Space data centers could leave more satellites than stars in the night sky, a Columbia-led report warns

AI & Compute

Columbia Report Warns Orbital Data Centers Could Outnumber Stars

A Columbia-led report warns filed plans for orbital AI data centers — up to one million SpaceX satellites — could put more satellites than stars in the night sky.

By
Sophie Lindqvist
Filed
Channel
AI & Compute
Read
4 min read

SpaceX has asked the Federal Communications Commission for permission to operate up to one million data-center satellites. A Columbia-led working paper, published Nov. 12 by Columbia Climate School, says that if filed plans are fully built, more satellites than stars could be visible at many times in many places around the world.

The report, titled Space-Based Data Centers Roadmap: The Case for Caution and dated November 2026, was led by David Sandalow, inaugural fellow at Columbia University's Center on Global Energy Policy. Nine co-authors from Columbia, Georgetown, Northwestern and Lawrence Livermore National Laboratory contributed. The authors call it a working paper that has not been peer reviewed — a caveat worth holding onto before treating its estimates as settled.

What would make these satellites so visible?

The answer is size and number. The report states SpaceX's AI1 satellite has a 70-meter wingspan, more than twice the size of a Starlink satellite. Some experts cited expect data-center satellites to reach 100 meters tip to tip because of their solar arrays and radiators.

Radiator coatings create a dual problem. They reflect sunlight, which makes them bright in visible light. They also shed heat as infrared, which makes each megawatt-class satellite a megawatt-scale infrared source for as long as it computes.

The authors note that darkening a radiator carries a cost. It cuts visible brightness but makes the radiator absorb more sunlight and cool less well. The infrared emission cannot be removed at all, because it is the waste heat itself. The report quotes the American Astronomical Society's filing against SpaceX's January 2026 application, which called the plan "an unprecedented threat to the visual integrity of the night sky."

How large are the filed plans?

Roughly 19,000 satellites orbit Earth today. Full implementation of proposed data-center plans would raise that number more than 50-fold, the report finds. Four companies besides SpaceX have filed for several hundred thousand more satellites since January 2026, including Blue Origin and Starcloud:

  • SpaceX: up to 1,000,000 satellites, filed with the FCC in January 2026
  • Orbital: up to 100,000 satellites
  • Starcloud: a proposed constellation of 88,000 satellites
  • Blue Origin: up to 51,600 satellites

SpaceX is also designing a 72-GPU rack for orbit, according to earlier reporting by MLQ.ai. Starcloud was selected this week for a federal mission-authorization pilot, MLQ.ai also reported.

Do the numbers add up?

The authors stress-test the claims and find gaps. SpaceX has suggested in regulatory filings that a 100-kW data-center satellite could weigh one tonne. The report's own estimate for the power system alone is about 2.2 tonnes.

Heavier satellites mean more launches. The authors estimate that maintaining an orbital fleet serving 10% of projected 2030 AI data center demand would require at least 700 Starship-sized launches per year — about two per day, in perpetuity, for as long as the fleet operates.

Even so, the report concludes that high costs are unlikely to block initial deployment. Companies may accept losses to establish market position. That logic has played out before in terrestrial cloud markets.

What else does the report flag?

Beyond the night sky, the authors list:

  • Debris fields that could make parts of low Earth orbit unusable for years
  • Casualties and property damage from de-orbiting tens of thousands of satellites per year
  • Ozone depletion from rocket exhaust and re-entering satellites
  • Added warming and interference with astronomy

The report says current governance is fragmented and not suited to constellations of this size. That is the quiet center of the argument: no regulator currently owns this problem end to end.

What do the authors recommend?

Their recommendations are specific. They propose phased deployment with licenses renewed annually based on brightness, de-orbit and emissions performance. They urge regulators to refuse licenses for constellations visible from Earth's surface. They call for an end to the FCC's categorical exclusion of satellite launches from National Environmental Policy Act review. They also propose an international space traffic management organization.

The authors point to SpaceX's own FCC statement that "initial operations under this license will be limited to a significantly smaller number of satellites and re-entries than the maximum authorized." They say SpaceX and other companies should strictly follow that principle.

The report is not peer reviewed, and its mass estimates have not been reconciled with SpaceX's filings. But with five companies seeking well over a million satellites combined, the burden of proof on launch cadence, thermal design and debris risk now sits squarely with the applicants.

Source: mlq.ai

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,000 articles

Related articles

« Previous article