The AI build-out's next siting decision is above the atmosphere

The AI build-out is migrating from where the power and water are scarce to where the constraints disappear. Orbital compute is no longer a research curiosity; it is the next siting decision for the AI-infrastructure race, with chip vendors now being named at the constellation-planning stage rather than the single-satellite demo stage.

SpaceX's Aug 4 Nvidia partnership is the first named silicon step in a plan filed with the FCC on Jan 30 for an orbital data center constellation of up to one million satellites in low Earth orbit. The Starmind AI1 payload will run NVIDIA Rubin GPUs and Vera CPUs, with first launches targeted for late 2027. SpaceX frames orbital processing as cheaper and lighter-footprint than Earth-based facilities. Both claims are company projections.

When the chip vendor gets named at the FCC-filing phase, the conversation shifts from "can AI run in space" to "who is supplying the silicon for the orbital build." SpaceXAI's separate Aug 24 adoption of NVIDIA's Vera CPU for agentic AI workloads points in the same direction: orbital and terrestrial stacks are converging on the same supplier base.

The conditional stakes: if orbital AI compute works at the scale SpaceX is filing for, the next gigawatts of AI capacity get sited above the atmosphere, and the energy-footprint debate moves with it. The 1M figure is a filing maximum, not a confirmed build, and the cost and footprint numbers remain SpaceX's to prove.