Oracle adds multiplanar RDMA fabrics for Vera Rubin and TPUs
Oracle announced Acceleron networking improvements alongside NVIDIA Vera Rubin and Google TPU architecture for AI cloud deployment. New capabilities include RDMA-based fabrics, resilient multiplanar networking, and distributed collective communications for moving data across large clusters.
Why it matters Oracle's multiplanar RDMA design and collective communications library provide alternatives to single-plane architectures for large distributed training, affecting how practitioners architect fault-tolerant AI infrastructure at hyperscale.
Oracle unveiled Acceleron AI networking advances at AI World 2026 on October 27, spanning three elements: NVIDIA Vera Rubin accelerators, Google TPU architecture, and Oracle's high-performance cloud networking. The company reported adding 850 MW of data center capacity and deploying more than 300,000 GPUs in fiscal Q1 2027. Oracle Acceleron combines dedicated network fabrics, programmable networking, direct data paths and host-level acceleration. New capabilities include the Oracle Collective Communications Library, which helps customers keep distributed AI workloads running when network paths fail by spreading collective traffic across OCI's backend Multiplanar Network. Transparent Front-End Multiplanar Networking brings multiplanar networking to front-end networks with built-in transparent path selection, failover, and transparent congestion control for front-end RDMA traffic. NVIDIA and Oracle announced plans to expand OCI Superclusters to multi-gigawatt scale using NVIDIA Vera Rubin NVL72 rack-scale systems. This represents a significant shift from single-plane RDMA architectures toward active multiplanar redundancy in production AI infrastructure, with implications for how practitioners design collective operations and handle network fault isolation.
Read the original at techblog.comsoc.org ↗