It is a common misconception that high-performance AI clusters require proprietary interconnects. This session proves that high-performance AI fabrics can be built entirely on open standards. We showcase a new production-ready OCP reference architecture utilizing Celestica’s OCP-inspired DS5000 (800G) and the roadmap for the DS6000 (1.6T) switches, orchestrated by Hedgehog’s AI fabric software. Leveraging SONiC and standard RDMA over Converged Ethernet (RoCEv2), this stack provides the deterministic low latency and massive throughput required for Blackwell-class GPU clusters. We will present engineering benchmarks demonstrating that this open-standard approach achieves performance parity with proprietary alternatives while offering superior flexibility and supply chain resilience. Attendees will explore the hardware-software co-design necessary to manage 1.6T signaling and high-density power profiles within the OCP framework, proving that open-source ecosystems are ready for AI scale.