Perplexity to deploy Nvidia’s Vera CPU for AI agents

AI search startup Perplexity confirmed on Tuesday that it plans to use Nvidia's ↗ 0.71% new Vera central processing unit, lending an early customer endorsement to the chipmaker's push into the data center CPU market long held by Intel ↘ 9.66% and AMD Advanced Micro Devices, Inc. ↘ 6.51%.

A CPU Built for Agents

Nvidia said it expects the Vera line to generate $20 billion in sales, positioning the chip as a more general-purpose complement to its dominant GPU business. The 88-core Arm-based processor is designed specifically for the repetitive "agent loop" that defines modern AI systems — the tool calls, code execution, data processing, and verification steps that happen between model inferences.blogs.nvidia+1

Perplexity tested Vera on a real coding workflow — cloning a repository and running its test suite in sandboxes — and found the chip completed the job approximately 1.5 times faster than x86 processors, while starting concurrent sandboxes up to 1.9 times faster. The company said it is now looking to deploy Vera in its upcoming production system.reuters+1

Expanding Beyond GPUs

The Vera CPU, which entered full production earlier this year, represents Nvidia's most direct challenge yet to the x86 server processors that have powered data centers for decades. Nvidia claims the chip delivers 1.8 times the sustained per-core performance of x86 chips under loaded agentic workloads, enabled by its custom Olympus core and up to 1.2 terabytes per second of memory bandwidth.nvidianews.nvidia+2

Perplexity joins a roster of early adopters. Nvidia began shipping Vera systems in May to companies including Anthropic, OpenAI, and Oracle Cloud Infrastructure. Reuters reported that the adoption marks a broadening of Nvidia's product line as AI companies like OpenAI and DeepSeek develop increasingly complex agent systems that strain traditional computing hardware.finance.yahoo+3

Performance Over Core Count

Nvidia's pitch centers on a design philosophy that prioritizes single-threaded speed over the high core counts favored by conventional server chips. The company argues that because each step in an agent's reasoning chain depends on the previous result, faster individual cores — rather than more cores — determine how quickly AI workloads complete. Nvidia has already outlined its next-generation Rosa CPU with the Rigel core as its successor to Vera, signaling a long-term commitment to the CPU roadmap.blogs.nvidia