A million satellites could rewrite the night sky to power AI on the ground

At dawn and dusk this decade, a permanent band of reflected light may settle over the sky where there used to be none. The light will not come from a city. It will come from a fleet of data-center satellites parked in a near-polar orbit chosen because it is the most power-efficient place to feed solar arrays, and the worst possible place for the night sky.

A modeling study posted to the research repository arXiv on August 3 by Aaron Boley, an astronomer at the University of British Columbia, is the first public-facing estimate of what three pending satellite filings would actually do at twilight. The study, which has not yet been peer-reviewed, simulated three configurations from proposals now in front of the Federal Communications Commission: a single ring from startup Cowboy Space Corporation, a crisscross of two rings from Blue Origin, and SpaceX's plan for crisscrossed rings plus an additional shell. (Scientific American)

The geography of the problem is the polar cusp, the band of orbit between day and night where a satellite receives almost constant sunlight. For solar-powered compute, that is the only spot worth sitting in. For anyone trying to see the stars, it is the worst spot in the sky: a low band that dawn and dusk observers already scan for auroras and the zodiacal light. Boley's team found that reflecting sunlight off large solar arrays in that orbit creates persistent fixtures of light visible from both remote dark-sky preserves and the most light-polluted cities.

The proposals are not theoretical. Cowboy Space, a well-capitalized startup, has asked the FCC for permission to operate a constellation. SpaceX and Blue Origin have filed for larger systems. Across the three filings, the proposed satellite count runs from 20,000 to one million, a number that puts the largest proposal on the same scale as every operational satellite ever launched, combined. The pitch on the ground is that orbital data centers escape the local opposition and environmental costs that have made new terrestrial AI data centers politically hard to site. Moving that burden to orbit shifts the cost onto anyone who looks up.

"People have lost the sense of the night sky because of the light-polluted cities, and now we're rewriting that story with artificial satellites," Boley told Scientific American. Light pollution has already erased the Milky Way from roughly a third of the global population, and the orbital proposals would add a sky-wide contributor to the same loss: one that no city council can vote out.

The FCC filings are proposals, not deployed hardware. The brightness numbers in Boley's preprint are model outputs, not measurements, and peer review is still ahead. Even so, the magnitudes are the point. The smallest of the three configurations, Cowboy Space's single ring, would already create a persistent twilight fixture of the kind astronomers currently work around. The next move belongs to the FCC, which has to decide whether satellite data centers fall under the rules it already uses for communications constellations, a question the current filings do not resolve.

The trade-off is being set in documents that few people will read, on behalf of an industry that everyone can already see. The night sky has been treated as a commons because no one had to apply for a permit to look at it. That premise is now under review, in three filings, at one agency.