Oracle and AMD have announced a major expansion of their long-standing, multi-generational partnership aimed at helping customers with their most advanced capabilities. From Calendar Q3 2026, Oracle will be the first hypercluster to offer a publicly available ai supercluster powered by 50,000 amd Series GPUS, with further expansion planned for 2027 and beyond.
The move builds on several years of integrated development, starting with the launch of AMD’s virtual environments on Oracle Cloud Infrastructure (OCI) in 2024 and the next release of OCI compatable with AMD COBUTE with AMD MI355X GPUS. These contributions feed into the zetascale OCI supercluster.
The new AI superclusters will use AMD’s “Helios” Rack Design, which includes AMD’s native MI450 GPUS, the “Venice” Epyc CPUS, and an advanced network processor codenamed “volcano.” This rack-scale configuration is designed to deliver high performance, power efficiency and capacity for large-scale AI training and AI workloads.
Mahesh Thiagagajan, President of Product Development, Oracle Cloud Infrastructure, said: “Our customers are developing some secure AMD services, and need to integrate OCI parameters with OPI. By working with long-term AMD partners – from EPYC to AMD AcvurAtors benefit – we continue to deliver the Best Prece-Performance, Open, secure and cloud-based in collaboration with AMD to meet the needs of customers in the next era of AI. “
Forrest Norrod, Executive Vice President and General Manager of AMD’s Data Center Solutions Group, AMD CPUS, and implementing the next generation of AI in the cloud. Together, AMD and Oracle accelerate AI with open, optimized, and secure systems designed for AI big data centers. “
The powerful MI450 GPU features are engineered to support complex tasks including AI and advanced AI. Each GPU delivers 432 GB of HBM4 memory and 20 / s of memory Bandwidth, enabling users to train and fit models 50 percent larger with memory. The “Helios” Rack Design allows for dense, cool, 72-GPU racks and includes Ualoe Scale-Up communication with a Nethernet-Out Network for low latency and high throughput in the GPU Cluster.
In addition, the supercluster architecture includes advanced EPYC “Venice” CPUS-based cores that provide privacy computing and security features. AMD Pencil DPUS enables scalable networks that provide high-rate data throughput, improve performance, and improve security. The design allows the GPUS to be equipped with 800 AI-NITS of 800 GBPS AI-Nics (VULCANO), supporting informal communication, with the speed of systematic planning with standards such as roce and UEC.
The architecture also features an open fabric of Ualink and Ualoe, which facilitates the distribution of data across the GPUS within the racks without moving the CPUS. rocm’s open software stack facilitates vendor flexibility and supports popular AI and HPC programming, while advanced partitioning and virtualization features enable GPU resource allocation.
Alongside the MI450 deployment, Oracle also announced the general availability of Compretes OCI instances powered by the AMD MI355X GPUS environment. This will be part of the Zettascale OCI Supercluster capable of scaling to 131,072 GPUS, providing customers with the best performance and flexibility for AI efficiency.
