Cisco unveils its Silicon One G300 switch chip, made using TSMC's 3nm tech, to take on Broadcom and Nvidia in AI infrastructure, set to go on sale in H2 2026
Context & Ripple Effects
Cisco is extending Silicon One from long-haul routing into the AI-networking buildout. Its earlier P200 chip and 8223 routing systems focused on faster transfers across long optical links, while later switch updates emphasized relieving AI-network bottlenecks.
The G300 puts Cisco’s networking silicon more directly against Broadcom and Nvidia while making TSMC’s 3nm process part of Cisco’s AI-infrastructure product roadmap. That raises the stakes from system-level competition to the chip layer.
First-order effects
- Cisco gains a new AI-infrastructure switch chip scheduled for H2 2026, giving its customers another Silicon One-based option for building AI networks.
- Broadcom and Nvidia face a more direct Cisco challenge in the switch-silicon segment; TSMC adds Cisco’s G300 to the set of advanced-node designs it manufactures.
Second-order effects
- Cisco’s system and optical-networking offerings can be positioned alongside proprietary switch silicon, building on its switch refresh aimed at AI-network bottlenecks.
- The launch gives AI-network buyers a potential additional supplier path, increasing pressure on incumbent platforms to compete on performance, interoperability, and availability rather than silicon alone.
Third-order effects
- If Cisco converts the chip into deployed systems, AI-networking competition could become less concentrated around a small set of merchant and accelerator-adjacent silicon suppliers.
- The broader shift is toward vertically integrated AI infrastructure, where networking vendors differentiate through custom silicon, systems, and optics together; adoption remains dependent on customer validation and supply execution.
The trend: AI infrastructure is broadening from accelerator procurement into a contest over the custom networking silicon and integrated systems needed to move AI workloads at scale.