SpaceX could need over 2 million Nvidia Rubin GPUs: a striking estimate

Analysts have started tying SpaceX to a striking long-range compute estimate: demand for more than 2 million Nvidia Rubin GPUs. Rubin is Nvidia’s next big accelerated-computing platform after Blackwell, and SpaceX hasn’t confirmed anything remotely close to that kind of volume. Still, analysts argue that this is the sort of capacity the company could end up needing if Starlink, Starship, satellite operations, robotics, simulation, autonomy software, manufacturing automation, and its other software-heavy systems keep expanding.

Getting past 2 million Nvidia Rubin GPUs would almost certainly take years. It would also mean several data-center campuses, not a single giant order. And the strain wouldn’t stop at chip supply. You’d also be looking at enormous demand for dense networking, memory bandwidth, custom racks, liquid cooling, real estate, and power on the scale of gigawatts.

For anyone who tracks AI infrastructure, that 2-million-plus estimate matters. It would put SpaceX in the same conversation as the biggest compute build-outs anyone has seriously floated. It would also send Nvidia a clear signal that demand is moving beyond the cloud providers, while tightening the market for top-tier chips and data-center capacity at the same time.

So treat the 2-million-plus GPU figure as a directional industry estimate, not an active purchase plan. Even so, it points to something real: SpaceX’s future may hinge as much on data-center scale as it does on rockets.