Las Vegas – October 28, 2026 -- Oracle said new NVMe block-storage attachment options across Oracle Acceleron bare metal and VM shapes improve performance by up to 3x, as it expanded its Oracle Cloud Infrastructure (OCI) networking suite for AI workloads at Oracle AI World.

Oracle extends NVMe block storage and RDMA connectivity to more AI workloads

The NVMe options target AI, database and other data-intensive workloads. Separately, Oracle Acceleron High-Performance Fabric adds RDMA connectivity on OCI front-end compute, aimed at inference, storage and HPC deployments.

Oracle Collective Communications Library reroutes traffic across the backend Multiplanar Network when links fail

The library spreads collective traffic across OCI's backend Multiplanar Network and steers it toward healthier paths. Oracle says workloads recover instantaneously when links, ports or NICs fail.

Transparent Front-End Multiplanar Networking brings the same multiplanar design to front-end networks. It adds built-in path selection and failover, plus transparent congestion control for front-end RDMA traffic.

Acceleron combines dedicated fabrics, a converged NIC and host-level zero-trust routing

Oracle Acceleron pairs dedicated network fabrics and more direct data paths with a converged NIC, which Oracle says raises performance and lowers cost. Host-level zero-trust routing and multi-planar designs address security and resilience in multi-tenant clusters.

Pradeep Vincent, executive vice president of Compute and AI Infrastructure at OCI, said that when thousands of accelerators work together, "one congested or failed path can have consequences far beyond a single connection."

The software architecture runs on a programmable foundation that unifies networking, storage and security functions across a broad range of hardware.

Oracle noted that the release, timing and pricing of the features described may change at its sole discretion.