NVIDIA has unveiled the Vera CPU, designed specifically for agentic AI and reinforcement learning. It delivers 50% faster performance and double the energy efficiency, already adopted by Alibaba, Meta, ByteDance, Oracle Cloud, CoreWeave, and Lambda.

Vera features 88 Olympus cores, high-bandwidth memory, and advanced multithreading, supporting large-scale AI deployments. Liquid-cooled racks can host over 22,500 concurrent CPU environments, allowing enterprises and research labs to scale agentic AI efficiently.

The CPU integrates with NVIDIA GPUs via NVLink-C2C and includes ConnectX SuperNIC and BlueField-4 DPUs to enhance networking, storage, and security. Early users like Cursor and Redpanda report major gains in AI agent throughput and real-time data processing.

High performance, energy efficiency, and GPU integration make Vera a new standard for faster, scalable, and responsive AI systems. The platform supports coding assistants, reinforcement learning, and large-scale data, making it suitable for enterprise and scientific use.