AI 行业情报分析：理解主题演进与主体角色

Investors have increasingly framed AI networking as a binary debate between copper and optics. Thus they are constantly thinking and evaluating the rotation in and out each theme. We think that the networking landscape can actually be viewed from a simpler lens. As GPU clusters scale, the importance of connecting them efficiently only increases, and that drives demand for more networking content of all types. Copper and optics serve different but complementary roles. Copper remains the preferred solution where it can meet reach, power, cost, and reliability requirements, while optics becomes necessary where bandwidth and distance push beyond copper’s practical limits. Nvidia’s approach has been consistent: use copper where they can, and optics where they must. In other words, this is not an either-or market. The growth of optical interconnect does not mean copper goes away, and the durability of copper does not mean optics will not ramp. As AI systems scale in size and complexity, both copper and optical connectivity should benefit from the increasing importance of moving data between GPUs, switches, racks, and clusters.