Perplexity registers for Nvidia’s Vera CPU as the chip manufacturer advances beyond AI accelerators.

      Perplexity has emerged as one of the first prominent AI companies to announce that it will utilize Nvidia's new Vera CPU, which Nvidia hopes will extend its reach beyond the accelerators that have helped it become the world's most valuable company. The AI search firm indicated its intention to run agent workloads on Vera, joining an initial group of users that Nvidia is eager to highlight.

      Vera is the processing component of Nvidia's Vera Rubin platform, an Arm-based architecture that features numerous custom cores designed to support the company's next generation of accelerators. This marks Nvidia’s most significant step into the conventional server processor market, a domain historically dominated by Intel and AMD, and the company has already identified Anthropic and OpenAI as some of its first users.

      Perplexity advocates for the chip based on its speed for specific tasks. Nate Kupp, the firm's vice president of enterprise infrastructure, stated that Vera executes agentic coding tasks approximately 1.5 times faster than the standard CPUs Perplexity had previously employed, noting that its single-threaded performance aligns well with the nature of the work.

      That work is substantial, as Perplexity now processes over 400 million search queries monthly, with each query passing through an inference pipeline that has relied on Nvidia hardware—such as H100 GPUs—and the company's Triton and TensorRT software from the outset.

      The importance of agentic features emphasizes why the CPU is crucial in this context. As an AI system integrates numerous model calls, tool usages, and code executions to accomplish a task, the general-purpose processor that orchestrates these operations can become a bottleneck comparable to the accelerator handling the complex computations.

      Vera itself is a custom Arm design that combines multiple Nvidia-built cores with fast, low-power memory and a high-bandwidth connection to Nvidia’s accelerators. The claim is that a chip optimized by the same manufacturer of the GPUs can transfer data between the two more smoothly than a standard processor.

      For Nvidia, this endorsement is as valuable as the order itself. The company has informed investors that it anticipates generating approximately $20 billion from Vera CPU sales this fiscal year, marking its initial foray into a general-purpose computing market it estimates to be around $200 billion in size.

      This initiative also has a defensive aspect. Some of Nvidia’s largest clients, including OpenAI, are developing their own AI chips, and a CPU line enables Nvidia to increase its sales per server rack, even in an environment where its accelerators face stiff competition. Additionally, it has been offering startups computational resources with deferred payments to secure demand.

      The broader ecosystem surrounding Vera Rubin is rapidly taking shape. On the memory front, Nvidia and SK Hynix have finalized a multi-year deal for HBM4 to supply the platform, among other supply agreements to support the roadmap.

      Perplexity’s decision also reflects the future direction of AI search. As competitors integrate generative responses into their offerings, a shift that is already transforming user search habits, the expense associated with scaling those responses has become a significant competitive issue in itself.

      Lowering inference costs is the key advantage. Faster CPUs allow a company to handle more queries within the same budget or to develop agentic features that combine numerous API calls without escalating expenses.

      For Perplexity, which has experienced rapid growth and raised substantial funds to support that expansion, controlling the cost per query is almost a matter of survival. Every efficiency gained from the hardware equates to a margin it doesn’t have to seek from investors or from increasing user charges.

      Neither party has disclosed when Vera will be available in large quantities, and Perplexity has not specified how much of its infrastructure will switch to the new CPU. The more challenging evaluation will occur later when Vera needs to gain the trust of enterprises that are already familiar with and reliant on Intel and AMD, while Nvidia’s CPUs remain relatively untested in that sector.