Vera Arrives: NVIDIA’s First CPU Built for Agents Lands at Top AI Labs
Original reporting by NVIDIA Blog
NVIDIA’s long-anticipated Vera CPU, purpose-built for the burgeoning era of agentic AI, has officially moved from announcement to production. Ian Buck, NVIDIA’s Vice President of Hyperscale and High-Performance Computing, personally delivered the first Vera systems to leading AI innovators Anthropic, OpenAI, and SpaceXAI on Friday, followed by Oracle Cloud Infrastructure on Monday. This pivotal moment signals a significant advancement in the infrastructure powering the next generation of AI.
Powering Agentic AI
Agentic AI, where models not only answer but actively *act* by performing complex, multi-step tasks, places unprecedented demands on computing hardware. While GPUs handle the core model training and inference, the intricate orchestration layers, tool calls, data retrieval, and constant real-time concurrent tasks are fundamentally CPU work. Traditional CPUs, designed with different priorities, struggle under this sustained load. Vera, introduced by Jensen Huang at GTC, addresses this challenge head-on. With 88 custom Olympus cores, 1.2 TB/s of memory bandwidth, and 50% faster per-core performance, Vera is engineered to keep agentic workflows moving efficiently at scale. This new class of CPU is poised to accelerate development and deployment, increasing the efficiency of the entire AI factory and enabling faster responses for users. OCI's plan to deploy hundreds of thousands of Vera CPUs by 2026 underscores its critical role in hyperscale enterprise AI.
The hand-delivery of NVIDIA’s Vera CPU systems to leading AI labs and cloud providers marks a pivotal moment, transitioning agentic AI hardware from ambitious announcement to tangible production. This dedicated CPU, designed from the ground up to address the unique demands of agentic workloads, underscores a significant evolution in the AI compute landscape. Vera’s specialized architecture, with its 88 custom Olympus cores and high memory bandwidth, is poised to unlock new levels of efficiency and responsiveness for tasks ranging from complex simulations and reinforcement learning to real-time data analysis and intricate code generation. Its immediate adoption by industry pioneers like Anthropic, OpenAI, and SpaceXAI, alongside hyperscale deployment plans from Oracle Cloud Infrastructure, validates its necessity and potential to accelerate the development and widespread deployment of sophisticated AI agents.
Broadening AI Horizons
Beyond its immediate impact, Vera signifies a broader shift in how AI infrastructure is conceived and scaled. As AI models move from simply answering queries to actively performing multi-step, concurrent tasks, the role of the CPU becomes increasingly critical for orchestration, tool calls, and long-context retrieval. Vera’s arrival ensures that the entire "AI factory" remains highly utilized, preventing CPU bottlenecks from hindering GPU performance and overall system efficiency. This extreme co-design approach, integrating Vera with GPUs and other NVIDIA technologies, solidifies the company’s vision for comprehensive, optimized AI systems that deliver unprecedented performance per watt. For enterprises, Vera’s widespread availability through cloud providers like OCI promises to democratize access to robust, production-grade agentic AI, fueling innovation and fundamentally transforming operational workflows across every sector. The age of intelligent agents, powered by purpose-built hardware, has truly begun.