Perplexity picks Nvidia's new chip for AI agent tasks

Perplexity AI is adopting Nvidia's new Vera CPU to run its AI agent workloads, becoming the latest company to back Nvidia's push into a server chip market long controlled by Intel and AMD.

Nate Kupp, Perplexity's vice president for enterprise infrastructure, described it as "a perfect match" for many of the company's core computing needs.

According to Kupp, Vera was found to be 1.5 times faster at completing AI agent code programming than normal servers during tests conducted by Perplexity.

However, Nvidia's marketing figures are even more impressive; they claim that their processor is 1.8 times faster than normal x86 processors in agentic AI, reinforcement learning, and data processing tasks. Perplexity has refused to comment on the number of units it intends to purchase.

Nvidia designed Vera specifically for the demands of AI agents, software that runs continuously through chained tasks, tool calls and coding loops without the pauses a human user naturally takes between commands.

It features 88 Olympus cores with up to 1.2 TB/s of memory bandwidth from LPDDR5X memory, and Nvidia is already manufacturing it, with systems shipping out to cloud providers and system builders starting this fall.

Perplexity will be a customer for Nvidia, which is trying to dominate a segment where Perplexity is a newcomer, although the ambitions of Nvidia seem greater than Perplexity's.

With the combination of Vera CPUs and their GPUs along with other networks and software stacks, Nvidia plans to take charge of the entire AI computing stack.

OpenAI, Anthropic, and Oracle are also joining forces to adopt Vera CPUs, with OpenAI, Anthropic, and SpaceX said to be some of the first companies to receive them.

Nvidia expects that the Vera processor will help the company bring in $20 billion in sales during its fiscal year-end. The company is looking to diversify its portfolio amid rising leverage from major AI labs.

OpenAI and DeepSeek are both developing custom processors, a trend that could eventually chip away at demand for Nvidia's most specialised hardware even as it expands into new categories like CPUs to offset that risk.