Market Share
~80% AI accelerator market
Key Product
H100, H200, Blackwell B200 GPUs
Bottleneck Status
Allocation queue 6–9 months
Full briefing▼ Expand
NVIDIA designs the AI accelerators that power the global AI infrastructure build-out, holding approximately 80% of the AI chip market. Its H100 GPU, built on TSMC's 4nm process with SK Hynix's HBM3E memory and Ibiden's substrates, became the defining product of the AI era — with list prices reaching $40,000 per unit and secondary market premiums well above that. The H200 and Blackwell B200 extend this lead with higher memory bandwidth and better inference performance. NVIDIA's CUDA software ecosystem creates powerful switching costs that reinforce its hardware dominance. US export controls have restricted NVIDIA's ability to sell its most advanced chips to China, leading the company to develop downgraded variants (H800, A800) — which were subsequently banned. NVIDIA's supply chain is itself a chokepoint map: every node it depends on — TSMC CoWoS, SK Hynix HBM, Ibiden substrates — is constrained.
Connected companies
Tap a chip to trace that company's chain.
Critical path — raw silicon to deployment
The tightest single-source dependencies, in order.
MATERIALS
Ibiden
FC-BGA substrates for AI accelerators
FOUNDRIES
TSMC
CoWoS advanced packaging, N3/N2 logic
MEMORY (HBM)
SK Hynix
HBM3E memory for H100/H200
CHIP DESIGNERS
NVIDIA
H100, H200, Blackwell B200 GPUs
SERVER ODMs
Super Micro
NVIDIA DGX-compatible GPU servers
AI CONSUMERS
xAI
Grok AI assistant, Colossus training cluster
Export controls touching NVIDIA
U.S. BIS Advanced Computing Export Controls (Oct 2023)
The U.S. Bureau of Industry and Security expanded controls on advanced AI chips and semiconductor manufacturing equipment, prohibiting sales of GPUs exceeding performance thresholds to China. The rules targeted NVIDIA A100/H100-class chips and required licenses for any chip enabling large-scale AI training. This effectively cut off China's cloud providers and AI labs from the latest U.S.-designed accelerators.
▲ 6 companies affected
U.S. BIS Advanced Computing & SME Export Controls (Oct 2022)
The original U.S. BIS advanced-computing rule, effective October 12, 2022, set performance thresholds on AI chips and restricted their export to China without a license. It also imposed comprehensive restrictions on semiconductor manufacturing equipment (SME) destined for Chinese fabs producing sub-16nm logic, sub-18nm DRAM, or 128-layer NAND. The rule marked the first time the U.S. explicitly used export controls to constrain China's capacity to produce advanced semiconductors, rather than merely restricting imports of finished chips.
▲ 6 companies affected
U.S. BIS Blackwell/GB300 Southeast Asia Transshipment Enforcement (May 2026)
A Bureau of Industry and Security memo dated May 31, 2026 affirmed that U.S. export license requirements apply to NVIDIA Blackwell-class chips — including the GB300 — whenever the ultimate purchasing entity is headquartered in China, regardless of where its subsidiary or shipping destination sits, closing a loophole that had let China-headquartered buyers acquire chips through shell entities in third countries. This follows a broader, already-documented pattern of advanced GPUs reaching China via re-export through Southeast Asian hubs — Singapore, Malaysia, and Thailand — where export-control enforcement infrastructure lags the U.S., Netherlands, and Japan, as seen in the $160 million Operation Gatekeeper smuggling case unwound in December 2025. On July 22, 2026, White House OSTP Director Michael Kratsios publicly alleged that Chinese AI lab Moonshot AI had acquired GB300-equipped servers and accessed GB300 chips staged in Thailand to train its models, and had built an internal platform to conduct large-scale distillation of U.S. models — specifically Anthropic's Fable — to develop its open-weight Kimi K3 model (released July 17, 2026, with 2.8 trillion parameters). Treasury Secretary Scott Bessent said sanctions and Entity List authority remain "on the table," but no formal U.S. enforcement action against Moonshot had been announced as of the allegation date, making this the first prominent case naming both a specific Southeast Asian transshipment corridor and a named Chinese AI lab in the Blackwell diversion debate.
▲ 6 companies affected
QWho supplies NVIDIA?
NVIDIA relies on 12 upstream suppliers across the AI chip supply chain.
Ibiden (Dominant supplier of IC substrates for advanced packages), TSMC (World's largest contract chip manufacturer), Samsung Foundry (Second-largest advanced foundry, competing at 3nm), SK Hynix (Leading HBM3/HBM3E supplier for AI accelerators), Samsung Memory (World's largest DRAM and NAND supplier, ramping HBM3), and 7 more.
QWhat does NVIDIA make?
Dominant AI GPU designer (H100, H200, B200)
Key products H100, H200, Blackwell B200 GPUs
▲Allocation queue 6–9 months