Skip to main content

NVIDIA's New Vera CPU: Built for AI Agents, 1.5x Faster

NVIDIA has officially launched its Vera CPU, a chip that breaks away from the company's GPU-dominated legacy. Unlike traditional processors from Intel and AMD, Vera is built from the ground up for the age of AI agents—autonomous programs that can write code, execute tasks, and manage complex workflows without human intervention.

Why Traditional CPUs Fall Short

AI agents operate in continuous loops, constantly evaluating and acting. But Intel and AMD CPUs were designed for the laptop era, optimized for intermittent human-computer interactions. When faced with the relentless demands of AI agents, they struggle to keep up. NVIDIA's blog explains the bottleneck: slower execution speeds reduce the number of effective evaluations per reinforcement learning cycle, lengthening training times and increasing latency. Worse, insufficient speed can lead to KV cache loss, driving up compute costs in a vicious cycle.

Vera's Performance Leap

Vera directly tackles these pain points. Its core speed is 1.8 times faster than traditional CPUs, allowing up to 85% of evaluation tasks to complete within the same time window. In real-world tests, Perplexity's Vice President of Infrastructure, Nate Kupp, found that Vera executed AI agent coding tasks about 1.5 times faster than conventional chips. Faster execution also means richer reinforcement learning feedback, generating higher-quality training signals and shortening overall training cycles.

Industry Adoption

Perplexity, an AI search startup, has already announced plans to adopt Vera. While the company hasn't disclosed purchase quantities, NVIDIA has confirmed that OpenAI, Anthropic, and Oracle are also on board. As leading AI labs develop their own chips, NVIDIA is expanding its hardware portfolio, positioning Vera as a key step from "GPU dominance" to "full-stack computing provider." If AI agents become the core of next-generation software, custom-built hardware like Vera could reshape the entire computing landscape.

Key Points

  • Vera is a new CPU architecture designed specifically for AI agents, not a GPU extension.
  • It offers 1.8x faster core speed and 1.5x faster coding task execution compared to traditional CPUs.
  • Early adopters include Perplexity, OpenAI, Anthropic, and Oracle.
  • The chip addresses bottlenecks in reinforcement learning loops, reducing training time and compute costs.