Perplexity chooses Nvidia’s Vera CPU for AI agent workloads

Perplexity plans to use Nvidia’s Vera CPU in its production systems, giving Nvidia another prominent customer as it expands beyond AI graphics processors into a market long led by Intel and AMD. Nvidia is positioning Vera as a data centre CPU built for AI agents, which run continuous task loops rather than conventional computing workloads.

Perplexity’s Vice President for Computer Enterprise and Infrastructure, Nate Kupp, said Vera was significantly faster for the company’s AI agent workloads than traditional processors. He said the chip completed AI coding workloads around 1.5 times faster than conventional CPUs, though Perplexity did not disclose how many processors it plans to buy.

Nvidia describes Vera as a CPU for AI factories, handling code execution, data processing, tool calling, memory management and result verification around AI models. The company says its custom Olympus CPU core delivers 50% higher instructions per cycle than the prior Grace architecture, with up to 1.2TB/s memory bandwidth and 3.4TB/s core-to-core bandwidth across 88 cores.

The shift adds pressure on Intel and AMD as customers increasingly buy integrated AI systems rather than standalone server CPUs. Nvidia has also disclosed Vera commitments from OpenAI, Anthropic and Oracle, while Forbes reported that OpenAI, Anthropic and SpaceX were among the earliest recipients of Vera systems.