Oracle and AMD partner on 50,000 MI450 GPU supercluster at DeepFest 2026

RIYADH — In a significant expansion of enterprise cloud capabilities, Oracle and AMD have formalized plans to deploy a large-scale AI supercluster on Oracle Cloud Infrastructure (OCI). The strategic agreement, announced against the backdrop of DeepFest 2026 in Riyadh, positions Oracle to become the first hyperscaler to offer a publicly available AI supercluster powered by an initial deployment of 50,000 AMD Instinct MI450 Series GPUs. This hardware rollout is scheduled to begin in the third quarter of 2026, targeting the most demanding AI training workloads. The collaboration underscores a broader shift in how major cloud providers are structuring their high-performance computing environments. By integrating AMD's latest silicon into OCI's established architecture, the initiative aims to address the growing bottleneck in AI infrastructure availability. AMD, participating as a Platinum Sponsor at DeepFest 2026, is actively showcasing this capability and partnership as a cornerstone of its enterprise strategy. The sheer scale of the planned deployment reflects the escalating computational demands of modern artificial intelligence development. As organizations increasingly rely on complex neural networks and foundational models, the necessity for robust, scalable infrastructure has become paramount. The decision to base this specific supercluster entirely on AMD silicon highlights an increasing diversification of hardware suppliers among top-tier cloud providers. This diversification strategy mitigates supply chain risks and provides enterprise customers with a broader array of compute options tailored to specific workload requirements. The expansion of Oracle Cloud Infrastructure with these specialized processors represents a substantial capital commitment to supporting next-generation enterprise AI workloads. This strategic move is expected to reshape the competitive dynamics of the cloud computing market by providing a viable alternative for organizations requiring massive, synchronized computational power. The technical foundation of this deployment relies heavily on AMD's specialized infrastructure designs. The supercluster will utilize the AMD "Helios" rack design, which has been specifically engineered to optimize the physical and thermal constraints of high-density GPU deployments. Managing the power consumption and heat dissipation of 50,000 high-performance accelerators requires sophisticated engineering at the rack and data center level. The Helios design addresses these challenges by providing a structured framework that maximizes cooling efficiency while maintaining the dense physical proximity necessary for rapid inter-processor communication. Each MI450 GPU within the cluster provides up to 432 gigabytes of high-bandwidth HBM4 memory, representing a substantial leap in capacity over previous generations. This memory density is critical for training large language models and processing complex datasets without encountering severe latency penalties. The ability to hold larger portions of a model directly in the GPU's memory significantly reduces the time spent transferring data between storage and compute units. To support the massive data throughput required by 50,000 interconnected GPUs, the system incorporates advanced "Vulcano" networking technology. This architecture enables direct hardware-coherent networking via the UALink protocol. The implementation of UALink is designed to facilitate rapid, low-latency communication between individual processors, effectively allowing the entire cluster to operate as a unified compute engine. This level of integration is essential for minimizing the overhead typically associated with distributed computing tasks across large-scale data centers. By ensuring that data can flow directly between any two GPUs in the cluster, the Vulcano networking architecture prevents the communication bottlenecks that often plague massive AI deployments. This seamless connectivity is a critical differentiator for organizations looking to train the next generation of foundational models, which require large-scale levels of synchronized computational power. The scale of this deployment represents a material commitment from both organizations to capture a larger share of the generative AI infrastructure market. For Oracle, the integration of 50,000 AMD Instinct MI450 GPUs reinforces its position as a primary destination for enterprise AI workloads. By offering a differentiated hardware stack, Oracle Cloud Infrastructure can attract customers who require massive scale but are seeking alternatives to existing market monopolies. This move also aligns with Oracle's broader strategy of providing specialized, high-performance cloud services that cater to the specific needs of data-intensive industries. The availability of such a massive supercluster on a public cloud platform democratizes access to resources that were previously only available to a handful of well-funded research inst