Oracle and AMD Launch Next-Generation AI Superclusters
Oracle and AMD recently unveiled a significant expansion of their long-term partnership, aimed at enhancing AI capabilities for their customers. Starting in the third quarter of 2026, Oracle will become the first hyperscaler to provide a publicly available AI supercluster powered by an impressive 50,000 AMD Instinct MI450 Series GPUs. Further plans for expansion are expected in 2027 and beyond.
Building on a Strong Foundation
This initiative is a continuation of the collaborative efforts between Oracle and AMD, which began years ago. It follows the earlier integration of AMD Instinct MI300X GPUs into the Oracle Cloud Infrastructure (OCI) in 2024, and the introduction of OCI Compute featuring AMD Instinct MI355X GPUs. These developments form part of the expansive zettascale OCI Supercluster, a cutting-edge cloud infrastructure designed to meet the growing demands of AI applications.
Advanced Architecture for AI Workloads
The new AI superclusters will utilize AMD’s innovative “Helios” rack design, which integrates the latest MI450 GPUs and the upcoming “Venice” EPYC CPUs alongside advanced networking technology codenamed “Vulcano.” This vertically optimized architecture is engineered for high performance and energy efficiency, making it ideal for extensive AI training and inference tasks.
Mahesh Thiagarajan, Executive Vice President of Oracle Cloud Infrastructure, emphasized the importance of robust infrastructure for ambitious AI applications. He remarked that combining AMD’s latest processor technologies with OCI’s secure and flexible platform empowers customers to push their AI projects further than ever before. Over the past decade, Oracle and AMD’s collaboration has consistently delivered competitive price-performance ratios, reinforcing their commitment to providing secure and scalable cloud solutions for this evolving AI landscape.
Powering AI Innovation
Forrest Norrod, Executive Vice President and General Manager of AMD’s Data Center Solutions Business Group, echoed Thiagarajan’s sentiments. He stated that their combined efforts are accelerating AI advancements in the cloud. With AMD’s state-of-the-art Instinct GPUs, EPYC CPUs, and advanced networking capabilities, Oracle customers are obtaining new tools for effectively training and deploying future AI solutions.
The MI450 GPU configurations are specifically designed for demanding workloads, including generative AI and high-performance computing (HPC). Each GPU boasts up to 432 GB of HBM4 memory and a remarkable memory bandwidth of 20 TB/s, which enables users to work with models that are 50% larger while remaining entirely in-memory. Moreover, the “Helios” rack design facilitates dense, liquid-cooled setups with up to 72 GPUs in a single rack, ensuring optimal connectivity and performance across GPU clusters.
Enhanced Security and Performance Features
The architecture of the supercluster also incorporates advanced head nodes based on the EPYC “Venice” CPUs, offering confidential computing and various security enhancements. AMD Pensando’s Data Processing Units (DPUs) aid in converged networking, enabling efficient data ingestion and boosting performance while enhancing security. The setup accommodates GPUs with up to three 800 Gbps AI-Network Interface Cards (NICs), supporting high-speed, lossless networking with established standards like RoCE and UEC.
Moreover, the supercluster design features an open UALink and UALoE fabric that simplifies data sharing across GPUs within a rack, eliminating the need to route traffic through CPUs. The open-source ROCm software stack supports a wide range of AI and HPC frameworks, allowing for vendor flexibility. In addition, advanced partitioning and virtualization capabilities promote secure multi-tenant resource sharing, helping to fine-tune GPU workloads.
Expanding Capabilities with OCI Compute Instances
In conjunction with the deployment of MI450 GPUs, Oracle has also announced the general availability of OCI Compute instances powered by AMD Instinct MI355X GPUs. These components will play a crucial role in the zettascale OCI Supercluster, which is designed to scale to an impressive 131,072 GPUs, offering customers flexible optimization options for their AI workloads.
Through these advancements, Oracle and AMD are not just pushing the boundaries of technology but also ensuring that their customers have the tools they need to thrive in a rapidly evolving digital landscape.
