SpaceX and Nvidia have officially partnered to deploy a space-optimized Vera Rubin NVL72 system into orbit, with the first launch targeted for Q4 2027 and significant scaling planned for 2028.
Announced on Monday, this initiative marks the first attempt to place a full rack-scale AI training and inference node in space, extending Nvidia's terrestrial "AI factory" architecture to SpaceX's Starmind satellite constellation. CEO Elon Musk confirmed that the company will exclusively use Nvidia hardware going forward, citing the Vera Rubin architecture as the "best AI computer" available for both ground and orbital operations.
The orbital system leverages the Vera Rubin NVL72 platform, which integrates 72 Rubin GPUs and 36 Vera CPUs connected via NVLink 6 into a shared-memory fabric. The Vera CPU, described as the first processor built specifically for AI agents, features 88 custom Olympus cores and 1.2 TB/s of memory bandwidth, claimed to complete agent tasks 1.8 times faster than comparable x86 processors. This infrastructure is intended to power the next generation of Grok's agentic AI applications, handling complex orchestration, code execution, and data processing directly in orbit to keep GPUs fully utilized.
The system aims to deliver up to 25 times more AI compute per GPU than previous space-based solutions like the H100, enabling real-time processing for geospatial intelligence and autonomous operations without relying on ground stations.
The core of this project is the Starmind AI satellite, which will house the adapted Vera Rubin NVL72 rack. Unlike traditional racks, this space-optimized version is designed to be simpler, lower cost, denser, and lighter, addressing unique orbital constraints such as power draw, thermal dissipation, radiation hardening, and physical integration.
This partnership extends xAI's infrastructure strategy, powering the Grok AI model both on Earth and in space. "SpaceX, in partnership with Nvidia, has designed a space-optimized Vera Rubin NVL72 system for launch to orbit in Q4 next year, with significant scale in 2028," Musk said. He has also said in a past X post the design is simpler, lower cost, and denser than traditional racks, making it suitable for both terrestrial gigawatt-scale data centers and orbital deployment.
This move aims to bypass terrestrial constraints like grid interconnect timelines and cooling limits while creating a unified computing ecosystem across Earth and orbit.
"@SpaceX is deploying NVIDIA Vera to accelerate the orchestration, code execution, and data processing that powers its next generation of agentic AI — keeping GPUs fed and agents acting fast. From gigawatt AI factories to orbit. One NVIDIA architecture, everywhere" Nvidia wrote on X.