Nvidia’s Vera CPU has arrived as a standalone cloud product, giving Oracle customers access to the company’s first hyperscale CPU deployment without requiring a Rubin GPU. Oracle and Nvidia announced general availability for Vera CPUs and Vera Rubin NVL72 systems at Oracle AI World, while outlining plans to expand OCI Superclusters to multi-gigawatt scale. The Oracle Cloud Vera rollout puts Nvidia’s Arm-based processor directly into a market still dominated by Intel Xeon and AMD EPYC.
The initial deployment reportedly includes 150,000 Vera CPUs, with bare-metal and virtual instance options aimed at workloads that can use dedicated CPU capacity without a GPU attached. Oracle specifically positions the machines for agent orchestration, retrieval, analytics and other CPU-heavy stages of AI applications. OCI Kubernetes Engine can scale CPU and GPU resources independently, while Oracle’s Acceleron infrastructure is designed to isolate workloads that execute untrusted, AI-generated code.
Vera’s pitch is less about replacing every general-purpose server than about feeding modern AI systems efficiently. Each chip combines 88 Olympus cores with Spatial Multithreading and an LPDDR5X memory subsystem rated for up to 1.2 terabytes per second of bandwidth. Nvidia says that figure delivers twice the bandwidth at half the power of traditional CPU memory systems, a particularly relevant advantage for database retrieval, large documentation searches and other memory-bound agentic tasks.
That design also explains why Nvidia is entering the CPU market through AI infrastructure rather than trying to win the entire server business at once. Vera uses a monolithic die and distributed cache-and-memory topology, allowing software configured for the architecture to keep many concurrent operations supplied with data. Early testing cited in the launch coverage found Vera’s memory bandwidth more than doubled that of a dual Intel Xeon 6980P setup using 8,800-MT/s MRDIMMs, although those results were produced on a Vera server rather than Oracle’s cloud hardware.
The processor will also appear inside Nvidia’s larger Vera Rubin systems. Oracle’s NVL72 configuration combines 72 Rubin GPUs with 36 Vera CPUs, linked through Nvidia’s sixth-generation NVLink and supported by ConnectX-9 networking, BlueField-4 DPUs and Spectrum-X Ethernet. Nvidia describes that rack-scale platform as a system for frontier-model training, high-throughput inference and agentic applications, while Oracle plans to expand its Superclusters beyond the initial availability footprint.
Oracle is not alone in moving quickly. Google Cloud has also made Vera Rubin generally available, and Nvidia’s launch materials identify AWS, Microsoft Azure, CoreWeave, Lambda, Nebius, Nscale and other providers as part of the broader Vera rollout. The important shift is that Vera is no longer merely a CPU component hidden inside Nvidia’s GPU racks: Oracle is making it available as an independent compute choice, giving developers a way to match high-bandwidth CPU capacity to the parts of an AI workload that do not need GPU acceleration.
