Nvidia details Vera CPU for AI data centers

Nvidia has disclosed new technical details for Vera, its next-generation server CPU, as it moves further beyond the GPUs that have powered its rise in AI infrastructure. The company said the first Vera processors were delivered to customers in June, including OpenAI, Anthropic and SpaceX.

Vera is Nvidia’s first server CPU developed from the ground up rather than based on Arm’s off-the-shelf core designs. The chip targets bottlenecks in agentic AI systems, with Nvidia claiming up to 50% better performance for AI agent workloads than traditional x86 processors from Intel and AMD.

Nvidia is emphasizing single-core performance, memory bandwidth and low latency over simply increasing core counts. Vera will be sold as a standalone chip, in dual CPU server configurations, in liquid-cooled racks containing 256 Vera processors, and alongside upcoming Rubin GPUs in Vera Rubin AI systems.

The move intensifies competition with AMD and Intel, whose shares have risen 128% and 149%, respectively, in 2026, compared with Nvidia’s 8% gain. Analysts cited in the report said Nvidia may be targeting specialized AI workloads rather than replacing conventional enterprise CPUs, while adoption among major cloud providers remains early.