NVIDIA Vera conquers Perplexity: the new Arm processor targets artificial intelligence workloads

Perplexity has confirmed its intention to use NVIDIA's new Vera CPU, developed for the Vera Rubin platform and designed to handle increasingly complex workloads related to AI agents. The general-purpose processor is tasked with coordinating - or orchestrating - activities, and can become a critical element as much as the accelerator responsible for parallel computations.

Perplexity's decision represents one of the first significant signals of interest in Vera, a CPU with which NVIDIA aims to diversify its business beyond the sale of platforms and GPUs. The company led by Jensen Huang has already identified early users of the processor such as OpenAI and Anthropic, along with other ecosystem partners like Meta.

Vera is based on a customized Arm architecture from NVIDIA. The processor integrates 88 Olympus cores and 176 threads, an increase from the 72 cores of the previous generation Grace. These features are complemented by high-speed memory and wide-bandwidth connections to accelerators, aimed at reducing bottlenecks in communication between CPU and GPU. Further technical details can be read here.

According to Nate Kupp, Vice President for Infrastructure and Enterprise Computing at Perplexity, Vera has proven particularly suited to the company's workloads, with performance in AI agent-based programming tasks about 1.5 times higher compared to traditional CPUs previously used.

The company manages over 400 million queries per month, each requiring an inference pipeline made up of numerous computational steps. In this context, improving hardware efficiency becomes essential for containing operational costs and increasing service capacity. A processor that is faster in managing secondary processes and coordination tasks allows for a greater number of requests to be executed with the same infrastructure, or for more advanced AI features to be introduced without a proportional increase in costs.

NVIDIA's entry into the processor sector represents a direct challenge to Intel and AMD, companies that have dominated the CPU market for servers and workstations for decades. NVIDIA has communicated to investors its goal of generating about $20 billion in revenue from Vera CPUs by the end of the current fiscal year, within a general-purpose processor market that the company estimates to be around $200 billion.

This move also has strategic significance. Many large customers in artificial intelligence, including OpenAI, are developing proprietary chips to reduce dependence on external suppliers. Thus, a CPU line allows NVIDIA to maintain a central role in complete AI systems, selling more components within the same rack even in the presence of increased competition in accelerators.

Perplexity has not indicated how many Vera processors it intends to purchase or when it will begin large-scale distribution. However, the initial adoption represents a significant first step for NVIDIA, which aims to transform the CPU into a new pillar of its business.