AI startup Perplexity has confirmed plans to use Nvidia’s Vera central processing units, as the chipmaker pushes into data center CPUs built for AI agents and a market long led by Intel and AMD.
TL;DR
- Perplexity plans to use Nvidia’s Vera CPU for AI agent workloads.
- Perplexity said Vera handled AI agent coding tasks about 1.5x faster than traditional CPUs.
- Nvidia says Vera is built for agentic AI, reinforcement learning, and data processing.
- Perplexity did not disclose how many Vera CPUs it plans to buy.
AI search startup Perplexity has confirmed that it plans to use Nvidia’s new Vera central processing units, as the chipmaker looks to expand beyond its dominant AI accelerator business and compete in a CPU market traditionally led by Intel and Advanced Micro Devices.
The move was confirmed by Perplexity Vice President for Computer Enterprise and Infrastructure Nate Kupp in an interview with Reuters. The company did not disclose how many Nvidia CPUs it plans to buy, so the scale of the deployment remains unknown.
The core reason behind Perplexity’s interest appears to be AI agent performance. AI agents do not behave like traditional human users who open an app, perform a task, and step away. They can continue running tool calls, code execution, sandboxing, retrieval, and verification loops without long idle periods, creating a different type of CPU demand.
Perplexity said Nvidia’s Vera CPU carried out AI agent coding tasks about 1.5 times faster than traditional CPUs. “Vera really stood out to us as just like a dead-on fit for a lot of the core workloads that we have,” Kupp said.
For Nvidia, this is more than a single customer win. The company has positioned Vera as “the CPU for agents,” a processor designed for the CPU-heavy work that happens around AI models, including code execution, tool use, sandboxing, analytics, data pipelines, orchestration, and reinforcement learning.
Nvidia said Vera is now in full production and can deliver 1.8x faster task completion compared with x86 CPUs across workloads such as agentic AI, reinforcement learning, and data processing. The company also said Vera builds on its Grace CPU business, which has seen nearly 2.5 million shipments to date.
The technical pitch rests on high single-threaded performance at scale. Nvidia says Vera uses 88 custom Olympus cores, Spatial Multithreading, and an LPDDR5X memory subsystem delivering up to 1.2TB/s of bandwidth. The goal is to reduce time spent waiting on CPU-bound steps and keep AI accelerators better utilized.
Nvidia has also said Vera systems will be available through system builders and cloud partners starting this fall. The company has named several AI labs, cloud providers, and infrastructure players as adopters or evaluators, including OpenAI, Anthropic, Oracle Cloud Infrastructure, ByteDance, CoreWeave, Lambda, Nebius, and Nscale.
Topics for more insights:
Nvidia has also put a financial target around the product. Reuters reported that Nvidia expects to generate $20 billion in sales from Vera by the end of its fiscal year, as it looks for growth beyond AI-specific chips while AI companies increasingly explore custom hardware.
The broader context is clear: as AI agents become more common, the infrastructure bottleneck may not be limited to GPUs. If agents need fast CPUs to run tools, execute code, and complete reasoning loops, Nvidia wants Vera to become a core part of the AI factory stack.
For Perplexity, the Vera plan signals an effort to optimize the infrastructure behind its AI products. For Nvidia, it adds another AI-native company to its CPU push, while keeping the company’s hardware story tied closely to the next phase of agentic AI.
