Most AI chip companies are fabless: they design chips but don't own factories. NVIDIA, AMD, Google, Amazon, Microsoft, Meta, Cerebras and Groq all send their designs to a foundry to be manufactured.
The company on the label is rarely the company that made the silicon.
THE DESIGNERS
Who designs them
🧑💻
NVIDIAData-center GPUs (H100, B200 and successors); the market leader
AMDInstinct MI300 and MI350 series GPUs
Cloud companiesGoogle TPU, AWS Trainium and Inferentia, Microsoft Maia, Meta MTIA
OthersIntel Gaudi; start-ups such as Cerebras, Groq, SambaNova and Tenstorrent; Huawei Ascend in China
Cloud companies design their own chips to cut costs and depend less on one supplier.
THE BUILDERS
Foundries, memory and tools
🏭
FoundriesTSMC makes nearly all leading AI chips; Samsung and Intel Foundry compete
MemorySK hynix, Samsung and Micron make HBM
ToolsASML (lithography), Applied Materials, Lam Research, Tokyo Electron and KLA
Design softwareSynopsys and Cadence make the EDA tools every chip is designed with
A single AI chip depends on dozens of companies in several countries.
SORT THEM
Designer, foundry, memory or tools?
Sort each company by its main role in making an AI chip.
Some companies, like Samsung and Intel, play more than one role.
WHY IT MATTERS
Concentration
🎯
Several steps depend on just one or a few companies: ASML for EUV, TSMC for leading-edge logic and CoWoS, three companies for HBM. That concentration makes the industry efficient but fragile, since a disruption at any one of them can slow AI chip production worldwide.
AI chips are a chain, and a chain is as strong as its rarest link.
UNIT 18 STUDY COMPLETE
🏢
Ready for the Fab Challenge?
You've sorted the companies that design AI chips from those that manufacture them, supply memory or build the tools.