Oracle aims to build an AI supercluster with more than 100,000 Blackwell GPUs, targeting an unprecedented scale of compute power. This investment extends Oracle's existing deployment of thousands of Nvidia GPUs across its Oracle Cloud Infrastructure cloud service, positioning it for large-scale AI model training, according to Network World.
Oracle is aggressively building some of the world's largest AI superclusters and is recognized as a leader in cloud platforms, but its on-demand H100 GPU pricing is substantially higher than some specialized competitors. This creates a trade-off for enterprises balancing cost and integrated capabilities.
Enterprises seeking cutting-edge, integrated AI infrastructure will increasingly turn to hyperscalers like Oracle, even if it means paying a premium for the comprehensive ecosystem and future-proofed scale, potentially shifting the competitive landscape away from pure-play GPU providers for large-scale projects.
OCI's Supercluster Prowess: Raw Power and Connectivity
Oracle Cloud Infrastructure (OCI) stands out for enterprises requiring scalable, integrated cloud solutions for AI and general workloads. Recognized as a Leader in the 2025 Gartner® Magic Quadrant™ for Strategic Cloud Platform Services, OCI has aggressively scaled its GPU infrastructure, according to Oracle. The platform already deploys thousands of Nvidia GPUs and plans for a cluster with over 100,000 Blackwell GPUs. It previously built an OCI Supercluster with 65,536 Nvidia H200 GPUs. This rapid expansion, coupled with the NVL72 interconnect providing 130 TB/s bandwidth and each GB200 NVL72 offering over one exaflop of training performance, positions OCI for extreme-scale AI model training, capable of scaling to 131,072 GPUs, as reported by Network World. This commitment to massive, interconnected GPU clusters suggests Oracle intends to be a primary enabler for the next generation of foundational AI models, where sheer compute power and seamless data flow are paramount.
While OCI offers high-performance AI infrastructure and significant GPU investments, its on-demand H100 GPU pricing is higher than niche providers. For example, H100 on-demand instances cost $10/GPU/hr, with an 8x H100 node at $80/hr, totaling $57,600 per month for 720 hours, according to Spheron. However, OCI's VM instances (4 vCPUs, 16 GB RAM) are $54/month, 2.3X less than AWS/Azure, and Kubernetes clusters (64 vCPUs, 512 GB RAM) are $3,507/month, 2.3X less than AWS/Azure, according to Oracle. GPU limit increases may require 1-3 business days.
