SpaceX
1 Million AI Satellites

The January 2026 FCC Application

On 30 January 2026, SpaceX submitted an application to the Federal Communications Commission seeking authority to launch and operate up to one million satellites under the name "SpaceX Orbital Data Center System." The FCC accepted the filing four days later and opened a public comment period running to 6 March 2026, with over 1,500 public comments ultimately submitted.

The application describes satellites operating as distributed AI computing nodes, powered by near-continuous solar energy in orbit. SpaceX argues that rising electricity costs and grid strain are making terrestrial data centres increasingly expensive, while falling launch costs — driven by Starship — make orbital computing viable for the first time. In the filing, SpaceX framed the project as "the first step towards becoming a Kardashev Type II civilisation — one that can harness the Sun's full power."

The proposed satellites would fly in orbital shells between 500 and 2,000km altitude, at 30-degree and sun-synchronous inclinations to maximise solar exposure. SpaceX requested waivers from the standard FCC milestone requirements — which typically require half a constellation to be deployed within six years — arguing these rules were designed to prevent spectrum warehousing rather than to constrain genuine deployment plans. No specific satellite size, mass, deployment schedule, or cost estimate was included in the filing.

The "AI Sat Mini" Revelation

At a March 2026 event in Austin, Texas, Musk showed an illustration of what he called the "AI Sat Mini" — the initial version of the orbital data centre satellite. The name is misleading. Based on the scale comparison Musk presented alongside SpaceX's Starship rocket (124 metres tall and the largest rocket ever built), each "mini" satellite would be over 170 metres long — larger than the International Space Station, which spans 109 metres.

Each satellite would provide 100 kilowatts of power for onboard AI processors, with large solar arrays dominating the structure and a 100 square metre radiator panel for heat rejection. Musk dismissed concerns about cooling — a common critique of orbital data centre concepts — saying "for some reason there's been a bizarre debate about radiators in space. It's safe to say SpaceX knows how to do heat rejection in space with 10,000 satellites in orbit." The "mini" label, he clarified, reflects plans for future even larger satellites providing a full megawatt of power.

When the size was revealed, astronomer Samantha Lawler — who had already modelled the brightness impact of the constellation — told PCMag: "We thought the size we assumed was ridiculous, but this graphic shows that we actually underestimated what SpaceX is planning to do." At this scale, each satellite would be an extremely bright object — not a faint point of light but a structured, reflective body larger than anything currently in orbit.

More Satellites Than Stars

The human eye can see approximately 4,500 stars in an unpolluted night sky. Astronomers at the University of Regina and the University of British Columbia ran simulations of what SpaceX's proposed constellation would look like and found that for large portions of the night and year, throughout the world, more satellites would be visible than stars. This is not a marginal degradation — it is a categorical transformation of what the night sky is.

The satellites would fly higher than standard Starlink — up to 2,000km — making them visible for longer periods each night, including well into the dark of midnight when Starlink satellites are typically in shadow. Musk estimated individual satellites would be hard to see from one to another, citing the vastness of space. Astronomers responded that vastness at orbital scale does not prevent visibility from the ground, and that simulations based on the filing's orbital parameters show tens of thousands of sunlit objects simultaneously visible.

The Royal Astronomical Society, the American Astronomical Society, the European Southern Observatory, and DarkSky International all filed formal objections. The RAS deputy executive director called the proposals — including both the SpaceX filing and Reflect Orbital's mirrors — a scheme that would "permanently scar the natural landscape" and urged the FCC to reject both.

When Orbit Becomes Unusable

Beyond the night sky, physicists and orbital debris experts have raised a more fundamental concern: Kessler Syndrome. Named after NASA scientist Donald Kessler, the scenario describes a cascade where collisions between satellites generate debris, which causes more collisions, which generates more debris — until portions of Earth orbit become completely unusable for generations. At 14,500 active satellites today, this risk is already being managed carefully. At one million, the collision probability mathematics change entirely.

Jonathan McDowell, astronomer and space object cataloguer at Harvard-Smithsonian, told The Register: "One million satellites are going to be a big challenge for astronomy, especially as they are in higher orbits which is worse for us." Higher orbits mean the satellites remain in sunlight longer each night, and take decades rather than years to naturally deorbit if attitude control is lost.

Reflect Orbital + SpaceX: A Combined Picture

The SpaceX filing arrived just weeks after Reflect Orbital's application for Eärendil-1 entered the FCC public comment process. Astronomers and environmental groups found themselves fighting two separate filings simultaneously — and began covering them as a single pattern rather than isolated proposals.

Why AI in Orbit?

The business logic behind orbital data centres rests on three claims: that space offers near-unlimited solar power at no fuel cost, that the vacuum of space provides natural cooling without water consumption, and that falling launch costs from Starship will soon make orbit cheaper than land for compute-intensive AI workloads. Musk is not alone in making this bet — Google is testing orbital AI data centres with two prototype satellites planned for early 2027, Jeff Bezos has predicted "giant gigawatt data centres in space within 20 years," and OpenAI's Sam Altman explored buying a rocket company to pursue the same idea.

Sceptics dispute each leg of the argument. Cooling in space requires radiation rather than convection — meaning enormous radiator panels, not passive cold. The ISS's solar panels cover half a football field and produce just 100 kilowatts. Replicating a standard 100-megawatt terrestrial data centre in space would require panels 500 to 1,000 times that size. Deutsche Bank estimates cost parity with terrestrial data centres won't arrive until well into the 2030s — years beyond Musk's "30 months" claim. MIT professor Olivier de Weck put it plainly: "Is that feasible? Yeah, I think it's feasible, but not next year and certainly not in three years."

The manufacturing scale required is equally staggering. To supply the AI chips for the constellation, Musk announced "Terafab" — a $20 billion chip fabrication facility in Austin aimed at producing 200 billion chips per year, 50 times the combined current production of all advanced chip manufacturers. Ars Technica estimated the bare-bones launch cost alone for one million satellites would exceed one trillion dollars — roughly SpaceX's entire estimated IPO valuation.