SemiAnalysis: AI Networking Not an Either-Or Between Copper and Optical Communication, Both Will Benefit from GPU Cluster Expansion

BlockBeats News, June 20th. Semiconductor and AI research firm SemiAnalysis published an article stating that investors are increasingly considering AI networks as a binary choice between copper and optical interconnects, continually evaluating the rotation between the two themes. However, it believes that as GPU cluster scale grows, the importance of efficiently connecting these clusters will only continue to rise, driving demand for various network content.

Copper and optical communications play different but complementary roles. When copper can meet distance, power, cost, and reliability requirements, copper remains the preferred solution; however, when bandwidth and distance exceed the actual limits of copper, optical communication becomes necessary. NVIDIA's strategy has always remained consistent: use copper when possible and only resort to optical communication when necessary.

This is not an either-or market. The growth of optical interconnects does not mean copper will disappear, and the resilience of copper demand does not mean optical communication will not ramp up. As AI system scale and complexity increase, the importance of data transfer between GPUs, switches, racks, and clusters rises, benefiting both copper and optical interconnects.