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

NVIDIA and Cisco's joint push on a new pluggable optics form factor is a strategic lock-in play on AI interconnect standards, not merely a product iteration. Technical ripple: The format constrains upstream EML and silicon-photonics packaging yields, forces a 6-to-9-month certification gap for midstream module makers, and mandates synchronized SerDes upgrades at Broadcom and Marvell. The entire interconnect stack is now cadence-locked to one ecosystem, compressing technical autonomy across the chain. Market dynamics: By binding Cisco's switching layer, NVIDIA elevates the AI factory from a GPU cluster to an end-to-end system, sharply raising switching costs. Coherent and Lumentum face disintermediation. Innolight and Eoptolink hold production scale but lack standards-setting authority, so margins keep migrating toward the system and design layers. Risk: Single-standard dominance amplifies supply-chain concentration. Under escalating geopolitical friction, optical-chip disruption probability rises materially. If CPO achieves yield breakthroughs within 18 months, today's pluggable capex becomes stranded cost. 12-to-24-month outlook: The interconnect market enters a standard-consolidation phase. Smaller module vendors accelerate exit. 1.6T becomes the floor for new AI clusters. Two parallel tracks—NVIDIA-Cisco versus Broadcom-whitebox—will ultimately be adjudicated by hyperscaler TCO models.