SpaceXAI, the combined AI arm of xAI and SpaceX, will deploy standalone NVIDIA Vera CPUs to power the next generation of agentic AI workloads behind Grok and put a space-optimised Vera Rubin NVL72 rack in orbit aboard its first Starmind satellite in Q4 2027, the companies said on August 24.
Why the CPU matters for agentic Grok
Agentic AI increasingly leans on CPUs to orchestrate tool calls, execute generated code, run simulations and shuttle data between model calls. NVIDIA’s new Vera CPUs are being adopted as a standalone accelerator alongside Rubin GPUs specifically to feed those inter-model orchestration steps at latency levels Grok's agent workloads need, and to keep Rubin GPUs fully utilised as SpaceXAI scales toward gigawatts of on-the-ground capacity.
A Vera Rubin NVL72 rack in space
NVIDIA and SpaceX have jointly designed a space-optimised variant of the Vera Rubin NVL72 rack for launch aboard the first Starmind satellite in the fourth quarter of 2027, with a broader ramp planned for 2028. The plan targets a distributed inference layer that pushes agentic Grok compute closer to end-users and offloads training and simulation workloads to on-orbit GPU capacity, adding an unusual space edge tier to NVIDIA’s AI-factory playbook.
Fits NVIDIA’s week of platform deals
The tie-up lands in a week of NVIDIA infrastructure headlines. On August 26 the company disclosed a Q2 FY27 with data-centre revenue up 117% year-on-year, and it followed with a 2-million-GPU AWS expansion and the Jetson Orin Nano 2 launch. Together they extend NVIDIA’s Vera Rubin platform from hyperscale data centres into edge robots — and now into low-Earth orbit.
Reporting based on coverage from NVIDIA, Tom’s Hardware, StorageReview and Wccftech.