Elon Musk Says SpaceX Will Use NVIDIA GPUs Exclusively and Plans 10 GW of AI Capacity

In a post on X, Elon Musk said SpaceX is standardizing its AI stack around NVIDIA GPUs, which he described as the best available option. He also said the company will use NVIDIA's Vera Rubin platform for AI workloads in its data centers, with deployments spanning the globe and space.

SpaceX's rollout starts with 2 GW of compute capacity planned by the end of 2026. By the end of 2027, Musk expects that figure to reach 10 GW, turning the project into a much larger buildout than the first phase alone.

For terrestrial data centers, SpaceX will use NVIDIA Vera Rubin NVL72 racks, codenamed Kyber. For orbital systems, it will rely on NVIDIA Space-1 Vera Rubin modules certified for space, using the same platform name across both environments but tailoring the hardware to where it will run.

Musk said space could become the most economically compelling place for AI in about 30 to 36 months. As he framed it, the advantage is practical: space avoids the usual limits of power, land, and cooling that constrain large data centers on Earth.

The setup raises a simple question: if SpaceX can really scale from Earth to orbit, what would stand in the way of others doing the same?