Nvidia's Vera CPU signals rise of computing infra built for agentic AI era

Nvidia introduced Vera, its first CPU designed exclusively for AI agents, at GTC Taipei, marking a shift from its traditional GPU dominance. The chip targets autonomous systems that run code, process data, and coordinate tasks—workloads increasingly critical as AI evolves beyond query-based responses. Vera uses Nvidia’s custom Olympus core, a departure from past reliance on licensed Arm designs, allowing tailored optimization for AI workloads. Unlike general-purpose CPUs, Vera prioritizes agentic tasks like orchestration, reinforcement learning, and memory management, delivering up to 1.8x faster performance than x86 processors. Nvidia CEO Jensen Huang emphasized that AI agents will drive future computing demands, positioning Vera as the foundation for this era. The chip’s 88 Olympus cores and 1.2TB/s memory bandwidth enhance efficiency for AI-driven workflows. Vera integrates seamlessly with Nvidia’s GPU ecosystem, addressing bottlenecks in data movement between CPU and GPU—a key limitation in modern AI systems. The design reflects Nvidia’s vision of data centers as ‘AI factories,’ where CPUs act as coordinators for large-scale model execution. This tight coupling aims to accelerate AI workflows by reducing latency in task coordination. The launch underscores Nvidia’s long-term strategy to expand beyond GPUs, leveraging Vera to capture CPU market share in AI infrastructure. Previous efforts like the Grace CPU used off-the-shelf Arm cores, but Vera’s custom architecture grants Nvidia full control over performance and integration. This move could reshape enterprise computing, particularly for cloud and AI-driven enterprises.