SpaceX Aims to Launch Up to 1 Million AI Data Center Satellites

SpaceX has asked the Federal Communications Commission (FCC) for permission to launch up to 1 million new satellites that will act as data centers to run AI. Let that number sink in for a moment. That’s 100 times more satellites than the company’s current Starlink constellation.

Elon Musk, founder of SpaceX, claims that data centers in space will be better than ones on the ground, where they take up land, devour electricity, and guzzle water for cooling. Space has lots of room, abundant solar power, and it’s cold. Musk wrote in a statement, “In the long term, space-based AI is obviously the only way to scale.”

That conclusion isn’t at all obvious to other experts, including Jonathan McDowell, an astronomer and space sustainability expert who recently retired from the Harvard-Smithsonian Center for Astrophysics. He’s not convinced that SpaceX’s plan is “a good idea or even a viable idea.”

“The industrial scale of this is staggering,” he says, adding that this may be a publicity stunt. “I don’t think it’s going to happen in the next few years.” SpaceX did not respond to requests for comment.

Blocking out the stars

To imagine a single orbital data center, start with a low-Earth orbit communications satellite. Move it up to a higher orbit of 1,000 kilometers (600 miles) to put it in a Sun-synchronous orbit that allows continuous access to solar energy. Increase the size of its solar panels and add computers and heavy radiators for cooling. It would be a “ginormous” structure, says McDowell, “probably 100 meters (330 feet) tip to tip.”

(SpaceX's application calls for altitudes ranging from 500 to 2,000 km, as well as orbits at both 30° inclination and Sun-synchronous. Recent details unveiled by Musk suggest the satellites' actual size might be even larger, 170 meters across.)

That high orbit is good for powering computers, but it’s bad if you care about dark skies and stargazing. Astronomer John Barentine (Dark Sky Consulting) says that these orbital data centers “would be visible probably all night long over much of the Earth over much of the year.”

Right now, there’s not enough technical detail available to calculate exactly how bright they would be, Barentine adds. But simulations based on the data have alarmed him. “It is so potentially transformational to the night sky that I think it would really endanger the hobby [of stargazing]," he says. The science of astronomy would become far more difficult, too, with increased visual and radio interference from so many satellites.

Blasting off and breaking down

Space does make it somewhat easier to power and cool a data center, McDowell acknowledges. But building and managing something meant to zip around the planet takes energy in other forms.

Currently, a SpaceX rocket blasts off every few days. Musk’s plan calls for hourly launches each carrying 200 tons. Those would be with the powerful Starship rocket, which belches out greenhouse gases and pollution. With our current technology, these emissions would outweigh any carbon savings from space-based solar power and cooling, a 2025 study found.

Once the satellites are up there, they need to avoid each other and all the space junk left over from the Cold War. New satellites are designed to dodge obstacles. But just “a few bad collisions” could trigger a runaway cascade of crashes, McDowell says, leading to a belt of debris that renders much of space completely unusable.

Finally, all of these satellites eventually need to be scrapped. The most common way to dispose of a satellite is to burn it up in Earth’s atmosphere, where it would leave behind metals. Scientists are still studying the environmental impact of our current rate of launches and re-entries, says McDowell. Increasing both by a factor of 100 would certainly raise the risks of damaging the ozone layer and polluting the upper atmosphere.

A way forward

Space is a resource that the whole world shares. China and the European Union have their own plans for orbital data centers, as do a number of other companies. So even if SpaceX never actually reaches its audacious million-satellite goal, some number of orbital data centers are likely to become a reality.

“Innovation in space is the next frontier,” says public policy expert Karthik Kannan (University of Arizona). He’s collaborating with engineers in a project called AZSCI (pronounced AZ-Sky), investigating how to build orbital data centers safely and sustainably.

“I enjoy the sky,” says Kannan. “It’s mesmerizing to look at the stars.” But he thinks that orbital data centers will be important for all kinds of future uses of space, including space travel. So it’s important to set guidelines and discuss policy surrounding our use of space now, he says.

In order for SpaceX to move forward with its plans, the FCC must grant permission. Through March 6th, anyone can submit a public comment regarding the proposal. The comment period for the SpaceX filing closes on March 6th, while the Reflect Orbital comment period closes on March 9th.

Editorial note (March 27, 2026): Information was added regarding the FCC application as well as the satellites' possible size.