SpaceX seeks FCC approval to launch a million data center satellites

The AI boom has been marked by a simple constraint that is only getting harder to outrun. You can design chips, you can expand factories, but electricity and cooling are quickly becoming the real friction in how fast new compute can come online.

Late last week, SpaceX put its boldest answer on paper. In its Jan. 30 filing with the Federal Communications Commission, the company asked for permission to launch and operate what it calls the "SpaceX Orbital Data Center System," a constellation intended to act as data centers for advanced AI models and the apps that run on them, with a grand ambition of up to one million satellites.

The filing reads like a bet that the lowest-cost compute for AI will come not from chips and data centers, but from orbit, where solar power is almost constant and heat is dumped directly into space. It is also, however, the reason the proposal is spreading so fast. It takes today's data center backlash over power consumption, water usage, and local opposition and replaces it with a space-scale alternative.

Reuters reported SpaceX has been in merger talks with xAI, a context that makes the orbital data center plan feel less isolated and more like part of a broader AI buildout.

What SpaceX is asking regulators to approve

SpaceX is applying for permission to deploy a system of up to one million satellites operating as orbital data centers. It would operate within narrow shells of up to 50 kilometers in width, between 500 to 2,000 kilometers in altitude.

It also includes requests to speed up the licensing process, including exemptions from certain processing rounds and flexibility on milestones and bonding requirements tied to how satellite systems are usually reviewed.

How the “orbital data center” would work

The proposal depends on networking. SpaceX explains that it would use high-bandwidth optical communications almost exclusively to route traffic through the network and into Starlink to reach authorized ground stations.

SpaceX argues that by operating in these specific shells, especially high-altitude sun orbits the satellites canstay in sunlight more than 99% of the time. Helping the batteries to provide steady power to “energy-intensive AI computing work.”

One engineering note worth including is cooling. Space is not a freezer. In a vacuum, you can't cool down by air; you have to use radiators to emit heat. And that creates a real design challenge, because you are talking about massive radiator surfaces adding weight and complexity.

Why “one million” is the number driving the buzz

The million-satellite number does two things. First, it signals ambition. Second, it gives SpaceX flexibility under a wide authorization. Reuters notes that SpaceX previously applied for 42,000 Starlink satellites but has deployed only a fraction of that, suggesting that “up to” figures are often just flexibility language.

However, in this filing, Starship is not a minor detail. SpaceX explicitly argues that fully reusable launch vehicles like Starship could make it possible to deliver “millions of tons of mass to orbit every year,” which is how on-orbit processing could reach unprecedented scale.

In this context, orbital computers are not just enabled by Starship; they are a payload concept that becomes increasingly compelling the more often you fly a megarocket.

SpaceX also attempts to quantify what that scale could mean. In one example, they describe launching one million tonnes per year of satellites generating 100 kilowatts of compute power per tonne, which could potentially add 100 gigawatts of AI compute capacity annually.

The risks and questions that come next

Even those sympathetic to SpaceX’s plans are likely to arrive at the same question first: orbital crowding. Collision risks become an even harder problem when constellation sizes increase exponentially.

SpaceX assures regulators that it would follow debris mitigation guidelines and dispose of satellites through atmospheric re-entry.

There is also a quieter business appeal beyond electricity politics. Even if federal licensing still applies, orbital compute avoids many of the fights that slow down terrestrial data centers, from local zoning disputes to water constraints to long grid-connection waits, because the hardware sits in orbit and doesn’t plug into any single local grid.

The friction doesn’t disappear entirely because ground stations still need to be built, but the main constraint shifts from city council meetings to spectrum allocation and launch economics.

SpaceX is telling Washington that the next phase of AI infrastructure is on a collision course with Earth’s energy limits and local resistance, and it wants the freedom to test a fundamentally different model.