Oracle Expands AI Cloud with Multi-Vendor Hardware Ecosystem Including Next-Gen GPUs and CPUs
Oracle Cloud Infrastructure is enhancing its AI infrastructure by integrating accelerators from NVIDIA, AMD, Intel, and other tech leaders. The company is deploying advanced systems like the NVIDIA Vera Rubin platform and AMD's 6th Gen EPYC with MI455X GPUs to support large-scale AI workloads. This expansion includes new superclusters leveraging high-speed networking and massive power capacity for next-generation AI training and inference.
Oracle Cloud Infrastructure (OCI) has unveiled a major expansion of its AI infrastructure, integrating multi-vendor hardware including NVIDIA’s Vera Rubin platform, AMD’s 6th Gen EPYC CPUs with MI455X GPUs, Intel’s upcoming Diamond Rapids Xeon 7 processors, Google TPUs, and specialized systems from Cerebras and SambaNova. The deployment, announced during Oracle AI World 2026, includes the rollout of multi-gigawatt-scale NVL72 rack systems and the launch of new OCI Compute E7 instances powered by AMD’s latest EPYC chips. These systems are designed to support large-scale AI training and inference with up to 50 PFLOPs of performance, leveraging HBM4 memory and advanced networking via NVIDIA Spectrum-X and BlueField-4 DPUs. The infrastructure is backed by gigawatt-class power and cooling, with Oracle Acceleron fabric enabling resilient, multiplanar connectivity across hundreds of thousands of GPUs. OCI also confirmed plans to deploy Google TPUs starting in 2027, marking the first time a non-Google hyperscaler will offer them at scale.
This move follows years of growing demand for flexible, high-performance AI infrastructure as enterprises adopt agentic AI and complex workloads. Prior to this, OCI had already deployed NVIDIA’s H100, H200, and GB300 systems, along with AMD MI300X and MI355X GPUs, signaling a long-term commitment to heterogeneous computing. The collaboration with Intel aims to expand Xeon-based compute on OCI, particularly for enterprise applications and inference,
Oracle’s push to unify multiple AI hardware vendors under one cloud platform could reshape how enterprises access and deploy AI infrastructure. By offering NVIDIA, AMD, Intel, Google, Cerebras, and SambaNova systems on a single, scalable fabric, Oracle may reduce the complexity of managing disparate AI stacks—potentially lowering barriers for mid-sized organizations to experiment with frontier AI. This could accelerate innovation in agent-based applications, real-time inference, and large-scale model training, especially in sectors like healthcare and finance where latency and reliability are critical. The integration of Google TPUs outside of Google’s ecosystem may also challenge the dominance of cloud-native AI platforms, possibly shifting market dynamics toward more open, flexible infrastructure models. A comparable precedent is Amazon Web Services’ early adoption of custom silicon (like AWS Trainium), which helped differentiate its AI offerings but also created vendor lock-in. Ora