Nvidia ships TSMC-backed CPO switches to tackle AI data center bottlenecks - Taipei Times

NVIDIA has begun shipping next-generation co-packaged optics (CPO) switches developed in collaboration with Taiwan Semiconductor Manufacturing Co. (TSMC), aimed at addressing data transmission and pow... Read original →

NVIDIA’s CPO rollout with TSMC marks a structural shift from electrical to optical interconnects in AI data centers. This move forces upstream silicon photonics, 3nm EUV packaging, and downstream thermal solutions into rapid co-evolution around integrated optical engines. Geopolitically, U.S. export controls on advanced packaging could soon target CPO modules, compelling NVIDIA to build redundant capacity outside Taiwan, China—raising near-term costs. In response, Broadcom and Marvell may abandon pluggable optics for in-house CPO designs, but without access to TSMC’s COUPE platform, they’ll face widening performance gaps. Within 18 months, CPO will become a de facto entry barrier for AI clusters; cloud operators lacking co-packaged optics will lose critical power-efficiency advantages. Meanwhile, TSMC solidifies its role as the invisible backbone of global AI infrastructure through silicon photonics integration.