AI firm Perplexity has chosen NVIDIA’s Vera CPU, highlighting its exceptional single-threaded speed that outperforms traditional processors by a significant margin in agentic AI coding tasks. The Vera chip’s design, tailored specifically for inference workloads, addresses the demands of AI models that rely heavily on efficient, low-latency CPU performance.
NVIDIA markets Vera as the world’s only large-scale “Max Single-Threaded” CPU, focusing on three crucial design pillars: strong per-core performance under load, sufficient memory bandwidth per core, and stable latency. These factors are critical for agentic AI workloads, which involve iterative decision-making loops requiring rapid, predictable processing times rather than simply increasing core counts.
Unlike conventional CPUs that depend on adding more cores to boost performance—often increasing resource contention—Vera optimizes each core individually, delivering higher instructions per cycle (IPC) with its custom Olympus cores. This approach allows AI agents to process tasks faster at the single-thread level. Vera also features high memory bandwidth paired with efficient power usage to sustain demanding inference operations.
Vera’s adoption is growing among major AI companies, including OpenAI, xAI, Oracle, and Anthropic, reflecting a broader industry shift toward CPUs tailored for AI inference rather than traditional GPU reliance. NVIDIA projects substantial revenue from Vera while positioning it as a leading CPU for AI in the current year. Perplexity’s endorsement underscores the chip’s practical advantage in real-world agentic AI scenarios, where speed measured in nanoseconds is paramount.