“By directly harnessing near-constant solar power with little operating or maintenance costs, these satellites will achieve transformative cost and energy efficiency while significantly reducing the environmental impact associated with terrestrial data centers,” SpaceX wrote in an application seeking U.S. regulatory clearance for a concept that pushes satellite networks beyond communications and into large-scale computing.
The submission requests Federal Communications Commission to grant approval of a constellation that is projected to have 1 million satellites that are to orbit the earth and utilize the sunlight to energize onboard processing. SpaceX presented this concept as a solution to a fundamental limitation of contemporary artificial intelligence, which is that the data centers that constitute its material infrastructure consume vast amounts of electricity and demand large cooling systems.
The mentioned number is impressive even according to the standards of ambitious satellite licensing. Space operators tend to seek approvals in excess of original deployment plans to maintain design flexibility; SpaceX earlier sought clearance of 42,000 Starlink satellites before increasing launches. The broadband network of the company has expanded to approximately 9,500 satellites, and an independent tracking indicated that the number of satellites in orbit is 9,357 Starlink satellites in orbit at the end of 2025.
The chargeability of the feasibility of the “compute in orbit” is also connected to one industrial lever one being the launch capacity. SpaceX claimed that the mass put into orbit by“fully reusable launch vehicles like Starship” could be sufficient to enable on-orbit processing at a scale significant to the terrestrial scale. Since 2023, Starship has made multiple test flights and is the core of SpaceX strategies regarding the implementation of heavier and more capable spacecrafts.
However, engineering questions become hard when post-launch. The solar power in space is readily available, but computing produces heat to dissipate, and radiating it into space (without air or water cools) is the only choice. Infrastructure TAI on the land has been roundly criticized by its resources footprint; an academic analysis estimated that by 2030 AI expansion would emit 244444 million tonne of CO2 per year and would use 7311125 million cubic meters of water each year, both figures are shown in a Nature Sustainability study that outlines state by state effects and mitigation strategies. The relocation of a portion of workloads out of planet deals with grid and water limitations in a single shot, but at the cost of thermal design, radiation tolerance and long life operations in an uncompromising environment.
Sketches of architectures targeting those constraints have been started by research groups. In another tethered concept proposal at the 2026 AIAA SciTech Forum, engineers at the University of Pennsylvania proposed a design that employs long, flexible cables to passively orient simple modular computing nodes and giant, “leaf-like” solar arrays. According to Igor Bargatin, the first design, “This is the first design that prioritizes passive orientation at this scale,” a system is intended to scale to thousands of nodes and provide up to 20 megawatts of computing power per structure with data being relayed via optical links.
The other fact that any large orbital computing platform needs to endure is the same one highlighted in that academic paper: one has to be able to endure constant hits of micrometeoroids and debris. “It’s not a matter of preventing impacts,” said Jordan Raney. “The real question is how the system responds when they happen.” Simulations by Penn indicated that disturbances could decrease within a long tethered structure and this restricted off-pointing to a few degrees in his simulations.
The filing of SpaceX comes to an already strained ecosystem due to scale. The significant constellations have been cautioned against by astronomers and experts in space safety as posing greater risks of collisions and making observations on the ground more difficult. Any movement to the idea of “data centers in orbit” would introduce mass, power systems, and radiators to an area where the satellite lifetime can be short and the plans to dispose of them are important, especially with the reentry of satellites entering the upper atmosphere.
Processing of space-based AI, that is, is no longer primarily a matter of whether or not a computer can be attached to a satellite bus. The new argument is over what workloads should be in the orbit, what architecture is capable of scaling without having active control, and how economics of launch, debris risk, and thermal physics are all limiting what can be realistically provided in terms of power by the “near-constant solar power.”