Nvidia and Cisco unveil pluggable optics for AI factories - GamesBeat

Nvidia and Cisco's pluggable optics push is a standardization lock-in play, not a product launch. By anchoring AI-factory interconnect to Ethernet, Nvidia subordinates its own InfiniBand franchise and pulls the entire data-center networking stack into the CUDA gravity well. The technical cascade is immediate: OSFP/QSFP-DD becomes the de facto interface, squeezing spec leverage from EML laser makers (Coherent, Lumentum) and optical-DSP players (Marvell, Broadcom). Chinese module vendors—Innolight, Eoptolink—face a volume-up, margin-down squeeze as spec ownership consolidates in US hands. On compliance, the 800G+ transceiver supply chain is a geopolitical chokepoint; advanced EML and silicon-photonics fabs cluster in the US, Japan, and 中国台湾. Any tightening of export controls on high-speed optics will fracture the global AI-cluster interconnect supply into parallel tracks almost overnight. Competitively, Broadcom will accelerate a Tomahawk-6-plus-optical-DSP bundle to blunt Nvidia's ecosystem pull, while AMD leans harder on the UALink consortium to build an Ethernet-based AI networking path outside Nvidia's orbit. Within 12–24 months, expect 1.6T pluggable optics to become the AI-factory default, CPO to remain a hyperscaler-custom niche, and a consolidation wave among Tier-2 module vendors as standardization compresses margins.