SpaceX Q3: AI Revenue Topped Rockets, Profit Hit $4.4B
Elon Musk's rocket company just had its best quarter ever — and the money barely came from rockets. SpaceX swung to a $4.4 billion third-quarter profit on $14.58 billion of revenue, blowing past Wall Street's estimates, and the engine behind the beat was an AI business that looks, on closer inspection, almost nothing like an AI lab. It looks like a landlord. One that rents out computing power to the very companies beating it.
Revenue jumped 161% from a year earlier and landed comfortably above the $12.72 billion analysts expected, according to LSEG data. The bottom line was the real shock: $4.4 billion of net income against a $1.1 billion loss in the same quarter last year and a $1.57 billion consensus profit. Per-share, it earned 45 cents, outstripping a 13-cent average estimate. That is not a beat. That is a different postal code from the estimates.
Photo: Starship igniting during its fifth integrated flight test. The rocket that defines SpaceX's image had almost nothing to do with its best quarter. (Steve Jurvetson / Wikimedia Commons, CC BY 2.0 — file photo, not from the earnings quarter.)
The quarter that flipped the story
Context matters here, because SpaceX's mood music was very different three months ago. The second quarter ended with a $541 million net loss on $7.81 billion of revenue — a result the market treated as a stumble, piling onto a stock that had already shed roughly half its value since its record June debut. This quarter, revenue nearly doubled sequentially. The swing from red to black took one quarter and about $7 billion of incremental revenue.
We compared the headline figures against the Reuters wire and two of its republications — every number matched, right down to the 45-versus-13-cents per-share spread. When a story is this dramatic, the first job is making sure nobody fat-fingered a decimal. Nobody did.
Musk, for his part, was in triumphant mode on the investor call. "We are in an inflection moment for SpaceX across launch, connectivity and superintelligence," he told investors. It's a line that tells you everything about where he wants the narrative to go: not a satellite-internet company that also does AI, but a superintelligence company that happens to own satellites.
The AI business that isn't an AI business
Here's the detail that should rewrite how you think about this company. SpaceX's AI revenue rose more than ninefold in the quarter to $7.6 billion. But almost none of that came from Grok, the chatbot, or any model at all. It came from renting out data-center capacity — compute contracts with Anthropic, Alphabet's Google, and a startup called Reflection AI.
Sit with that for a second. Two of SpaceX's paying AI customers are companies whose models are, by Reuters' own account, currently overshadowing Grok. SpaceX is, in effect, selling shovels to the miners who are beating it — and the shovel business is now bigger than most actual AI labs' entire revenue lines. The company describes its AI operation as spanning Grok, social-media platform X, and a fast-growing data-center operation. But Reuters' verdict is blunt: SpaceX is making its AI money like a cloud provider, not an AI lab.
There is something deliciously pragmatic about this, and something slightly absurd. The world's loudest AI maximalist is, this quarter, mostly a compute landlord. The deeper numbers, reported by the specialist press from the company's disclosures, sharpen the picture: the Super Intelligence segment turned a $431 million operating profit after a $1.26 billion operating loss last quarter and a $1.72 billion loss a year ago. But strip out the accounting gloss and the picture wobbles. The adjusted figure SpaceX prefers — $4.95 billion — excludes depreciation and stock pay, and depreciation is where the landlord business bleeds: $2.47 billion in the quarter, up from $922 million a year earlier, as the data centers and chips bought over the past year start wearing out on the books. Over the first nine months of 2026, the AI segment still lost $3.3 billion at the operating level on $10.9 billion of revenue.
The company talks like a power company now. SpaceX said its "nameplate compute" — a term borrowed from power plants — reached 1.7 gigawatts on September 30, up from 1.4 in June and 0.6 a year ago, as it keeps building the Colossus II data center in Memphis. A rocket company that measures itself in gigawatts, like a utility. That single phrase tells you more about SpaceX's future than any render of a Mars colony.
The $16.4 billion asterisk
Now the part of the quarter that should keep you honest about that profit. Capital spending nearly tripled to $19.6 billion, and the AI business consumed $16.4 billion of it — more than twice what the AI segment brought in. Year to date, AI capex is roughly $40 billion against $10.9 billion of AI revenue. SpaceX says spending will stay near that level for the near future.
Read that twice. The company is spending money on AI infrastructure at more than three times the rate the AI business generates revenue. This is not a criticism, exactly — it is the bet. But it is worth saying plainly: a $4.4 billion profit that arrives while the AI unit is simultaneously told it "will need sustained investment before it can consistently generate profits" is a company profit, not an AI-business profit. The operating losses at the AI unit, Reuters reported, remain large. The quarter's triumph is real, but it is financed by the other two engines.
Musk told investors that Grok Bot — "our always-on agentic SI system," in the company's words — is the "fastest-growing product I've ever seen." It is worth noting, with a raised eyebrow, that the company gave no subscriber or user numbers for Grok at all. "Fastest-growing" without a base is the kind of line that works on an investor call and dies in a spreadsheet. Grok, Reuters noted, remains overshadowed by OpenAI, Anthropic, and Google. For now, the AI story at SpaceX is compute contracts, not the chatbot.
Starlink: still the engine room
Buried under the AI fireworks, the old business had a quietly excellent quarter. Starlink and the broader connectivity operations remain, in Reuters' words, the company's main financial engine. The subscriber base kept expanding across consumer, enterprise, aviation, maritime, and government services — with Starlink now installed on more than 3,800 aircraft, according to TESLARATI's read of the disclosures, and consumer subscribers at roughly 13.6 million, up 89% year over year, with consumer ARPU improving to $69.
Photo: A Starlink v2 user terminal. Starlink and the connectivity business remain SpaceX's financial engine — the profitable base the AI bet is built on. (Steve Jurvetson / Wikimedia Commons, CC BY 2.0 — file photo.)
But investors are watching the pricing question closely. SpaceX has said expanding beyond the wealthiest countries could push average revenue per user lower over time, and analysts are keen to know whether that pressure is starting to show up as the company enters more international markets with lower-priced plans. It is the classic scale-versus-margin tension, and it is the one number that could spoil future quarters: can the connectivity engine keep growing without getting cheaper per customer?
What this quarter actually proves — and what it doesn't
Let me take a position, because the polite read of this quarter is colossally wrong. The polite read is that SpaceX is now an AI company and the numbers prove it. The numbers prove something narrower and arguably more interesting: SpaceX has stumbled into one of the best businesses in tech almost by accident. The world is desperate for compute; SpaceX, thanks to the xAI integration and Musk's mania for building data centers, had compute to rent; and the rent is enormous. That is a fine business. It is also, to be blunt, not superintelligence. It is DigitalOcean with a rocket on the logo.
None of which means the bet is bad. The land-grab logic is real: 1.7 gigawatts today, Musk promising to more than double Colossus 2's Nvidia chips before year-end, and a management team willing to spend $16.4 billion in a single quarter on the thesis that compute demand keeps compounding. SpaceX's finance chief called it "a clear inflection in our Super Intelligence business" and promised discipline — the kind of line that sounds reassuring until you remember the $2.47 billion depreciation charge sitting right underneath it.
The honest summary: SpaceX's AI unit is growing at a pace that would make a hyperscaler blush, it is finally profitable at the operating level this quarter (barely, and with accounting help), and it is still being carried by the satellite business. Whether the "superintelligence" story justifies the spend is the open question — and it will not be answered by renting GPUs to Anthropic.
The week ahead
Two threads to watch. First, the long-rumored SpaceX–Tesla merger is back in the conversation — Barclays floated a December timeline just last week, and investors were listening for any hint on the call. Tesla reports its own results Wednesday. Second, Starship remains the company's most important unanswered question: Reuters notes it has yet to enter commercial service, and last month's first orbital test flight ended hours earlier than planned after an engine issue. The vehicle that is supposed to launch the higher-bandwidth Starlink satellites and, eventually, orbital AI data centers still has to prove it can fly reliably.
The space funding market, at least, is euphoric — space startups banked a record $20.8 billion in Q3 across the industry, and SpaceX's print is the exclamation point. And the AI-money arms race shows no sign of cooling: even as SpaceX rents out compute, OpenAI is racing to lock up $30 billion in its own funding round.
SpaceX is a rocket company that made its quarter as a landlord. Musk can call it superintelligence if he likes. The invoices say data-center rent.
Management is already selling the next chapter: CFO Bret Johnsen told the earnings call that fourth-quarter capex will run $35–40 billion and guided Q1 2027 revenue at $33–37 billion — against a ~$22.2 billion analyst consensus — citing visibility into its growing pipeline of compute contracts. On the other side of the bet, Gartner estimated last week that more than $3.5 trillion in AI infrastructure spending ‘could be exposed to impairment’ by end-2030. The Q1 guide is the number to watch: it says SpaceX expects the landlord invoices to keep growing faster than the sceptics think.
The Q3 numbers, answered: the questions everyone is asking
Why is SpaceX calling its AI business "Super Intelligence"?
SpaceX relabelled the segment in this quarter's disclosures, and the name is political: US President Donald Trump signed an executive order renaming artificial intelligence as "superintelligence" in federal usage, a move Musk has endorsed. The brand follows the White House, but the business underneath is still what the main piece describes — data-centre rent, not models.
Did Starship actually make money this quarter?
Yes — for the first time. Starship Flight 14 flew the company's first revenue-generating mission, deploying 26 Starlink V3 satellites. SpaceX said the flight added roughly ten times the data capacity of a single Falcon 9 Starlink launch. It is the clearest proof yet that Starship is becoming a business, not just an R&D expense.
How profitable is the AI business, really?
More profitable than last quarter, less than the headline suggests. The Super Intelligence segment turned a $431 million operating profit after a $1.26 billion operating loss last quarter and a $1.72 billion loss a year ago. But SpaceX ploughed $16.4 billion of its $19.6 billion quarterly capex into AI — roughly 85% of all investment — and the company still posted a net loss of $413 million for the first nine months of the year. Adjusted EBITDA of $8.8 billion is not free cash flow.
Sources: Reuters (Akash Sriram, Oct 20) — primary reporting on the Q3 results; SRN News (Reuters relay) for the investor-call quotes, per-share figures, and capex detail; TESLARATI on Grok Bot comments, Starlink subscriber figures, and space-segment revenue; madrobot.blog for Super Intelligence segment operating detail (operating profit, depreciation, nameplate compute, Cursor-deal stock pay); OfficeChai/TradingView for IPO history context. Figures are company-reported and estimate comparisons come from LSEG/IBES data as reported by Reuters.
Comments
Post a Comment