Oracle Opens Vera Rubin Cloud at Gigawatt Scale

Oracle is treating agent infrastructure as more than a GPU problem: keeping accelerators busy requires dedicated CPU capacity for everything happening between model calls.

• –Customers can scale GPU and CPU resources independently, potentially reducing expensive idle accelerator time.
• –Vera Rubin NVL72 brings 72-GPU rack systems, while planned HGX NVL8 configurations offer a smaller x86-compatible deployment path.
• –Vera CPUs are designed for orchestration, sandbox execution, vector retrieval, and other agent workloads that compete with inference for infrastructure.
• –CUDA compatibility gives developers a relatively smooth path from existing NVIDIA deployments to Rubin-based OCI clusters.
• –The multi-gigawatt roadmap is strategically important, but its value will depend on Oracle’s ability to deliver power, cooling, networking, and reliable capacity at scale.