Oracle Cloud Infrastructure (OCI) announced the deployment of a next-generation AI infrastructure designed to support frontier-scale AI by integrating accelerators from multiple vendors, including NVIDIA, AMD, and Intel. The company stated that OCI now spans more than 200 cloud services, with plans to leverage large-scale deployments of hardware across the next decade.
For NVIDIA, OCI will deploy a multi-gigawatt scale of Vera Rubin NVL72 systems for AI training, inference, and agentic applications, with general availability starting immediately. The deployment will also include NVIDIA Vera CPU servers and NVL8 "HGX" systems. OCI Superclusters combine these NVL72 rack-scale systems with Oracle Acceleron networking, NVIDIA Spectrum-X Ethernet networking featuring NVIDIA Spectrum-6 switches, and NVIDIA BlueField-4 DPUs to train and serve large AI models. Each rack connects 72 GPUs via NVIDIA NVLink scale-up networking, and OCI’s network scales across hundreds of thousands of GPUs.
AMD's 6th Gen AMD EPYC CPUs will power OCI's Compute E7 instances. The Helios AI rack will utilize AMD's enterprise architecture, featuring the latest Instinct MI455X GPUs. OCI Compute E7 instances, powered by these processors, are designed to provide advanced capabilities for agent orchestration, concurrent agent sandboxes, retrieval services, and in-memory caches, supporting web, API, and enterprise application tiers. OCI's infrastructure already incorporates deployments from both NVIDIA (including P100, V100, A100, H100, H200, GB200, GB300) and AMD (including MI300X, MI355X).
Intel is also expanding its presence through a multiyear collaboration with Oracle. This involves deploying the next-generation OCI X14 platform using upcoming Diamond Rapids "Xeon 7" chips. These chips are intended to boost AI performance via dedicated general-purpose and AI-specific acceleration, alongside Intel AMX (with FP8, FP16, BF16 support) and Intel APX (Advanced Performance Extensions).
Furthermore, OCI announced it will bring Google TPUs to the platform, positioning itself as the first hyperscaler outside of Google to deploy them, starting in 2027 and scaling to GW-scale over multiple years. The latest generation TPUs feature on-die Inter-Chip Interconnect (ICI) routing paired with Optical Circuit Switching (OCS), allowing the control plane to dynamically route optical paths around faults in real time without job restarts, sustaining over 97% goodput across thousands of chips.
OCI also plans to leverage Cerebras and SambaNova to accelerate AI inference at low latency. Additionally, during its Q1 FY2027 earnings, the company reported delivering an additional 850MW of data center capacity.
