NVIDIA Corporation

What they make

2 inputs NVIDIA Corporation supplies

Click an input to see every good that depends on it, every country that produces it, and every other company in the supply chain.

Where they make it

1 facility

Intelligence

What's known

Sourced claims about this company's role in supply chains — chokepoints, concentration, incidents, dual-use connections.

• Did you know2023

  The H100 GPU that trains ChatGPT is physically identical to the H100 GPU that the US Department of Energy uses to simulate nuclear weapons at Los Alamos, that pharmaceutical companies use to run molecular dynamics for drug discovery, and that financial firms use for options pricing Monte Carlo simulations. The US government classified H100 as a military-controlled export item in October 2023, restricting sales to China and other adversary nations, precisely because the civilian AI chip and the defense simulation chip are the same piece of hardware. This dual-use classification means that every H100 NVIDIA sells into a commercial AI data center is subject to the same export control framework as weapons-grade technology -- and NVIDIA, a chip designer with no defense contracts, is now a gatekeeper for national security policy through its allocation decisions.

  US Bureau of Industry and Security ↗

• Chokepoint2023

  NVIDIA is a fabless semiconductor company -- it designs all chips but owns zero fabs. 100% of NVIDIA GPU production is manufactured by TSMC in Taiwan, with H100/H200 on N4 process and Blackwell on N3E. In FY2025, NVIDIA paid TSMC an estimated $23.4B, making it TSMC's largest customer at 19% of TSMC revenue. This means the global AI infrastructure buildout -- estimated at $500B+ in annual capital spending by 2025 -- is physically bottlenecked by the production capacity of a single fab complex in a country 100 miles from mainland China. NVIDIA has no capacity redundancy: AMD and Intel also use TSMC for leading-edge nodes. A Taiwan Strait crisis that halted TSMC production for 6 months would freeze global AI infrastructure expansion more completely than any conceivable cyberattack or sanctions regime.

  TSMC ↗

• Origin2023

  NVIDIA was founded in 1993 to make graphics chips for video games, and its AI dominance was not a strategic plan -- it was an accident of CUDA. In 2006, NVIDIA engineer Ian Buck pitched CEO Jensen Huang on CUDA (Compute Unified Device Architecture), a general-purpose programming layer for GPUs. Huang approved it as a research project. For years CUDA was a niche tool for researchers who realized GPUs could parallelize non-graphics workloads. It was Stanford researchers and Google Brain scientists -- not NVIDIA's sales team -- who discovered that GPUs trained neural networks orders of magnitude faster than CPUs. By the time the AlexNet moment (2012) proved deep learning worked at scale, CUDA had an 8-year head start on any competitor. NVIDIA did not build an AI company; academic researchers turned a video game chip company into the infrastructure backbone of AI by writing papers about CUDA performance.

  NVIDIA Corporation ↗

• Concentration2026

  NVIDIA's entire AI accelerator product line — every H100, H200, B100, B200, and GB200 — is manufactured exclusively at TSMC Taiwan. NVIDIA is a fabless company: it designs chips at its Santa Clara California headquarters but owns no fabs. NVIDIA surpassed Apple as TSMC's largest customer in 2025 ($23.4B, 19% of TSMC revenue, +62% YoY). This means that the global AI infrastructure buildout — estimated at $300B+ in data center capital expenditure in 2025 — is physically constrained by and entirely dependent on a single company (TSMC) on a single island (Taiwan) that faces both seismic and geopolitical risk. If Taiwan were blockaded tomorrow, new AI chip deliveries would cease within weeks (TSMC inventory pipeline) to months. Every major US cloud provider (AWS, Azure, Google Cloud, Oracle) has AI infrastructure directly dependent on Taiwan remaining stable, peaceful, and accessible.

  CNBC ↗