Hardware & Networking: OCP Insights: Day 1 Highlights Increasing Complexity in Scaling AI Infrastructure

We attended Day 1 of OCP 2026 in San Jose, California, where the overarching theme across the keynotes and executive sessions was the increasing complexity of scaling AI infrastructure, particularly with the rapid growth in inference and agentic workloads driving greater emphasis on system-level efficiency, networking, power delivery, and open architectures rather than simply adding accelerator capacity. Networking was particularly prominent, with discussions spanning scale-out, scale-across, and scale-up, including growing adoption of Ethernet in the latter and the need to interconnect increasingly large clusters across multiple data centers. Meanwhile, power availability and efficiency remain critical constraints, driving innovation around 800V DC architectures, liquid cooling, and software-based management to maximize compute capacity within existing power budgets. Connectivity was another major focus, with advancements across copper, NPO/CPO, and emerging optical technologies aimed