SpaceXAI adopts NVIDIA Vera CPUs to accelerate agentic AI

SpaceXAI is deploying NVIDIA Vera CPUs to power its next generation of agentic AI workloads, specifically to expand the infrastructure supporting Grok. The Vera CPU is designed to handle the intensive orchestration, code execution, and data processing required by AI agents, ensuring that GPUs remain fully utilized for model inference.

SpaceXAI intends to extend this computing architecture into orbit with its first-generation Starmind AI satellite, which will utilize an optimized NVIDIA Vera Rubin NVL72 system.

Technically, the NVIDIA Vera processor is built to address the specific needs of agentic workflows, which involve reasoning, tool invocation, and memory access. Unlike traditional cloud CPUs that focus on aggregate core counts, Vera utilizes a monolithic compute die with 88 custom Olympus cores. This design aims to reduce latency and provide the predictable single-thread performance necessary for tasks such as Python execution, database management, and complex simulations.