NVIDIA and Cisco intro pluggable optics form factor for AI factories - Fierce Network

NVIDIA and Cisco defining the pluggable optics form factor for AI factories is not a product announcement—it is an ecosystem lock-in play that mirrors the TCP/IP standardization moment. Whoever controls the socket harvests the entire transceiver value chain. The immediate structural consequence: CPO commercialization gets pushed to 2027 at the earliest, handing a two-to-three-year revenue runway to 800G/1.6T pluggable module makers, while simultaneously concentrating supply-chain risk on China-based laser-chip and module capacity and amplifying geopolitical exposure for US hyperscalers. Broadcom and Marvell are now forced to design switch ASICs within NVIDIA's interface envelope, eroding their pricing leverage. Over the next 18 months, the profit pool migrates upstream to EML/DFB laser chips and silicon-photonics engines, while Huawei's parallel optical interconnect standard in the closed domestic market becomes the only alternate universe. The real question is not who wins the form-factor war—it is who owns the photonics IP layer beneath it.