Nvidia unveiled its Vera ARM-based CPU at the RTX Spark launch, claiming up to 1.8x performance gains over competing x86 processors. The chip packs 88 Olympus cores with Spatial Multithreading delivering 176 threads per socket, marking a significant shift toward ARM architecture in high-performance computing.

Vera supports up to 1.5TB of LPDDR5X memory with 1.2TB/s bandwidth, optimized for AI inference workloads. The processor handles agentic AI, reinforcement learning, and data analytics either standalone or paired with Rubin GPUs. Nvidia's NVLink-C2C interconnect enables 1.8TB/s communication between CPUs and GPUs in configurations like the NVL72 system.

Major AI companies have already committed to Vera deployment. Anthropic's Claude, OpenAI's ChatGPT, and SpaceXAI's Grok will leverage these processors alongside hyperscalers including ByteDance, CoreWeave, and Oracle Cloud Infrastructure. This represents Nvidia expanding beyond GPU dominance into full server stack solutions.

System integrators Dell, HP, Lenovo, Supermicro, and others will build Vera-based servers, while the NYSE explores the technology for processing its 1.1 trillion daily messages. Nvidia's ARM push directly challenges Intel and AMD's server processor stronghold.