NVIDIA and Cisco unveil new optics format to scale AI factories - Latest news from Azerbaijan

This is a standards war disguised as a product launch. NVIDIA and Cisco are writing the interconnect grammar of AI factories—control the format, control the supply chain. The bottleneck in AI infrastructure has decisively shifted from raw FLOPS to interconnect bandwidth. This new optical interface forces a full redesign cycle across transceivers, switch ASICs, and data center cabling. Upstream, optical chip designers (Coherent, Lumentum) and module manufacturers (Innolight, Eoptolink) face a 12-to-18-month redesign window, with line-switching costs flowing directly into hyperscaler capex. On compliance: US export restrictions on advanced optical components are already fragmenting the global market. If this format becomes the de facto standard, China's domestic optical communications chain risks a 'standards lockout'—echoing how PCIe historically squeezed native interconnect alternatives. Competitive response: Broadcom will almost certainly accelerate its CPO roadmap as a counter. Arista may champion an alternative specification. The real battleground sits in OIF and IEEE standardization votes, not in press events. 12-to-24-month outlook: CPO transitions from lab to volume production, silicon photonics migrates into switch chip packaging, and AI factory architectures converge on two or three dominant formats. Optical interconnect is becoming the highest-margin, deepest-moat layer in the AI infrastructure stack.