The Grok AI model is scaling toward gigawatt compute capacity using a new processor architecture designed to handle complex tool orchestration and code execution. SpaceXAI is deploying Nvidia Vera CPUs for these agentic AI workloads and utilizing the Vera Rubin NVL72 rack-scale system for its first-generation Starmind AI satellite. This partnership establishes a uniform computing foundation from terrestrial data centers to orbital infrastructure, standardizing the software ecosystem across both environments.
The Vera CPU features 88 Olympus cores and up to 1.2 TB/s of memory bandwidth, which Nvidia says provides 1.8x faster task completion than x86 CPUs for certain agentic tasks. The processor is intended to offload CPU-heavy inference work to ensure that GPUs remain focused on model workloads. To enable orbital deployment, the companies are adapting the system to withstand the specific thermal, power, and reliability constraints of space.
Key sources
- SOURCE@nvidia“keeping GPUs fed and agents acting fast”x.com
- SUPPORT@sawyermerritt“power, thermal management, bandwidth, reliability and physical integration impose dramatically different constraints from conventional data centers”x.com
- SUPPORT@wallstengine“scales toward gigawatts of compute”x.com
- SUPPORT@firstsquawk“SPACEXAI STARMIND SATELLITE TO USE VERA RUBIN NVL72”x.com
- SUPPORT@wallstengine“The chip features 88 NVIDIA-designed Olympus cores and up to 1.2 TB/s of memory bandwidth”x.com
- SOURCE@nvidianewsroom“The first CPU built for agents is now at work at massive scale.”x.com
- SOURCE@elonmusk“significant scale in 2028”x.com
- SUPPORT@business“initially launch in the fourth quarter of next year”x.com