SpaceXAI is expanding its use of NVIDIA hardware with plans to deploy Vera CPUs and Vera Rubin systems across its next generation agentic AI infrastructure.
The company says Vera CPUs will support workloads that depend heavily on orchestration, code execution, data processing and simulations, while Vera Rubin servers will be used to expand compute capacity for Grok.
SpaceXAI is also working with NVIDIA on an orbital version of the platform for its first generation Starmind AI satellite, extending the same architecture beyond terrestrial data centers.
SpaceXAI plans several NVIDIA deployments
| Platform | Planned use |
|---|---|
| NVIDIA Vera CPU | Agentic AI orchestration and CPU heavy processing |
| Vera Rubin servers | Grok AI infrastructure |
| Space 1 Vera Rubin | Orbital AI computing |
| Starmind satellite | First generation SpaceXAI AI satellite |
| Planned AI capacity | 10 MW by 2027 |
| Longer term direction | Gigawatt scale AI infrastructure |
The announcement follows an earlier commitment by SpaceXAI to use NVIDIA GPUs for its AI work.
The company has also outlined plans to bring around 10 MW of AI computing capacity online by 2027 using NVIDIA hardware.
Vera CPUs will handle agentic AI orchestration
SpaceXAI is adopting NVIDIA’s Vera CPU because agentic AI requires significant CPU resources in addition to GPU acceleration.
AI agents can repeatedly execute code, call external tools, process data and coordinate several operations between model requests.
Those tasks can create substantial CPU and memory bandwidth requirements even when the main AI model runs on GPUs.
SpaceXAI says Vera will allow more of this orchestration work to take place without using GPU resources for tasks better suited to the CPU.
That should allow the accelerators to remain focused on model inference and other highly parallel workloads.
Grok infrastructure will move toward Vera Rubin
Vera Rubin servers will also become part of the infrastructure supporting Grok.
SpaceXAI plans to expand toward much larger AI installations over time, potentially reaching gigawatt scale.
NVIDIA positions Vera Rubin as a full stack platform that combines compute, networking and software rather than treating the GPU as an isolated component.
That approach is particularly relevant for large AI clusters where processor performance alone is only one part of the system.
Memory bandwidth, networking, storage and workload scheduling can all become limiting factors as infrastructure scales.
Starmind will take Vera Rubin into orbit
The partnership also includes Space 1 Vera Rubin hardware designed for orbital computing.
SpaceXAI plans to use the platform in its first generation Starmind AI satellite.
The module combines CPU and GPU capabilities through a high bandwidth connection and is designed to operate using solar power.
According to the supplied information, the orbital module is intended for tasks such as geospatial intelligence, autonomous operations and real time on orbit analytics.
NVIDIA claims the Space 1 implementation can provide up to 25 times the AI compute capability of an H100 for orbital workloads.
The architecture is also intended to remain compatible with the broader Vera Rubin software ecosystem, which could make it easier to develop applications that move between Earth based and space based systems.
NVIDIA is extending Vera Rubin beyond data centers
The broader Vera Rubin platform is being positioned for more than conventional AI server deployments.
NVIDIA is working with several companies on orbital computing applications, with SpaceXAI becoming one of the latest organizations to adopt the architecture for a satellite project.
For SpaceXAI, the strategy connects three areas under the same hardware family.
Vera CPUs will handle CPU intensive agentic work, Vera Rubin servers will expand Grok infrastructure, and Space 1 Vera Rubin will support Starmind in orbit.
The main significance is not simply that SpaceXAI is buying more accelerators. The company is adopting NVIDIA’s broader CPU, GPU and systems architecture across both terrestrial AI infrastructure and orbital computing.
If those plans progress as described, Vera Rubin could become a common computing foundation across SpaceXAI’s large scale AI facilities, Grok workloads and its first Starmind satellite.
Discussion (0)
Be the first to comment.