SpaceX wants to burn up 4-ton orbital data centers. An astronomer says the safety math is "nonsense."

A SpaceX satellite the company calls "Starmind" weighs just under 4 metric tons, packs AI compute hardware into orbit, and is supposed to come down by burning up in the atmosphere. In a 12-page letter to the Federal Communications Commission, SpaceX says every piece that survives re-entry will hit the ground at under 15 joules of kinetic energy, the FCC's threshold for "no risk to humans" (PCMag; archive of SpaceX ex parte letter, ICFS File No. SAT-LOA-20260108-00016).

University of Regina astronomer Samantha Lawler, a long-time critic of mega-constellations, says the underlying casualty-risk table is "nonsense" and must be transparent enough for independent parties to replicate. The dispute is the first public test of whether the FCC's verification regime scales to a new class of orbital hardware: not a communications satellite, but an orbital data center.

The 15-joule figure works out to roughly the kinetic energy of a 1.7-inch hailstone falling, SpaceX has said in past filings. NASA Debris Assessment Software treats impacts below that limit as posing no casualty risk, and SpaceX's letter says no component of the satellite exceeds the limit, putting the satellite's "aggregate casualty risk... at 0, below the 0.0001 (1-in-10,000) threshold" (archive of SpaceX ex parte letter).

SpaceX lists AI compute hardware, solar panels, the thermal control system, the avionics assembly, the thrusters, and the "satellite aggregate," a catch-all for the rest of the mass, as components that will survive re-entry. The bus structure, propulsion, and laser optical communication links are expected to burn up completely.

Lawler does not contest the 15-joule threshold. Her objection is that the casualty-risk table in the filing comes from SpaceX, and the underlying calculations should be transparent and replicable by independent parties, she said. Without that, the operator is asking the public to take a "no risk to humans" claim on faith, on hardware that has no disposal precedent.

Starlink disposal is the easier case. SpaceX has deorbited thousands of Starlink satellites, about 260 in the last six months by Tom's Hardware's count, and the company has published a demisability playbook for them (Tom's Hardware; Starlink Approach to Satellite Demisability). An orbital data center is a different object: roughly an order of magnitude heavier per satellite, packed with economically valuable compute, and designed to be brought down deliberately at end of life.

The Space Bureau accepted the ex parte filing under DA 26-113. The question on the desk is whether the casualty math the operator submits is enough to clear the 0.0001 threshold for the new satellite class (FCC DA 26-113 filing acceptance notice). Independent analyst Yuan Chang has read the application in detail, and his read is that the calculation is being left largely to SpaceX to assert (Yuan Chang blog). Lawler's argument is the same: stop taking the table as asserted and let outside parties check the work.

The first public comment cycle on the orbital data center disposal approach is open under ICFS File No. SAT-LOA-20260108-00016. The verification template the FCC settles on for these satellites will be inherited by every orbital data center that follows, including ones built by companies that have not yet filed. The question is whether "no portion of the spacecraft will impact the surface of the Earth with a kinetic energy above 15 joules" is a fact the agency can verify, or a line in a letter the agency accepts.