Oracle Cloud

Full briefing

Oracle Corporation (NYSE: ORCL) operates Oracle Cloud Infrastructure (OCI), which has emerged as a major force in AI infrastructure despite entering the cloud market later than AWS, Azure, and Google Cloud. Oracle's differentiation is its focus on GPU-dense, high-performance computing clusters specifically optimized for AI training workloads. In May 2024, Oracle CEO Safra Catz and Chairman Larry Ellison unveiled a cluster of 131,072 NVIDIA H100 GPUs — the largest publicly disclosed single AI training cluster in the world at that time — interconnected using NVIDIA's Quantum-2 InfiniBand at 3,200 Gbps. This cluster was built on OCI's 'Supercluster' architecture, which uses RDMA (Remote Direct Memory Access) networking to achieve near-zero latency between GPU nodes. Oracle's AI cloud strategy is distinct from AWS and Azure in several ways: (1) Oracle focuses on 'clean slate' AI infrastructure with no legacy compute debt; (2) OCI offers a 'bring your own model' approach without the same proprietary AI service lock-in as competitors; (3) Oracle has signed multi-year deals with AI training companies (CoreWeave, X/Twitter's Grok, Cohere) that use OCI as sovereign AI infrastructure outside the AWS/Azure duopoly. OCI also offers Arm-based compute via Ampere A1 virtual machines — these use Ampere Computing's Altra processors (80-core Arm Neoverse N1, fabricated by TSMC) providing cost-efficient CPU compute alongside GPU instances. This TSMC fabrication link makes Oracle a downstream customer of TSMC's advanced manufacturing. Oracle's AI infrastructure investment is accelerating: it has committed to $8+ billion in capital expenditures for cloud data center expansion through 2025, with GPU density as the primary driver.