The Federal Communications Commission (FCC) accepted a filing from Elon Musk-led SpaceX to put AI data centers in orbit, which is seeking approval to launch a constellation of up to one million solar-powered satellites.

The filing, submitted on January 30, outlines what SpaceX describes as an “Orbital Data Center system” designed to operate in low-Earth orbit (LEO).

On February 2, SpaceX announced it had acquired Musk’s AI startup xAI. In the announcement, Musk argued that AI advances are increasingly constrained by the demands of large terrestrial data centers. The deal creates a business valued at $1.25 trillion, making it the world’s most valuable private company.

The FCC filing describes satellites deployed in orbital shells between 310 and 1,200 miles in altitude. The proposed system would connect to the company’s Starlink constellation via high-bandwidth optical links, with Starlink relaying traffic to ground stations through its laser mesh network. Operating in low-Earth orbit would reduce latency compared with geostationary satellites, though it would remain higher than terrestrial fiber connections.

In the announcement, Musk said that advances in AI are increasingly constrained not only by the immense power and cooling demands of terrestrial data centers, but also by the physical space they require. Musk argued that global electricity supply and land availability cannot scale fast enough to meet future AI demand without straining communities and the environment.

The “only logical solution,” he said, is to move compute infrastructure into space, where satellites can harness near-constant solar energy and avoid terrestrial resource constraints.

Musk said that launching up to one million tons per year of AI-focused satellites using SpaceX’s Starship launch system could add hundreds of gigawatts of compute capacity annually, eventually reaching terawatt scale and reducing the cost of AI processing.

However, industry experts have questioned the technical and economic feasibility of the proposal.

Jonathan McDowell, an astronomer and astrophysicist at the Harvard-Smithsonian Center for Astrophysics, said the launch cadence is extremely challenging, but depends on how long each satellite lasts before needing to be replaced, which SpaceX has not specified.

“If they last only five years like Starlink, I think it's not doable, but they are probably planning closer to 10 years," McDowell explained. "That's still about five Starship launches a day every day for this project alone. I'm not saying they can't do it, but it seems like a stretch.”

McDowell added that it would be possible to have a working system meeting only part of the world's data needs with as few as 10,000 satellites, but questioned whether that would be profitable.

Martin Barstow, professor of astrophysics and space science at the University of Leicester's School of Physics & Astronomy, said he was skeptical about the ability of the proposed system to generate sufficient power compared with the solar arrays on the International Space Station.

“ISS generates 240kW, but single data centers on the ground require megawatts of power,” Barstow said. “Total demand is going to be several hundred gigawatts worldwide.”

Barstow added that AI-based processing on satellites could improve autonomy and enable more intelligent data processing before sending it to the ground. But putting general processing into space would require a huge amount of infrastructure and cost.

“Similar to the arguments around space solar power, it is easier to do it on the ground," Barstow said. “One mantra of my career, when carrying out a cost-benefit analysis of putting experiments in space, is, if it can be done on the ground, you don’t do it in space.”