SpaceX wants to build a data center in space. Not a metaphorical one. An actual constellation of up to 1 million satellites capable of running AI workloads from low Earth orbit, powered by the sun and cooled by the vacuum of space.

The project is called Starmind, a name Elon Musk confirmed on June 23, 2026. It represents a dramatic pivot from SpaceX’s existing Starlink network, which currently has over 10,000 satellites beaming internet connectivity to Earth. Starmind isn’t about connecting people to the web. It’s about processing artificial intelligence computations where no one has processed them before.

From internet satellites to orbital supercomputers

Each Starmind satellite would carry onboard processors designed for machine learning inference, the process by which trained AI models generate outputs like text, images, or decisions. Solar arrays will provide the energy, and the natural vacuum handles cooling, eliminating the need for the kind of grid-scale power contracts that have turned data center operators into some of the largest electricity consumers on the planet.

The satellites will communicate with each other using optical inter-satellite laser links, the same technology that already connects Starlink satellites but repurposed here for shuttling AI data between orbital nodes and back down to Earth.

SpaceX has filed an application with the Federal Communications Commission for what the filing describes as an “orbital data-center system” involving solar-powered satellites with optical interlinks. The first test satellites, internally designated AI1, are projected to begin launching in early 2027.

The xAI acquisition ties it together

SpaceX acquired xAI, Musk’s artificial intelligence company, back in February 2026. That deal brought xAI’s AI models and engineering talent under the same corporate umbrella as SpaceX’s satellite manufacturing and launch capabilities. Starmind represents the merger of those two entities’ core competencies.

The scale is worth pausing on. Up to 1 million satellites would make Starmind roughly 100 times larger than the current Starlink constellation. For context, every satellite ever launched by every country in human history totals somewhere in the low tens of thousands.

What this means for investors and the broader market

The regulatory pathway is its own challenge. The FCC application is just the starting line in the US. International spectrum coordination, orbital debris management, and the sheer logistics of manufacturing and launching that many satellites will require sustained regulatory engagement across multiple jurisdictions.

The biggest risk is execution. Manufacturing AI-capable satellites at this scale has never been attempted. Starlink satellites already have a finite operational lifespan and require regular replacement launches. Adding onboard AI processors increases complexity, cost per unit, and the potential for hardware obsolescence before satellites even complete their orbital lifetimes.

Investors watching this space should track the AI1 test satellite launches in early 2027 as the first real proof point.