Nvidia and Cisco unveil new optics design to help scale AI factories

Nvidia and Cisco's new pluggable optics form factor is less a product launch than a strategic stalling maneuver against co-packaged optics (CPO). Broadcom has spent two years positioning CPO as the endgame, attempting to fuse optical engines with switch ASICs to lock in customers. Nvidia's counter-signal is unambiguous: the pluggable paradigm, re-engineered for density, can absorb the power and cabling penalties of 1.6T without forcing operators to abandon existing O&M tooling. Downstream, this preserves the EML laser and AWG chip supply chain for module makers like Innolight and Eoptolink—no architectural reset required. But the 200G-per-lane transition shifts the bottleneck to electrical interface loss, intensifying R&D pressure on TIA and driver ICs. On the compliance front, the sovereign-AI wave is spawning parallel supply chains across the Middle East, Southeast Asia, and Europe. Export licensing for 200G/lane EML chips has stretched to six-to-nine months, creating latent fragility for assemblers dependent on Lumentum and Coherent capacity in Taiwan, China. Competitively, Broadcom will likely accelerate CPO options on Tomahawk 6, forging a two-track market. Nvidia leverages its NVLink ecosystem to anchor the pluggable camp; Cisco cashes in on its installed-base refresh cycle. Twelve-to-twenty-four-month outlook: 1.6T pluggables hit volume production in H2 2025. CPO will first penetrate sub-2-meter GPU-to-GPU links. The true inflection lands at the 3.2T node in 2027, where thermal density hits physical limits and CPO may become the only viable path.