SpaceX Maps Out Starmind, Its Orbital AI Data Centers

HAWTHORNE, Calif. — SpaceX is turning one of Elon Musk's most ambitious ideas into hardware: data centers that live in space. The company's Starmind program pairs its launch and Starlink expertise with Nvidia's most advanced chips to process artificial intelligence workloads in orbit, and Musk has now put a timeline on it, with the first AI satellites expected to fly as soon as 2027 and reach meaningful scale the following year.

The plan centers on Starmind AI1, the first satellite in a planned megaconstellation. Each spacecraft will carry Nvidia's Rubin GPUs and Vera CPUs for what SpaceX calls datacenter-class compute in orbit. Musk told investors the company is committing to Nvidia's Vera Rubin architecture, calling it the best AI computer available and describing SpaceX as effectively exclusive to Nvidia for its first-generation orbital hardware.

Why Put a Data Center in Space

The logic is deceptively simple. In a sun-synchronous orbit, a satellite bathes in near-constant sunlight and can radiate heat directly into the vacuum of space, sidestepping the two constraints that increasingly bottleneck data centers on the ground: power and cooling. SpaceX says that approach drops cooling overhead by an order of magnitude while removing the zoning, land, and grid-connection hurdles that slow terrestrial builds.

Each Starmind satellite is a large machine, with a deployed height around 30 meters, a 75-meter wingspan, up to 250 kilowatts of peak compute payload, and roughly 210 kilowatts of solar output. Processed results would be beamed back to Earth over high-bandwidth optical laser links tied into the Starlink constellation, which already reaches customers across the globe. It is, in effect, a computer in the sky with a fiber-fast connection home.

Building at Scale

SpaceX has filed with the Federal Communications Commission for a constellation of up to one million Starmind satellites, and plans to mass-produce them inside a dedicated Gigasat Factory in Bastrop, Texas. Getting anywhere near that number will lean on the same manufacturing muscle SpaceX is pouring into ramping Starship production, since only a low-cost, high-cadence launch pipeline makes an orbital compute network economically viable.

The payoff Musk describes is enormous. He has set a target of bringing roughly 10 gigawatts of AI compute online by the end of 2027, spanning both ground and space, and has told staff AI could become the dominant source of SpaceX revenue within a few years. Early demand is already visible: SpaceX's AI division leases compute capacity to major customers, and the company reported AI revenue climbing sharply as those deals ramp.

As Teslarati reported, the Nvidia partnership was unveiled alongside SpaceX's first earnings report as a public company, sending its shares sharply higher on the day. Starmind remains an audacious bet, and the engineering challenges of operating server-grade hardware in orbit are real. But with a chip partner locked in, a factory taking shape in Texas, and a launch date on the calendar, SpaceX has moved orbital AI from concept toward a countdown.