AI factories are being built in the open. As workloads shift toward agentic reasoning and inference-time scaling, the network has become the decisive variable in GPU utilization, token economics, and time-to-first-token. Building these fabrics at gigawatt scale demands both engineered performance and shared standards.
This keynote explores how NVIDIA Spectrum-X Ethernet, the world's first Ethernet platform purpose-built for AI, advances the OCP community's mission of open, efficient, and scalable infrastructure. Spectrum-X Ethernet is built on the open Ethernet standard and aligned with OCP-developed specifications.
For operators building sovereign AI, neocloud platforms, and hyperscale training and inference systems, the question is no longer whether to build on open foundations, it’s how to scale them. This session offers a builder's view of how open standards and engineered performance compose into the fabric of the AI factory.