The Small Connection Behind a Bigger AI Network

News brief · Announced 2026-10-12

On October twelfth, twenty twenty-six, NVIDIA and Cisco announced plans for a new pluggable optical-module format called USFP. The companies describe it as a way to connect more computing systems within a limited amount of network space. Their release also identifies TE Connectivity and Amphenol as participants in building the supporting hardware ecosystem.

The engineering problem is not simply how much calculation a processor can perform. A large AI system must also move information between machines. Expanding connections can add equipment, power demand, and operational complexity. Our interpretation is that the announcement highlights networking as an important part of scaling AI infrastructure, alongside processors and memory.

However, a proposed format and its expected advantages are not the same as widespread adoption. The companies also describe plans for an open industry specification through a future community-led project. Readers should distinguish the announcement, the intended specification, and demonstrated deployments. Supplier participation makes the proposal more concrete, but it does not independently establish how widely the format will be adopted or what every installation will achieve.