Oracle to Launch AI Supercluster with 50,000 AMD GPUs in 2026

Oracle will introduce a publicly available Artificial Intelligence (AI) supercluster featuring 50,000 AMD Instinct MI450 Series GPUs, beginning in calendar Q3 2026. This initiative marks Oracle as the first hyperscaler to deploy such a system.

During Oracle AI World, Oracle and AMD announced expansion of their collaboration to advance AI capabilities. Oracle Cloud Infrastructure (OCI) serves as the launch partner for this supercluster, aligning with growing demand for enhanced AI processing power.

The supercluster is designed to address the limitations of current AI clusters, enabling customers to utilize flexible and scalable computing solutions for large-scale AI training and inference. The AMD “Helios” rack design will support this supercluster, integrating next-generation EPYC CPUs and advanced networking technologies.

Mahesh Thiagarajan, executive vice president at Oracle Cloud Infrastructure, stated, “Our customers are building some of the world's most ambitious AI applications, and that requires robust, scalable, and high-performance infrastructure.” He emphasized the importance of this partnership in delivering optimized solutions for demanding workloads.

Forrest Norrod, executive vice president at AMD, noted that the collaboration delivers new capabilities for AI model training and deployment, providing a foundation for large-scale data centers. The AI supercluster will enhance performance, scalability, and efficiency, as it features DPU-accelerated converged networking and advanced memory management.

In addition to the AI supercluster, OCI announced General Availability (GA) of OCI Compute with AMD Instinct MI355X GPUs, expanding options for customers aiming to leverage AI capabilities at scale.