SUPPLY CHAIN ATLAS Individually researched profile

NVIDIA — suppliers and downstream connections.

Accelerated computing platform: Independently verifies manufacturing or delivery methods, upstream and downstream links, and partnerships; read each item alongside its evidence date and undisclosed details.

Checked against public sources2026-09-05Selected relationships · not a complete supply chain

Select a company for explanations and evidence

Understand the manufacturing and delivery model

NVIDIA designs computing platforms, while external partners produce wafers, memory and some systems. Chips may pass through integrators such as Dell before reaching cloud customers; the product-adoption chain is not the payment chain. Beyond data centers, engineering-design, healthcare and industrial partnerships are examined separately. Partnership investments are not treated as GPU orders.

FY2026 Annual Report: Manufacturing and Competition ↗Delivery of GB200 systems to CoreWeave ↗NVIDIA and Synopsys expand engineering-design collaboration ↗Johnson & Johnson MedTech and NVIDIA surgical-AI collaboration ↗

What has been checked, and what is still missing?

Upstream

Checked TSMC's and Samsung's manufacturing roles, SK hynix's and Micron's memory generations, and contract manufacturing by Hon Hai, Wistron and Fabrinet. System assembly is distinguished from wafer processes.

Not yet covered: Supply shares for each HBM generation and specific factories and packaging allocations for each GPU generation remain incomplete. Unnamed direct customers in the annual report cannot be identified.

FY2026 Annual Report: Manufacturing and Competition ↗U.S. chip and AI-system manufacturing footprint ↗Micron discloses volume production of HBM4 for Vera Rubin ↗Multi-year memory technology collaboration with SK hynix ↗

Downstream

Checked Azure, AWS, Google, OCI and Dell/CoreWeave cases, adding the different adoption stages of Amgen's research computing, Synopsys EDA and Johnson & Johnson's surgical AI.

Not yet covered: Healthcare memoranda, R&D investments and deployment plans are not large commercial purchases. Demand is not counted twice across system manufacturers and final deployers.

GB200 cloud instances become generally available ↗EC2 P6-B200 powered by B200 becomes generally available ↗AI collaboration with Alphabet and Google ↗Expanded AI inference collaboration with Oracle ↗Amgen's deCODE plans to deploy the Freyja computing platform ↗NVIDIA and Synopsys expand engineering-design collaboration ↗Johnson & Johnson MedTech and NVIDIA surgical-AI collaboration ↗

SUPPLY SIDE

Upstream supply & manufacturing

10 companies · 10 documented relationships

Current focusNVDA

NVIDIA

NVDA · NASDAQ

Designs GPUs, networking and accelerated computing systems and offers software platforms including CUDA; wafer fabrication, packaging and some system manufacturing are outsourced to partners.

GPUs & systemsNetworkingCUDA software

How to read this map

Connections indicate disclosed relationships, not purchasing amounts; read each connection's label.

Supply/adoption means a partner role or product use. Qualification means technical compatibility, not an order. An agreement is a signed arrangement, not proof of delivery.

Downstream includes integrators, qualification partners and agreement counterparties, not necessarily direct customers in financial reports. Different matters may connect the same companies.

DEMAND SIDE

Downstream adoption & partnerships

13 companies · 14 documented relationships

Technology & business partnerships No assumed buyer–seller direction

Ownership & business affiliation Excluded from supply and sales relationships

FOLLOW THE PRODUCT

Products, integration and deployment are different steps.

This is one disclosed delivery example, not the only purchasing route.Dell announcement ↗

01 / Supply dependence

Supplier names, but no purchasing shares

The annual report confirms manufacturing partners, but does not quantify dependence on any one company. We do not create unsupported concentration scores.

Manufacturing section ↗

02 / Demand & competition

Product adopters may also design their own chips

NVIDIA's annual report identifies some cloud providers' in-house computing solutions as competition. Adopting NVIDIA products and exploring alternatives can happen at the same time.

Competition section ↗

03 / Disclosure limits

We do not guess unnamed major customers

Cloud providers in the map are not matched to unnamed customers in financial reports, nor used to estimate sales. Product adoption, payment routes and revenue attribution need not coincide.