投资者常将AI网络视为铜缆与光缆的二元选择,但SemiAnalysis认为二者互补。GPU集群规模扩大,高效连接驱动各类网络需求增长。铜缆在满足距离、功耗、成本与可靠性时仍是首选,光缆则在带宽和距离超出铜缆极限时必需。Nvidia策略一致:能用铜缆用铜缆,必须用光缆才用。因此光学增长不取代铜缆,铜缆持久也不意味着光学不增长。随着AI系统规模与复杂度提升,两者都将受益于GPU、交换机、机架与集群间数据传输需求的增加。
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.