NVIDIA and SpaceX Are Putting Vera Rubin GPUs Into Orbit

SpaceX and NVIDIA announced a deeper strategic partnership, with SpaceX planning to deploy NVIDIA's Vera Rubin architecture and NVL72 systems in its upcoming Starmind AI satellites — a concrete step toward running frontier-grade AI compute in orbit rather than only in terrestrial data centres.

Why Put a Data Center in Space

The core constraint on terrestrial AI buildout right now is not chips — it's grid power and cooling capacity, both of which are increasingly hard to secure fast enough near major population centres. Orbital compute sidesteps grid interconnection entirely by drawing on continuous solar power and radiative cooling, at the cost of far higher deployment complexity and no easy hardware servicing.

This Is Still Early, Not Imminent Mainstream Infrastructure

Running NVL72-class systems in orbit is a genuine engineering announcement, not yet a production data-centre replacement — launch costs, radiation hardening, and thermal management at that power density are unsolved problems at scale. Treat this as NVIDIA and SpaceX hedging against the terrestrial power ceiling years in advance, not a near-term alternative to conventional cloud capacity.

Why It's Still Worth Watching

The fact that NVIDIA is willing to put its most advanced architecture into an orbital pilot at all is itself informative — it's a signal of how seriously the industry now takes the power-constraint problem covered in our recent piece on AI's data-center power crisis, and a reminder that the next several years of AI infrastructure competition will be fought as much on power and cooling engineering as on chip design.