SpaceX Unveils Starmind, a Network of Orbital AI Data Centers That Could Eventually Grow to 1 Million Satellites

SpaceX has revealed new details about Starmind, an orbital computing system for artificial intelligence workloads. On the project’s official page, the company describes it as a network of satellites with onboard AI modules that will draw near-continuous solar power in orbit, link to one another via laser connections and send computing results back to Earth through Starlink infrastructure.

The idea behind Starmind is to move part of AI computing into space and thereby sidestep the constraints facing ground-based data centers: power shortages, expensive land, complex cooling and long construction timelines for new sites. SpaceX is betting that in a sun-synchronous orbit, satellites will be able to draw steady power from the Sun and shed heat more efficiently in a vacuum.

The first generation of spacecraft is called AI1. According to SpaceX, a single such satellite is designed to carry a computing load of up to 250 kW at peak and about 175 kW on average. Once deployed, it will stand about 30 meters tall, with a solar array span of roughly 75 meters. Computing will happen onboard, and the results will travel over high-speed laser links to the Starlink network and on to users on Earth. That detail matters: Starmind is not meant to replace Starlink but to serve as a separate computing layer that uses Starlink primarily as a transport network.

SpaceX just shared a new visualization of what its proposed 1 MILLION-satellite Starlink constellation could look like The architecture is completely different from a traditional data center SpaceX VP Michael Nicolls explained that Starmind would organize satellites into tightly coordinated formations, combining their compute into larger AI workloads Each cluster could have roughly 10 satellites, connected to one another with around 10 terabits of bandwidth, while still linking into the wider constellation

— X Freeze (@XFreeze) October 6, 2026

Thousands of spacecraft first, up to a million in theory

The most striking figure comes not only from SpaceX’s visualizations but also from an official regulatory filing. A public FCC notice dated February 4, 2026 states that the agency has accepted for review a SpaceX application for an Orbital Data Center system of up to 1 million satellites. This concerns a new non-geostationary constellation that the company filed on January 30, 2026.

There is an important caveat, though: 1 million spacecraft is the upper limit of the requested configuration, not an approved deployment plan. The FCC has only accepted the application for review and sought comments on it. SpaceX’s near-term plans are far more modest: on its page, the company writes about manufacturing and launching the first thousands of AI satellites starting as early as late 2027.

To that end, SpaceX is building the Gigasat Factory in Bastrop, Texas — a dedicated facility for mass-producing such spacecraft. The company also stresses that Starmind’s architecture will not be tied to a single accelerator supplier, meaning the computing modules in the system can change as chips evolve.

The main question isn’t the idea — it’s the scale

The concept looks ambitious even by SpaceX’s standards. If the project does reach deployment, it won’t be just another communications satellite constellation but an attempt to turn low Earth orbit into a distributed computing layer for AI.

Yet scale is precisely what remains the biggest challenge here. Even SpaceX itself states outright that a critical condition for Starmind is cheap, high-volume delivery of cargo to orbit, which means the project’s success is directly tied to Starship. On top of that, expanding such a system will inevitably run into questions of orbital safety, congestion in low orbits, coordination with regulators and the long-term resilience of the entire near-Earth infrastructure.

That is why the news about “a million satellites for AI” is best read not as a nearly finished plan but as a long-term bid on a future direction for SpaceX. For now, the company has shown the architecture, the basic specs of the first generation and its intention to scale the system. But the path from the first thousands of spacecraft to a full-fledged orbital computing network will be far harder than the concept presentation itself.