NVIDIA Unveils Vera CPU Built for the Age of Agentic AI

NVIDIA has taken step beyond GPUs with the launch of Vera, a new CPU the company says was designed from the ground up for agentic AI and reinforcement learning. Announced on March 16, 2026 at GTC, Vera is being positioned as the world’s first CPU purpose-built for a new class of AI workloads where systems do more than generate responses they plan, reason, orchestrate tools and act with greater autonomy. 

According to NVIDIA, Vera delivers 2x the efficiency and 50% faster performance than what it describes as traditional rack-scale CPUs. That matters because the AI race is no longer just about model training. As enterprises move deeper into inference, AI assistants and autonomous agents, the hardware underneath these systems increasingly needs to handle orchestration, memory movement, data processing and real-time coordination at massive scale. Vera is NVIDIA’s answer to that shift. 

The company says Vera is built to support the growing demands of “AI factories,” where thousands of models, agents and services are expected to run together. NVIDIA said a Vera CPU rack integrating 256 liquid-cooled Vera CPUs can sustain more than 22,500 concurrent CPU environments, with each running independently at full performance. That gives the launch a much bigger meaning than just another chip release. NVIDIA is effectively arguing that the future of AI infrastructure will require specialized CPUs just as much as specialized GPUs.

Vera is also arriving with major ecosystem backing. NVIDIA says companies including Alibaba, Meta, Oracle Cloud Infrastructure, Dell, HPE, Lenovo and Supermicro are already working with the platform, a sign that the company wants Vera to become part of mainstream enterprise and cloud AI deployments rather than remain a niche in-house component. The rollout also ties into NVIDIA’s broader Vera Rubin platform strategy, which the company presented as the next frontier for agentic AI systems. 

The timing is important. At GTC 2026, NVIDIA made it clear that inference and autonomous AI agents are becoming the next major battleground in AI computing. Reuters reported that CEO Jensen Huang used the event to argue that NVIDIA now sees at least a $1 trillion AI chip revenue opportunity through 2027, while unveiling products including Vera as part of its deeper push into inference infrastructure. 

For NVIDIA, Vera is about proving it can shape more of the full AI stack. The company built its dominance on GPUs, but modern AI data centres increasingly need tightly integrated systems that combine compute, networking, storage and orchestration. Vera suggests NVIDIA wants to own more of that foundation, especially as AI agents become heavier users of system-level compute and coordination rather than just raw matrix math. 

After all said and done, this is just NVIDIA betting that the age of agentic AI will need new kinds of CPUs not just faster old ones. And if the company’s claims hold up in the real world, Vera could become one of the more important pieces of infrastructure behind the next generation of enterprise AI.