Elon Musk's SpaceX has quietly filed with the US Federal Communications Commission for an audacious expansion: an "Orbital Data Center system" composed of up to one million satellites distributed between roughly 500 km and 2,000 km altitude. The company describes the constellation as a new class of on‑orbit compute designed to run advanced artificial‑intelligence models and deliver large‑scale AI inference to billions of users by routing compute via optical links into its existing Starlink network.
SpaceX says the satellites would operate in narrow orbital shells no wider than 50 kilometres, use abundant solar energy in space, and be linked by laser communications to move compute in near real time. The filing frames the architecture as an answer to the power, cooling and land constraints that now slow terrestrial data‑centre growth, and envisions multiple satellite hardware variants tailored to different shells.
Delivering a million satellites depends on two ingredients SpaceX already markets as core strengths: the Starship heavy launcher and access to capital. Starship is presented as the only practical fulfilment route because of its high payload and low marginal launch cost, and the company is reported to be preparing a public listing in part to finance what would be an unprecedented buildout.
The regulatory hurdles are immediate and substantial. The FCC recently approved 15,000 second‑generation Starlink satellites, but that clearance came with explicit deployment deadlines and scrutiny over orbital debris. A proposal two orders of magnitude larger will attract intense technical, environmental and diplomatic review, with space‑traffic management and collision risk likely the prime sticking points.
Industry players and cloud providers will also watch closely. If anything like this scheme were allowed to proceed it would reshape the economics of large‑scale inference and edge compute, forcing terrestrial cloud operators to respond while creating new manufacturing, launch and spectrum markets. Competitors such as OneWeb and Amazon's Project Kuiper, as well as national space agencies, may contest the plan on commercial and national‑security grounds.
Beyond business and engineering lies geopolitics and governance. A privately owned orbital compute layer presents dual‑use concerns — the same capacity that accelerates global AI services could be repurposed for military missions or subject to export‑control and data‑sovereignty disputes. Whether regulators authorize a million satellites, a reduced architecture, or none at all, the filing will accelerate debates about space governance, responsible de‑orbiting, and how to share increasingly crowded low‑earth orbits.