NVIDIA and Cisco Introduce New Pluggable Optics Form Factor to Scale AI Factories - semivision

This is not a form-factor release—it is an interconnect ecosystem lock-in. By fusing GPU cluster fabric with a unified optical spec, NVIDIA is hardwiring the AI factory's vascular system into its own architecture. Upstream, EML laser and silicon-photonics design anchors get rewritten. Downstream, module makers such as Innolight and Eoptolink face a 12-to-18-month production-line transition where compliance adaptation and yield ramp will compress 2025 Q3–Q4 margins. Cisco's Silicon One gains sticky lock-in; Broadcom's Tomahawk risks shedding 15–20% of AI datacenter switch share if its next generation cannot match the optical port spec. The deeper signal: CPO commercialization likely slips to 2027, since pluggable optics retain a cost edge at 1.6T—leaving the discrete transceiver segment at least 24 months of a golden window. For AMD, if MI400 cannot plug into the same optical standard, the dual-vendor AI factory narrative collapses. The real winner is not the module maker—it is whoever defines the interconnect spec.