SpaceX has put orbital AI compute in a corporate planning document, which is a bigger deal than a Musk post. Its filings say it expects to begin deploying orbital AI compute satellites “as early as 2028.” Its SEC filings also include a risk factor saying orbital compute may not be commercially viable, because the plans rely on unproven technology and on Starship being reusable. That is a company hedging its own plan, which matters when you read the next claim.
On October 30 (Pacific time), Musk made a much bigger claim. Replying to a chart of investor Shaun Maguire’s SpaceX financial model, he wrote that orbital compute deployed per month “probably exceeds terrestrial around the end of 2028 and then far exceeds it in 2029.” The charts he replied to assume SpaceX’s ground-based compute grows from about 1 gigawatt in early 2026 to 21 gigawatts by the end of 2028. So the comparison is monthly additions against monthly additions, not total capacity against total capacity. That’s a lower bar than “orbit overtakes Earth,” but it’s still a very high one.
The company’s own timeline is more cautious than the forecast. An SEC-filed interview with CFO Bret Johnsen described demonstrations as possible in 2027 and meaningful amounts of compute in space targeted for 2029, if the company stays on schedule. Musk has said the first Starmind AI1 satellites should launch in the fourth quarter of 2027. Those aren’t strictly contradictory, because deployment pace and total capacity are different things. But a forecast that monthly orbital additions pass Earth’s about a year after the first launch is a steep ramp.
The filings do give some numbers. One report of the prospectus says early satellites are expected to produce about 100 kilowatts of compute power, with a longer-term plan to scale much higher. SpaceX has also asked the FCC for permission to fly up to a million compute satellites. What’s missing is the number that would test Musk’s claim: how much compute SpaceX plans to deploy each month, and in what unit. Musk didn’t define “orbital compute” in the post, so the measure could be power, chips or something else.
If the unit is power, which Musk didn’t say, the arithmetic is rough but instructive. One commenter under the post estimated orbital compute would need to pass 3 to 4 gigawatts a month to beat all terrestrial additions. At 100 kilowatts per satellite, 3 gigawatts is about 30,000 satellites a month. That’s my division, not a SpaceX figure, but it shows the scale the forecast implies.
There’s no customer in the filings either. The reporting I found says SpaceX hasn’t named an organization contracted to use an orbital satellite, and that its own investor materials call the effort early-stage and unproven at commercial scale. One analysis also notes that Musk’s own Starship schedule has already slipped, with weekly flights now guided for the first half of 2027.
None of this says orbital compute won’t work. It says SpaceX has disclosed an intent and a start date, and Musk has made a forecast that needs a capacity metric nobody has published. Until SpaceX defines the unit and the monthly target, the late-2028 crossover can’t be checked, only waited for.
