SpaceX (NASDAQ:SPCX) CEO Elon Musk is betting the cheapest place to scale AI computing may soon be above Earth, where satellites can tap near-continuous solar energy.
His Starmind vision would knit as many as one million satellites into a vast computer circling the planet. Maintaining it could require more than nine Starship launches every day, according to one aerospace engineer’s calculation.
Musk said this week the first satellites, carrying chips from Nvidia Corp. (NASDAQ:NVDA), may launch in the fourth quarter of 2027, with “significant scale” planned for 2028.
The Replacement Math Gets Brutal
Christopher Smith, a senior aerospace engineer at UC Berkeley’s Space Sciences Laboratory, ran the numbers in an analysis reported by the Financial Times. Replacing aging hardware every five years across one million satellites means roughly 200,000 new satellites a year.
Under Smith’s assumption that each Starship could carry 60 satellites, that works out to more than 3,300 flights annually, or about 9.1 launches every day. That is one launch roughly every two hours and 38 minutes, around the clock.
Starship has flown just 13 tests since 2023, several ending in explosions, and has yet to demonstrate full reusability.
SpaceX’s Own Filing Points to Thousands
SpaceX’s own IPO prospectus goes further. The company says deploying 100 gigawatts of orbital compute each year would require thousands of Starship launches, moving roughly one million metric tons to orbit annually.
Even at SpaceX’s projected payload capacity of 100 to 200 metric tons per flight, moving that much mass would require roughly 5,000 to 10,000 launches a year.
The filing also warns the plan relies on unproven technology and may never become commercially viable.
Traders Don’t Buy Musk’s Timeline
As of Wednesday, Kalshi traders give a 51% chance that Starship reaches space four times in 2026, a 39% chance of five flights, and just 2% odds of more than five.
A separate market gives an 11% chance of a 1-Mega Watt data center going live in orbit before 2029, and 18% before 2030.
Kalshi and Benzinga have an existing data collaboration agreement.
Google Says Possible but Not Yet Economic
Alphabet Inc. (NASDAQ:GOOGL) is exploring the same idea, but its research suggests the economics remain years away.
The company’s Project Suncatcher research found no fundamental physics blocking AI in orbit. The company radiation-tested its TPU chips and achieved 800 gigabits per second in each direction in a bench-scale optical-link test.
Google projected that launch prices could fall below $200 per kilogram by the mid-2030s. At that level, orbital data centers could become roughly comparable with terrestrial facilities on an energy-cost basis. Heat, ground links and in-orbit reliability remain open problems.
SpaceX and Google agree on the destination, but not the timetable: Musk targets significant scale in 2028, while Google’s model suggests competitive economics may not arrive until the mid-2030s.
Image: Shutterstock
Login to comment