Xin Lang Cai Jing·2026-02-10 10:34Core Insights - Cisco has launched a new chip and router aimed at accelerating information transmission in large data centers, competing for a share of the $600 billion AI infrastructure spending market [1][2] Group 1: Product Launch - The Silicon One G300 switching chip is expected to be released in the second half of this year, designed to efficiently interconnect AI system training and operational chips through hundreds of thousands of links [1][2] - The chip will be manufactured using TSMC's 3-nanometer process and will feature new "buffer dampening" characteristics to prevent network congestion during spikes in data traffic [1][2] Group 2: Performance Expectations - Cisco anticipates that the new chip could enhance the speed of certain AI computing tasks by 28%, partly due to its ability to automatically reroute data around network failures in microseconds [1][2] - The focus is on overall end-to-end network efficiency, especially when dealing with tens of thousands of connections, which frequently encounter congestion issues [1][2] Group 3: Competitive Landscape - The networking sector has become a critical competitive arena in the AI field, with Nvidia recently releasing a system that includes a network chip among its six core chips, directly competing with Cisco's products [1][2] - Broadcom is also competing in the same market with its "Tomahawk" series of chips [1][2]