Particle.news
Download on the App Store

Nvidia Starts Shipping Vera CPU to Anthropic, OpenAI, SpaceXAI and Oracle

The custom chip targets CPU-heavy orchestration for agentic AI workflows.

Overview

  • Nvidia moved Vera into full production and hand-delivered the first systems to Anthropic, OpenAI, SpaceXAI and Oracle, with Oracle’s unit shown at its Santa Clara center on Monday.
  • SpaceXAI is evaluating Vera for reinforcement-learning and agent-based simulation pipelines after Elon Musk reviewed the system on site.
  • Oracle Cloud Infrastructure plans to deploy hundreds of thousands of Vera CPUs starting in 2026, making it the first cloud provider to commit at hyperscale.
  • Vera is Nvidia’s first custom data-center CPU, built for tasks like tool-calling, retrieval, sandboxing and long-context state, and it features 88 Olympus cores with 1.2 TB/s memory bandwidth and roughly 50% faster per-core performance under load than conventional designs.
  • Beyond standalone servers, Vera will host Rubin GPUs in Vera Rubin NVL72 racks using NVLink-C2C and a unified memory layout that Nvidia says delivers about 2x the energy efficiency of traditional setups.