Starship Flight 14 Made Orbit. SpaceX's Space Data Centers Need About 100 Launches per GW
Starship Flight 14 reached orbit, but SpaceX's own ratio puts space data centers at about 100 full launches per GW, and the pads it's cleared to use allow 69 a year.
Starship has flown four times in the last year. By SpaceX's own ratios, 1 GW of computing in orbit takes about 100 full Starship loads, and the launch pads it's cleared to use today allow 69 flights a year.
Starship finally made it to orbit on Monday, on its 14th flight, and it brought cargo that SpaceX can actually sell. The ship let go of 26 Starlink V3 satellites before an engine problem cut the flight to about three hours. SpaceX's stock closed down about 2% that day anyway.
I get why the market shrugged. SpaceX only had to reach orbit once to prove it could, and its plan for data centers in space depends on doing it over and over. By the math in SpaceX's own filings, every GW of computing it wants to put in orbit takes about 100 full Starship launches. Starship has flown four times in the last twelve months.
The Starship launch math behind SpaceX's space data centers
We went through the big version of this in our orbital data center explainer. Elon Musk told Dwarkesh Patel in February that 100 GW of compute in orbit takes "on the order of 10,000 Starship launches." SpaceX's IPO filing says 100 GW a year would take "approximately one million metric tons of payload annually."
That ratio works out to about 100 kW of computing per tonne you launch. A Starship V3 is designed to carry more than 100 tonnes to low orbit. So a full Starship carries about 10 MW of AI satellites, and 1 GW takes about 100 trips. (That's my arithmetic on SpaceX's numbers, and it lands right where Musk's 10,000 does.) SpaceX hasn't published what one AI1 satellite weighs, so the filing's ratio is the best thing there is to work from.
Monday's flight also wasn't full. A fully loaded Starship can carry up to 60 V3 Starlinks, and each one weighs about two tonnes. Flight 14 took 26, so a bit under half a load, which is a sensible thing to do on the first trip to orbit.
How often Starship flies today
Starship has flown three times this year, in May, July and on Monday, a little over two months apart each time. SpaceX's Gigasat Factory in Bastrop, Texas, is supposed to be turning out 1 GW a year of space AI compute by late 2027. If the 100-launch ratio holds, that one factory's output needs about two Starship flights a week, every week.
The permits are tighter than I expected. The FAA cleared Starbase for up to 25 Starship launches a year in May 2025. Its environmental review in January allows up to 44 a year from Launch Complex 39A in Florida, and SpaceX still needs a launch license there. That's 69 a year between the two sites, which doesn't cover one factory's output even if every flight carried data centers and none carried Starlinks. SpaceX knows this, because it's planning a $100 billion spaceport in Louisiana with more than a dozen towers, and that site's first launch is 2029 at the earliest.
Musk's answer is speed. He told analysts on SpaceX's first earnings call in August that "probably a year from now we will be doing at least one flight a day, possibly more." One a day gets you 100 launches in a little over three months. SpaceX has run a rocket at that kind of pace before, since it flew 165 orbital launches in 2025, almost all of them Falcon 9. But Falcon 9 throws away its upper stage on every flight. Starship has to bring both stages back and fly them again for any of this math to be cheap, and it hasn't caught a ship yet.
The next two flights are penciled in close together. NASASpaceflight's provisional schedule has Flight 15 no earlier than October 19, with the first try at catching the ship on the tower, and Flight 16 from Florida no earlier than October 30. NSF called those dates "super provisional" itself. Still, two flights eleven days apart would beat this year's pace by a lot.
Where the launch cadence shows up in revenue
For $SPCX the money right now is Starlink. Each V3 satellite adds about 1 terabit per second of download capacity, and SpaceX sells that capacity as Starlink subscriptions. So every Starship full of Starlinks earns money while SpaceX practices the flight rate that AI1 needs. Musk has said the first AI1 goes up in the fourth quarter of 2027, so compute in orbit has no revenue line before then.
Rocket Lab ($RKLB) comes into this through its solar panels. It introduced silicon solar arrays in February meant to power "gigawatt-scale space-based data centers," and 1 GW of chips in orbit needs at least 1 GW of panels to run them. It hasn't named an orbital compute customer yet, though. Its record $2.36 billion backlog at the end of June comes from launches and spacecraft, and its second-quarter results don't mention data centers at all. Neutron, its new medium rocket, still hasn't flown. Rocket Lab is aiming to get it to the pad in the fourth quarter, and in August it said the window for a 2026 debut is narrowing. The stock fell 2.4% on Monday for reasons that had nothing to do with Starship.
SpaceX needs about 100 launches for each GW, and it has flown Starship four times in the last year. I'll be counting the days between Flight 14 and Flight 15 more closely than anything that happens on either one.