SpaceX enters deal with NVIDIA

---

Elon just confirmed the design is “significantly simpler, lower cost, denser and lighter than a traditional rack.”

That’s the practical part. 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.

NVIDIA’s post the same day calls Vera “the first CPU built for agents.” SpaceX is deploying it to accelerate the orchestration, code execution, and data processing that powers its next generation of agentic AI — keeping the GPUs fed and agents acting fast. From gigawatt AI factories to orbit. One NVIDIA architecture, everywhere.

Vera Rubin NVL72 is NVIDIA’s rack-scale AI system: 72 Rubin GPUs paired with 36 Vera CPUs. Vera is the CPU side, purpose-built for the agentic workloads (tool use, code, data processing, orchestration) that sit around the model inference.

Starmind is SpaceX’s planned constellation of solar-powered AI compute satellites. Results get beamed back to Earth via Starlink lasers. The official SpaceX Starmind page and NVIDIA’s August 24 press release both state that the first-generation Starmind AI1 satellite will be based on an optimized version of this Vera Rubin NVL72 system, adapted for orbital power, thermal, and integration constraints.

NVIDIA’s release also notes SpaceXAI is expanding the AI infrastructure behind Grok on the Vera Rubin platform as it scales toward gigawatts of capacity.

In orbit you pay for every kilogram and every watt. A traditional data-center rack carries overhead that doesn’t survive the launch mass and vacuum-cooling reality. Elon’s design claim is the engineering answer to that. Earlier primary statements also indicated the same simplified Starmind V1 design (minus solar and radiator) is intended for ground data centers because of the efficiency gains.

Owners already live with efficiency trade-offs every day — range, charging speed, hardware iterations. This is the same mindset applied at the AI factory scale.

Per the official SpaceX Starmind page: SpaceX plans to produce advanced AI chips at scale in its new fab facility, a joint project with Tesla called TERAFAB, using a modular architecture that enables seamless integration and continuous upgrades of next-generation processors.

Those chips are aimed at both the high-power space data-center use case and the edge/inference needs of Optimus and Cybercab. The same silicon effort supports the orbital compute and the vehicle/robot AI hardware Tesla owners care about.

Nothing in the primary statements says this NVL72 system will update the FSD version in your Model Y or Cybertruck next month. The stated targets are first orbital launch in Q4 2027 and significant scale in 2028. Those are the numbers from Elon and NVIDIA. Treat them as targets.

This is a concrete step in the shared Elon AI stack. Denser, simpler, lower-cost compute — on the ground or in orbit — is what eventually improves the models that train FSD, run Grok, and power Optimus. Tesla is in the joint chip fab that supports it.

Q4 2027 for the first space-optimized system. 2028 for scale. Same architecture from the AI factories all the way to orbit.

Which part of this AI compute push are you watching most — the orbital scale or the Terafab chips for Optimus and Cybercab?

Primary sources only: Elon X posts, NVIDIA official release and X, SpaceX Starmind page and X