The AI Infrastructure Stack
Overview  /  Tier V Silicon
Layer 11

Chip Design Software & IP

The duopolies you can’t design a modern chip without.

What this layer does

Before a chip can be fabricated, it has to be designed. That requires EDA software (electronic design automation: simulation, place & route, verification), pre-built IP blocks (CPU cores, memory controllers, PCIe/NVLink interfaces), and photomasks that physically encode the design on the wafer. Every accelerator at TSMC — Nvidia’s, Google’s TPU, AWS Trainium — runs on Synopsys or Cadence software and licenses Arm or Synopsys IP.

This is one of the most beautifully structured industries in tech: a Synopsys/Cadence duopoly in EDA, Arm in CPU IP, and three photomask producers. Customers can’t switch off easily, the products are mission-critical, and AI is increasing both seat count and licensing intensity.

Sub-categories

Analysis coming soon — will cover: SNPS+CDNS duopoly economics, ARM royalty rate trajectory and licensing changes, RISC-V threat (real but slow), why this is arguably the highest-quality public exposure to AI silicon growth, and Synopsys-Ansys deal implications.