NVIDIA Vera Standalone CPU Enters Production for Agentic AI Deployment Featuring Olympus Cores and High Memory Bandwidth for Enterprise Cloud Scaling
NVIDIA has begun the rollout of its Vera standalone CPU into the production space. After CEO Jensen Huang announced the chip in March as a multi billion dollar growth prospect for the company, the hardware has arrived to its first set of customers. NVIDIA VP Ian Buck delivered the systems directly to Anthropic, OpenAI, SpaceXAI and finally OCI to start the production workload of next generation agentic artificial intelligence.
The Vera CPU signifies a turning point in the way that hardware serves up the infrastructure that feeds AI factories. From simple to action oriented models, the computational load now heavily favors the CPU. As noted by Ian Buck when delivering the systems to the clients, while GPUs are responsible for accelerated computation, control logic, and calls for data are managed by the CPU, thus the new architecture was designed specifically to handle such orchestrating functions and large context state management.
The Vera CPU contains 88 custom NVIDIA designed Olympus cores, offers 1.2TB/s memory bandwidth, and shows a 50% higher per core performance under heavy load than previous generations. The design allows for tasks that typically require significant computational power and processing time to be resolved more rapidly, hence more efficiently in an AI factory setting. It was designed to achieve rapid task completion in high throughput reasoning scenarios, where a model needs to generate Python code or execute a simulation to reach the right conclusion.
The emergence of agentic AI necessitates the arrival of the next CPU moment for the AI factory: The shift from a computing model centered around response to action. The Vera is engineered to meet that need and keep computation flowing efficiently at scale.
The Vera CPU is also a host processor on the Vera Rubin NVL 72 platform, and connects to Rubin GPUs with a 2nd generation NVLink C2C. This system leverages a unified memory architecture to efficiently utilize all of the components, feeding the GPUs the proper data and instructions to be executed while maintaining higher computational unit utilization due to its energy efficiency relative to traditional data center infrastructure at 2X energy efficiency than previous methods.
Oracle Cloud Infrastructure (OCI) becomes the first provider to incorporate Vera at the hyperscale level. The company plans to deploy hundreds of thousands of units starting in 2026, to assist with the production workflow of agentic AI. According to OCI head of product management, Karan Batta, the sustained performance delivered by Veras architecture is exactly what will be required for the next generation of enterprise AI. The technology provides enterprises with proven, on demand infrastructure that can efficiently perform agent based simulation and RL workloads at scale.
The shipment of hardware to facilities like Anthropic and SpaceXAI demonstrates the growing industry demand for specialized CPU hardware. James Bradbury, at Anthropic, has indicated that scaling computational power is essential to continuing the advancements in large scale models, particularly for agentic purposes. Through its 88 Olympus cores and substantial memory bandwidth, Vera sets the stage to become the orchestration engine for acting AI agents.
