Staff Reports
SpaceX just asked the federal government for permission to turn low Earth orbit into a giant computer farm. The idea sounds like science fiction: racks of AI chips humming above the clouds, linked by lasers, drawing power from the sun. But the FCC has officially accepted the filing and opened the comment period, so this dream — or expensive hobby — is now on the public record.
What SpaceX is asking for
SpaceX’s application to the FCC asks to build an “Orbital Data Center” system of up to one million non‑geostationary satellites. The filing lays out satellites from about 500 km to 2,000 km, optical inter‑satellite links, solar power, and requests for regulatory waivers so the plan can be reviewed outside the usual process. The filing even tosses in lofty language about reaching a “Kardashev II‑level civilization,” which sounds great at cocktail parties but doesn’t solve heat or radiation problems. FCC Chairman Brendan Carr put the filing on people’s radar and opened the public comment period — which is exactly what should happen when someone proposes to carpet orbit with compute nodes.
Big physics problems won’t read a PR deck
The engineering issues are brutal and plain. GPUs used for AI dump kilowatts of heat per rack. In space there is no air to carry heat away, only radiation — and radiators need surface area and mass. Experts call this the “physics wall”: radiator mass and size scale badly as you cram in more compute. Add space radiation that flips bits and ruins components, and you have to overweight satellites with shielding or redundancy. Professor Benjamin C. Lee and other engineers tell the same story: these aren’t software bugs you fix with an update. They are hard, physical limits.
Money, Starship, and whether customers will pay
SpaceX’s own investor filing for a planned IPO warns the idea is “in early stages” and “may not achieve commercial viability.” Translation: the company is hedging against lawsuits and investor fury if this turns out to be more fantasy than profit. The whole plan also depends on Starship delivering routine, ultra‑cheap heavy lift. If Starship stays late or expensive, launching thousands or millions of radiator‑heavy satellites becomes a money pit. And even if the logistics work, who will pay a premium for orbital compute versus cheap terrestrial data centers that already have power, cooling, and maintenance crews?
Regulators, industry and taxpayers should pay attention
This FCC filing is exactly the moment for sober questions. The public comment window should invite input from defense, commercial cloud providers, debris experts and operators of existing satellite constellations. Space traffic, collision risk, debris mitigation and spectrum use are not theoretical. If the FCC grants waivers without rigorous analysis, we risk a repeat of the “move fast, break things” mindset — except broken things up there don’t stay on a server rack, they become debris for a generation.
Bottom line
Innovation matters. But grand plans and glossy slides do not replace physics or markets. SpaceX can build amazing things, and maybe parts of this will work for niche on‑orbit processing. But turning orbit into a second Earth full of data centers is a bet on unproven hardware, unproven economics, and an unproven launch cadence. Regulators and investors should demand evidence, prototypes, and hard numbers — not sci‑fi slogans. Let’s cheer American ambition, but not hand out regulatory fast‑passes for ideas that still need to survive basic science and a cold balance sheet.
