Musk’s Space Megaplan Triggers Regulatory Firestorm

man in black cap looks ahead indoors — Photo: cristiano barni / Shutterstock

Elon Musk’s push for up to one million orbital data-center satellites could radically reshape AI—and raise huge questions about cost, risk, and control in space.

Story Snapshot

SpaceX has asked federal regulators for permission to deploy up to one million solar-powered “data center” satellites in low Earth orbit.
The plan relies on the Starship rocket, Starlink laser links, and new high‑power “AI Sat” designs that do not yet exist at scale.
Experts warn that key parts of the project—heat management, chip reliability, and trillions in costs—are still unproven.
Regulators moved the filing into public comment, so Americans can weigh in on orbital crowding, space junk, and who controls AI compute in space.

What SpaceX Actually Filed With Federal Regulators

SpaceX formally submitted an application to the Federal Communications Commission to operate up to one million satellites as orbital data centers between about 500 and 2,000 kilometers above Earth. The filing describes narrow orbital “shells,” only about 50 kilometers thick, stacked in low Earth orbit to host these solar-powered compute nodes. SpaceX says each satellite will plug into the existing Starlink network through high-bandwidth laser links, then hand traffic down to approved ground stations on Earth.

The company claims this giant constellation could deliver roughly 100 gigawatts of AI compute capacity by launching about one million tons of satellites, each providing around 100 kilowatts of power per ton dedicated to processing. The satellites would lean on the sun almost full time, with radiative cooling systems and an expected operational lifespan of about five years before controlled deorbit to limit long-term debris. SpaceX’s filing argues this will be the “most economical and energy-efficient” way to meet exploding AI demand.

The Hardware Vision: Starship, AI Chips, And “AI Sat Mini”

SpaceX tells regulators it will use its heavy-lift Starship rocket to deploy the initial orbital shells of the data center constellation. Starship’s high payload capacity is central to the proposal, because lifting hundreds of thousands of satellites demands regular, low-cost launches. SpaceX also plans pilot tests using upgraded Starlink V3 satellites later in 2026, turning some of them into early compute nodes if the FCC signs off. These pilots are meant to prove basic functionality before any mass rollout.

Beyond the pilots, the company and outside reporting describe a new class of satellite sometimes called “AI Sat Mini,” designed to carry roughly 100 kilowatts of electrical power for onboard AI processors. To stay safe and reduce clutter, SpaceX says each satellite will operate for only about five years, then deorbit under its own drag or remaining propellant, similar to current Starlink practice. The architecture leans heavily on proven Starlink laser-link designs, but scales them up for high-density computing instead of only broadband internet.

Where The Physics And Engineering Get Tough

Analysts highlight that much of the technology required for mass orbital compute does not yet exist in finished form. Current leading AI chips, such as those made by Nvidia, are built for Earth and are not hardened against space radiation; engineers warn they suffer random “bit flips” from high-energy particles on every low Earth orbit pass. Large-scale error-correcting software for millions of chips in space has not been demonstrated in real conditions, so reliability over years is an open question.

Thermal management may be even harder. Detailed technical reviews estimate that a flagship AI satellite would need to reject more than twice the heat handled by the International Space Station, but through only about one quarter of the radiator area. That implies roughly eight times more heat rejection per square meter than any spacecraft has managed so far, which pushes the limits of known space cooling designs. A new radiation-hardened “Tera F” AI chip has been talked about, but it has not been built or tested publicly, forcing designers to assume swap‑able modules until that chip exists.

Money, Timeline, And Wall Street Skepticism

Financial research firms and banks question whether this project can pay off on any realistic timeline. One independent analysis estimates the full one-million-satellite vision could cost between two and five trillion dollars over time, far beyond typical commercial space budgets. Gartner has gone as far as to label orbital data centers “peak insanity,” arguing they could cost around three times more than equivalent data centers on Earth when launch, maintenance, and specialized hardware are included.

SpaceX’s own confidential S‑1 filing for its stock market debut admits that orbital AI compute “involves significant technical complexity and unproven technologies, and may not achieve commercial viability.” Other bank research, such as from Deutsche Bank cited in filing analyses, suggests true cost parity with Earth-based data centers may not arrive until well into the 2030s, not within two to three years as Musk has suggested. This gap between bold public claims and careful investor warnings is driving sell ratings and lower price targets on SpaceX shares.

Regulators, Critics, And The Battle Over Crowded Skies

The Federal Communications Commission’s Space Bureau accepted SpaceX’s filing and quickly moved it into a public comment process, with experts openly doubting technical feasibility. Comment notices describe a network of up to one million non-geostationary satellites and list multiple waivers SpaceX seeks, including relaxed timelines and bond requirements if it fails to launch on schedule. By granting waivers, regulators could give SpaceX more freedom on cadence and accountability than smaller rivals receive, raising fairness questions.

Mainstream outlets and policy groups warn about wider consequences. Some experts quoted by Brookings worry that a million orbital data centers could “ruin the night sky, generate more space junk, and increase the risk of orbital collisions,” and they also question whether the “green” solar pitch is just marketing. Others compare the plan to earlier “big idea” infrastructure projects that never scaled, noting that large non-broadband constellations have rarely gone from regulatory filings to full operation because of launch cost and space-environment limits.