SpaceX reported $14.6 billion in revenue for the September quarter. Of that, $7.56 billion came from renting out computing power, which makes the AI unit bigger than Starlink for the first time.
SpaceX's Q3 in five lines
The company filed results on October 20 for the quarter ending in September. Revenue rose 161% from a year earlier and beat the $12.86 billion that analysts tracked by Visible Alpha expected.
Metric | Q3 2026 | Consensus estimate |
|---|---|---|
Revenue | $14.6B | $12.86B |
Operating profit | $2.79B | $2.33B |
Super Intelligence revenue | $7.56B | $6.47B |
Connectivity revenue | $5.25B | Not published |
Net income | $4.4B | Not published |
Fortune described the operating result as the company's first operating profit since it went public in June. That framing is Fortune's, and it holds up against the prior quarter, when SpaceX posted an operating loss of $143 million, which makes the swing look less like a trend than like one very good quarter.
Earnings per share came in at $0.35.
Super Intelligence is a compute landlord, not a model lab
The AI segment is called Super Intelligence, and its business is renting servers.
Not a model lab.
Customers named in coverage include Anthropic, Google, Grok, and X, which puts a mix of model developers and social platforms on the same invoice.
That's worth being precise about, because "AI revenue" invites the wrong picture. SpaceX isn't training frontier models or selling tokens. It's selling GPU time, and the margin depends on utilization, power contracts, and chip supply rather than on model quality.
The segment brought in $7.56 billion, against $6.47 billion expected. It accounted for about 52% of company revenue.
Two growth rates get quoted for it, and they measure different things. Year over year, revenue rose roughly ninefold from $813 million in the same quarter last year. Quarter over quarter, it nearly tripled from $2.56 billion in Q2. Both are accurate. Neither is the whole story.
Starlink grew 69%, and still came up short
The connectivity segment, which is mostly Starlink, posted $5.25 billion in revenue, up 69%. Operating income for the segment nearly doubled to $2.37 billion.
Then the number that spoiled the mood: 1.6 million net new consumer subscribers in the quarter, against the 2.29 million analysts expected.
Seven hundred thousand missing subscribers is a demand signal, not an accounting detail. Starlink still added 1.6 million people, which is real growth. It just wasn't the growth the models assumed, and the growth rate is what the valuation leans on.
SpaceX's launch business brought in $1.77 billion, up 6%. That's the smallest of the three segments now, which would have been an absurd sentence to write two years ago.
Why operating profit and net income differ by $1.6 billion
Operating profit was $2.79 billion, ahead of the $2.33 billion analysts expected. Net income was $4.4 billion, which is larger.
The gap sits below the operating line. SpaceX didn't break out every line in the summary coverage, so the exact composition isn't public in the reporting so far, and the $1.6 billion difference is best treated as a direction rather than a precise figure.
Why does the distinction matter? Because the operating number is the one that says whether the core business pays for itself. The net number can be flattered by things that won't repeat.
Last year's comparison is brutal in a useful way. A net loss of $5.5 billion a year ago against $4.4 billion of net income now is the kind of swing that turns a growth story into a profitability story, at least for one quarter.
Capex of $19.6 billion, mostly aimed at AI
SpaceX spent $19.6 billion on capital projects in the quarter. AI Weekly reported that $16.4 billion of that went to AI infrastructure, a figure attributed to the publication rather than the filing summary.
That ratio explains the strategic bet. The company is building data centers at a pace that assumes demand keeps arriving, and the AI segment's revenue is the evidence it currently does.
Capital intensity of this scale changes the risk profile. A launch company fails if rockets fail. A compute company fails if a customer leaves mid-contract, or if power costs rise, or if the next chip generation arrives on someone else's schedule, none of which the company fully controls.
What analysts expect from SpaceX's spending
Visible Alpha's Melissa Otto expects capital spending to climb from about $70 billion in 2026 to $197.4 billion in fiscal 2028. She also projects debt rising from $50.7 billion in 2025 to $350 billion by 2029.
Those are sell-side forecasts, not company guidance. Treat them as a picture of what the market is pricing rather than what will happen.
If they're roughly right, SpaceX is becoming a leveraged infrastructure company with a rocket division attached. That's a different animal from the one that IPO'd in June.
The Grain deal, the debt, and a promise about stock
Three items sit outside the quarter and matter to the next one.
The company agreed this month to buy 800MHz airwaves from Grain Management, in a deal the Wall Street Journal put at about $8 billion, with FCC approval still outstanding and no closing date announced. More airwaves means more capacity for connectivity, which is the slower-growing half of the business now.
Musk said the day before earnings that he plans to give SpaceX stock to every American child next year, a promise with no structure attached yet. Rough estimates put the recipient pool around 72 million minors, and media calculations value the pledge near $10.5 billion.
And the debt number from Otto's forecast is the one to keep in view. A company borrowing to build data centers is making a bet that compute demand stays tight for years.
For this quarter, the bet paid. Compute revenue passed Starlink, operating profit turned positive, and the launch business became a rounding error by comparison. Whether that holds depends on customers who can build their own clusters if the price stops making sense.
