Oracle used AI World 2026 in Las Vegas to announce an expansion of its AI infrastructure strategy, spanning new high-performance networking, multi-gigawatt NVIDIA Vera Rubin plans, and collaborations with Google, Cerebras, Intel and OpenAI. OCI is broadening its portfolio of GPUs, specialized inference accelerators and CPUs while expanding the power, cooling, storage and networking needed to operate them.

The central networking announcement is an expansion of Oracle Acceleron with RDMA-based fabrics, resilient multiplanar networking and distributed collective communications. Oracle also plans NVIDIA Vera Rubin NVL72 superclusters at multi-gigawatt scale; Google TPUs and Cerebras inference are planned for OCI in 2027; Intel and Oracle are collaborating on future Xeon-based instances; and OpenAI integrations span cloud infrastructure and enterprise software.

The scale of Oracle’s infrastructure expansion is extraordinary. Oracle Cloud Infrastructure (OCI) is pursuing a multiyear buildout of AI data centers designed to support some of the world’s largest computing clusters, with individual projects requiring hundreds of megawatts and planned deployments extending into gigawatt-scale territory. The company reported adding 850 MW of data center capacity and deploying more than 300,000 GPUs during its fiscal first quarter of 2027. Oracle’s remaining performance obligations reached $664 billion, reflecting substantial contracted future business that will require continued investment in computing infrastructure. The expansion encompasses not only accelerator hardware but also electrical substations, power distribution, advanced cooling systems, high-density racks, optical connectivity, Ethernet switching and storage. Securing sufficient electricity, constructing facilities and integrating these systems are increasingly central to Oracle’s ability to bring new AI capacity online.

Oracle’s buildout is also closely tied to the infrastructure requirements of major AI customers, including OpenAI, and to its partnerships with NVIDIA and other semiconductor suppliers. Supporting multi-gigawatt AI deployments requires a coordinated approach to site development, power procurement, construction schedules, equipment delivery and network architecture. At these scales, networking becomes a fundamental constraint: thousands of accelerator systems must exchange data with predictable latency and high availability, while storage and CPU resources must support increasingly complex inference and agentic workloads. Oracle’s strategy is to combine large-scale physical infrastructure with a more diverse portfolio of computing architectures, including NVIDIA GPUs, Google TPUs, Cerebras inference systems and Intel processors. The announcements at Oracle AI World illustrate how the company intends to connect these investments into an integrated AI cloud platform.

Oracle Acceleron: A New Networking Foundation for AI

Oracle Acceleron combines dedicated network fabrics, programmable networking, direct data paths and host-level acceleration. The latest additions extend RDMA connectivity to front-end compute workloads and introduce mechanisms for maintaining collective communications when network components fail. Oracle’s Collective Communications Library distributes traffic across OCI’s backend Multiplanar Network, allowing alternative paths when links, ports or NICs fail.

Oracle is extending multiplanar networking to front-end infrastructure with path selection, failover and congestion control for RDMA traffic. Backend fabrics support accelerator communications, while front-end networks connect compute to storage and services. Oracle also claims up to threefold NVMe block-storage performance improvements in supported configurations.

NVIDIA: Scaling Vera Rubin to Multi-Gigawatt Infrastructure

Oracle and NVIDIA plan to expand OCI Superclusters using Vera Rubin NVL72 systems at multi-gigawatt scale. NVLink connects 72 GPUs within a rack-scale system; Spectrum-X Ethernet connects systems across the cluster; BlueField-4 DPUs and Oracle Acceleron support infrastructure services. Oracle also plans HGX Rubin NVL8 configurations and NVIDIA Vera CPUs for general-purpose agentic workloads. OpenAI plans to use Vera CPUs in production on OCI. The announced multi-gigawatt capacity is planned, not all operational today.

Google: TPU Infrastructure Comes to Oracle Cloud

Google plans to make its eighth-generation TPU portfolio available through OCI beginning in 2027, making Oracle the first external hyperscale cloud provider selected to offer Google TPUs. TPU 8t targets large-scale training, while TPU 8i targets inference and reasoning applications. Oracle will provide power, cooling and networking. Initial deployment locations, accelerator counts and capacity commitments were not specified.

Cerebras: Wafer-Scale AI Inference Coming to OCI in 2027

Oracle and Cerebras Systems plan to bring Cerebras’ high-speed inference technology to OCI in 2027. The Wafer-Scale Engine integrates compute and memory on a wafer-scale processor to reduce data movement, targeting low-latency token generation for reasoning and agentic workloads. Such applications may require repeated model calls for reasoning, tool invocation and verification, making per-step latency significant. The companies have not specified initial deployment capacity, pricing or OCI regions.

Intel: Expanding Xeon Infrastructure for Enterprise and Agentic AI

Oracle and Intel announced a multiyear collaboration covering Xeon-based compute and future OCI X14 instances planned around Intel’s Diamond Rapids server processors. Intel described AI-relevant enhancements including AMX support for FP8, FP16 and BF16 and Advanced Performance Extensions. CPU infrastructure remains important for embedding generation, reranking, application execution, orchestration, security and tool-based agent workflows. No firm availability date for Diamond Rapids OCI X14 instances was announced.

OpenAI: Extending the Partnership into Enterprise Applications

Oracle and OpenAI expanded their relationship into Oracle AI Workspace, connecting ChatGPT and OpenAI models with Oracle applications, databases and cloud services. Related initiatives include AI Security Mesh, App Composer with Codex, and access to OpenAI products through Oracle Marketplace for eligible U.S. customers. These enterprise software integrations are distinct from the companies’ underlying compute-capacity collaboration.

Analysis: Oracle’s AI Infrastructure Strategy Moves Beyond GPUs

Oracle’s announcements illustrate a shift toward heterogeneous AI infrastructure combining NVIDIA GPUs, Google TPUs, Cerebras inference systems and increasingly capable Intel and NVIDIA CPUs. Each architecture addresses different requirements across training, inference and agentic application execution. Oracle Acceleron’s RDMA, multiplanar networking and collective communications provide infrastructure for moving data across large clusters. The operational challenge is coordinating compute, memory, storage and networking while preserving predictable latency, utilization and availability.

The announcements also underscore the physical requirements of AI cloud expansion: electricity, cooling, Ethernet switching, optical interconnects, storage and data center facilities. Oracle previously reported 850 MW of additional data center capacity and more than 300,000 GPUs delivered during fiscal Q1 2027, OCI revenue of $7.4 billion (up 121% year over year) and remaining performance obligations of $664 billion. These previously reported metrics provide context rather than representing delivery of the newly announced partnerships. Key execution milestones include delivery of planned capacity, integration of multiple accelerator platforms and reliable operation at scale.