Perplexity Bets on Nvidia's Vera CPU for Agentic AI Workloads

Perplexity, the AI search engine handling over 400 million queries per month, is adopting Nvidia's new Vera CPU to run its agent workloads. The company claims Vera delivers 1.5x faster single-threaded performance on agentic coding tasks compared to traditional CPUs, as stated by Nate Kupp, Perplexity's VP of enterprise infrastructure.

What is Vera?

Vera is the CPU half of Nvidia's Vera Rubin platform. It's a custom Arm-based design with dozens of Nvidia-built cores, paired with fast, low-power memory and a high-bandwidth link to Nvidia's accelerators. The key pitch is reduced data movement friction between CPU and GPU when both are designed by the same company.

Why CPUs Matter for AI Agents

Agentic AI systems chain together multiple model calls, tool uses, and code executions. The CPU orchestrates this pipeline and can become a bottleneck. Faster single-threaded performance directly reduces latency per query, allowing Perplexity to serve more queries for the same cost or enable more complex agentic features without exploding costs.

Nvidia's Market Play

Nvidia expects about $20 billion in Vera CPU sales this fiscal year, its entry into the $200 billion general-purpose server processor market long dominated by Intel and AMD. The move is partly defensive: major customers like OpenAI are designing their own AI chips, so a CPU line lets Nvidia sell more per rack even where its accelerators face competition. Nvidia has also been offering startups compute now and payment later to lock in demand.

Perplexity's Infrastructure

Perplexity already relies heavily on Nvidia hardware: H100 GPUs, Triton inference server, and TensorRT. Adding Vera CPUs tightens that integration. The company's 400 million monthly queries each go through an inference pipeline that benefits from any per-query efficiency gain.

The Bigger Picture

As AI search rivals fold generative answers into everything, the cost of running those answers at scale becomes a competitive differentiator. Cheaper inference is the lever. Faster CPUs let a company serve more queries for the same spend, or push into agentic features that string together many calls without the bill spiralling. For Perplexity, which has grown fast and raised heavily, holding down cost per query is existential.

What's Next

Neither Nvidia nor Perplexity has disclosed volume shipping dates or how much of Perplexity's fleet will move to Vera. The harder test: winning over enterprises that already trust Intel and AMD in a way they don't yet trust Nvidia's CPUs. But with early adopters like Anthropic, OpenAI, and now Perplexity, Nvidia has a credible beachhead.