersUnimicron

L3 hub: 66% foundry share, 90%+ advanced AI chips, CoWoS packaging

South Korea

SK HynixSamsungSK Siltron

HBM memory capital: SK Hynix ~57-62% share, Samsung ~22%

Netherlands

ASML

Single point of failure: EUV lithography monopoly, ~19% equipment market

USA (California)

NvidiaAMDBroadcomApplied MaterialsLam ResearchKLASynopsysCadenceAristaCiscoSupermicro

Chip design, EDA, equipment, networking HQ hub

USA (East / Central)

AWS (Virginia)Dell (Texas)Vertiv (Ohio)Caterpillar (Illinois)Cummins (Indiana)Corning (New York)Coherent (Pennsylvania)

Data center operations, power/cooling manufacturing, fiber optics

Europe

ASML (NL)Schneider Electric (FR)ABB (CH)TE Connectivity (CH)Stulz (DE)Rittal (DE)Siltronic (DE)Merck KGaA (DE)Submer (ES)Alfa Laval (SE)ARM (UK)Mistral AI (FR)

Strong in equipment (ASML), power/cooling, IP (ARM), and emerging AI (Mistral)

Israel

Nvidia/Mellanox (Yokneam)

InfiniBand networking R&D and manufacturing hub for AI cluster interconnects

04 | Critical Freight Lanes for AI Supply Chain

Netherlands to Taiwan
ASML EUV lithography machines, 180 tons per unit, 40 containers, white-glove logistics
Oversized / Charter
Japan to Taiwan
Silicon wafers (Shin-Etsu, SUMCO), photoresists (TOK, JSR), ABF substrates (Ajinomoto, Ibiden)
Temp-Controlled Air
South Korea to Taiwan
HBM3e memory stacks (SK Hynix, Samsung) for CoWoS packaging at TSMC
Secure Air Freight
Taiwan to California
Finished AI chips (Nvidia B200, AMD MI350X, Broadcom ASICs, Google TPU dies)
High-Value Air
California to US East/Central
GPU modules to Dell (TX), assembled server racks to AWS (VA), Google (IA), Meta (LA, TX)
White-Glove Trucking
California to US Equipment Fabs
Applied Materials, Lam Research, KLA tools and spare parts to TSMC Arizona, Intel Arizona
AOG Express Air
Europe to Global DCs
Schneider Electric (FR) UPS/PDU, ABB (CH) switchgear, Stulz (DE) cooling to DC campuses
Heavy Freight / Sea+Air
Israel to Global
Nvidia/Mellanox InfiniBand switches and networking cards from Yokneam to GPU clusters globally
Express Air