Cisco and Arista: One Customer Defected, One Didn'tNarrow moat

Broadcom's merchant-silicon model created the modern networking industry: rather than each vendor designing its own chips, Broadcom sells Tomahawk and Trident to everyone, and the vendors compete on software and systems. It is an enormously profitable arrangement, and it contains an obvious hazard — the customers know exactly how valuable the silicon is.

Arista is the model customer. Its switches run on Broadcom silicon, its own moat lives in the EOS software layer above it, and it has built a very successful business without ever needing to design a switch chip. Broadcom captures the silicon margin; Arista captures the systems margin; neither threatens the other. Arista's own pages in this collection treat that dependence as its principal supply risk.

Cisco went the other way. It develops Silicon One in-house, and in February 2026 launched the G300 — a 102.4-terabit switch chip aimed squarely at Broadcom's Tomahawk 6 and Nvidia's Spectrum-X — and, critically, sells it as merchant silicon to other vendors rather than reserving it for its own boxes¹. That is a customer becoming a supplier in the same market.

The reason this matters more than a single product: Cisco has proved the defection is possible for a company with sufficient scale. Watch whether any other large systems vendor or hyperscaler follows. Broadcom's switching position rests on it being uneconomic for customers to design their own — an argument that weakens each time somebody does.