Every time a large AI model trains, it guzzles power. A single hyperscale data center can consume as much electricity as a small city, generate mountains of waste heat, and require vast tracts of flat land and rivers for cooling. As the demand for computing power climbs faster than the grid can supply it, engineers have begun looking up.
An orbital data center is exactly what it sounds like: clusters of computers riding inside satellites circling Earth at altitudes between 500 and 2,000 kilometers. The pitch is seductive. In space, solar panels receive near-constant, atmosphere-free sunlight. Waste heat radiates directly into the cold vacuum of space, sidestepping the cooling bottlenecks that throttle terrestrial servers. And there is effectively unlimited room to expand.
At the center of the conversation is SpaceX, which in early 2026 filed with the U.S. Federal Communications Commission for authorization to launch up to one million compute satellites — a constellation designed to act as a distributed AI data center rather than a communications network.
The challenges are stubborn. Launching a tonne of hardware into orbit still costs hundreds of thousands of dollars, and a million-satellite system would demand an unprecedented manufacturing and launch cadence. Hardware in orbit cannot be swapped out easily: a failed processor chip today means a dead piece of infrastructure for the rest of its mission. Radiation from the Van Allen belts can damage circuitry, and maintaining a fleet of this scale across thousands of orbits raises complex questions about space traffic and debris.
Skeptics, including OpenAI founder Sam Altman, have argued that orbital data centers make sense someday — but that current launch costs and the difficulty of repairing components in space make them impractical for today's computing needs. Space advocates counter that, if cooling and energy costs on Earth continue to spike, orbit may simply become the path of least resistance.
For now, space-based computing remains a long-term bet rather than an imminent revolution. The grid is not yet so desperate that rocket-powered cooling makes financial sense. But as AI demand keeps climbing, the question is no longer whether orbit will one day hold some of our clouds — only when.