Nvidia and Oracle say Vera Rubin NVL72 and standalone Vera CPUs are generally available on OCI. Scaling CPUs independently from GPUs aims to reduce waits while agents execute code, search and use other tools between model calls. The companies also describe multi-gigawatt supercluster expansion and a smaller NVL8 configuration, with NVL8 still a planned offering. Vendor CPU benchmark gains are not the same as measured OCI improvements, and more CPU capacity will not resolve bottlenecks in data systems or remote services.