NVIDIA Vera CPU is a data center processor built for agentic AI workloads, featuring NVIDIA's custom Olympus core architecture, deployable as a host CPU alongside GPUs or in standalone CPU systems. The Olympus core combines a 10-wide instruction pipeline with deep out-of-order execution and a graph prefetcher to handle branch-heavy, pointer-chasing agent code. Spatial Multithreading statically partitions execution resources between two hardware threads, reducing latency interference from background tasks. The memory subsystem adapts LPDDR5X into the SOCAMM2 module format, delivering up to 1.2 TB/s of bandwidth and 1.5 TB of capacity under 40W. A monolithic 88-core die connected by NVIDIA's second-generation Scalable Coherent Fabric provides 3.4 TB/s of core-to-core bandwidth within a single NUMA domain. Vera connects to NVIDIA Rubin GPUs via NVLink-C2C in the Vera Rubin NVL72 platform and is projected to reach market in the third to fourth quarter of 2026.