SpaceX filed with the Federal Communications Commission for authorization to launch up to 1 million satellites as part of what amounts to a data center megaconstellation — a distributed orbital compute network following the merger of SpaceX and xAI into a single entity. An FCC filing is not a product announcement. It is a regulatory claim on orbital slots and radio frequencies. SpaceX is staking out spectrum before competitors can.
The scale is the news. Starlink's current operational constellation is approximately 6,000 satellites. The proposed data center constellation would be 167 times larger. At current Starship launch rates, deploying 1 million satellites would take decades. The filing is a ceiling on ambition, not a production schedule. But the ambition it describes — a distributed orbital compute fabric covering every point on Earth's surface with AI processing capability — is now a formal regulatory position, not a tweet.
Current Starlink operational satellites: approximately 6,000. Proposed megaconstellation: up to 1 million satellites. Current Starship launch cadence: dozens of flights per year. Years to deploy 1 million satellites at current rate: measured in decades. The FCC filing secures orbital slots; the production timeline is a separate and much harder problem.
The feasibility question depends almost entirely on launch cost per kilogram. At current economics — roughly $100 per kilogram to LEO with Starship — a 1-million-satellite constellation using even modest 100kg compute satellites would cost $10 trillion in launch costs alone, before hardware. That number is not achievable at current pricing.
Musk's bet is that Starship's full reusability program drives launch costs down by two orders of magnitude — toward $10 per kilogram or below. At $10/kg, the launch cost for a 100-satellite, 100kg compute network becomes $100 million: competitive with a mid-scale terrestrial facility. The thesis requires a cost reduction that has not yet been demonstrated but that SpaceX's engineering trajectory makes plausible over a 5-to-10-year horizon.
One million compute satellites, each drawing even modest power, represents a significant thermal rejection challenge — solved entirely by passive radiation to deep space. Each satellite would feature solar arrays for power and radiative panels for heat rejection. No coolant. No pumps. No water. The thermal architecture of a 1-million-satellite constellation is fundamentally simpler than a single 100 MW terrestrial data center.
This simplicity is a structural advantage that the terrestrial cooling industry cannot replicate. A satellite that fails thermally is replaced; a cooling system failure in a ground-based hyperscale facility triggers emergency protocols and potential SLA violations. The orbital model trades maintainability for thermal simplicity at scale. Whether that trade is commercially viable at the system level depends on the launch cost curve — the same variable that determines everything else about this thesis.
China's CASC filed its own five-year plan for 2,800 orbital data center satellites by 2030. StarCloud raised $170M at a $1.1 billion valuation for the same concept. The race for orbital compute spectrum and orbital slots is accelerating. The FCC filing is SpaceX staking its claim early, before the ITU orbital slot allocation process becomes a bottleneck. Spectrum and slots are finite. The company that files first has a structural advantage in a market that does not yet exist commercially — but will.