Nvidia Ships First Vera CPUs to OpenAI, Anthropic, and SpaceXAI
Nvidia has begun shipping its first custom CPU to major AI organizations, marking its formal entry into the data center processor market as a full-stack AI infrastructure provider.
The first Vera CPUs have reached leading AI firms and cloud infrastructure partners in California. As part of the rollout, early recipients can begin evaluating the processor ahead of wider commercial deployment.
Built for Agentic AI Workloads
The Vera CPU represents Nvidia’s first fully custom data center processor, featuring 88 Olympus cores with Spatial Multithreading, enabling 176 total threads. Additionally, the chip delivers 1.2 TB/s of memory bandwidth through a 1024-bit LPDDR5X interface, offering significantly higher per-core bandwidth than competing designs.
According to developer documentation, the processor delivers up to 50% faster sandbox performance compared to rival platforms. As a result, it specifically targets agentic AI workloads that demand strong single-core performance, deterministic execution, and minimal latency.
Earlier this year, the broader Rubin platform was projected to deliver inference at one-tenth the token cost of previous-generation systems.
Full-Stack AI Infrastructure Expansion
The initial deliveries come ahead of larger production shipments expected in the second half of 2026. Consequently, cloud providers are preparing to deploy Vera CPU infrastructure at scale.
The chip is designed to work alongside Rubin GPUs within the NVL72 server architecture, which combines 72 GPUs through NVLink 6. Moreover, this move places Nvidia in direct competition with established data center CPU players by leveraging its custom Arm v9.2 architecture.
