A rocket company that makes 52% of its money from AI is a strange thing to type, but here we are. SpaceX’s third-quarter numbers show a business that has quietly become a neocloud operator with a launch division attached.
According to its regulatory filing, as reported by TechCrunch, SpaceX posted $14.6 billion in revenue. The renamed Super Intelligence segment brought in $7.5 billion from the Cursor coding tool, the Grok bot, and a neocloud business now sitting on 1.7 GW of compute. That segment earned $431 million in profit, and would have earned more without $1.2 billion in one-time stock compensation tied to the Cursor acquisition.
The compute is the real story. Selling GPU capacity puts SpaceX in the same conversation as CoreWeave and Oracle, while Grok and Cursor put it up against OpenAI, Anthropic, and GitHub Copilot. Few rivals own both the model and the infrastructure underneath it. That is the pitch, anyway.
Read the headline carefully, though. About $1 billion of revenue came from Bitcoin price recovery and unspecified related-party income. Net income of $4.4 billion looks great against a $1 billion loss a year ago, but the first nine months of the year still show a $413 million loss. And capital spending was $19.6 billion this quarter, with CFO Bret Johnsen guiding to $35 billion to $40 billion in the fourth. That is a lot of cash going out the door.
Musk’s argument for spending it is that chips are no longer the bottleneck. Power, cooling, and infrastructure are. He is spending four days a week in Memphis trying to bring Colossus II online, and he says SpaceX is making its own turbine components with its metal casting capabilities because suppliers are quoting 12 to 18 month delays. If true, that is a real edge. Anyone who has tried to book power for a large cluster knows how tight it is. But it is a claim from the person selling the story.
The rocket side hasn’t gone away. The next Starship test in November will try to recover both stages, and Musk says reflying both is the actual milestone. Starlink reached 13.6 million subscribers, and he expects AI agents to push connectivity demand higher. Starship also matters to the orbital data center idea, which remains unproven.
Then there is the forecast. Musk predicted annual recurring revenue of $500 billion to $1 trillion by the end of 2027. Johnsen’s guidance is far more grounded: $33 billion to $37 billion in revenue for the first quarter of 2027. Even the top of that range annualizes to roughly $148 billion.
So does this matter? Yes. SpaceX is a serious AI infrastructure player with real revenue, not a side project. But the trillion-dollar number is noise until the capex starts converting into recurring contracts, and the fourth quarter will show how fast that happens.
