SpaceX Q3: AI revenue hits $7.56B, 52% of the company
SpaceX's AI business stopped being a side bet this quarter. In the same 24 hours, the AI buildout's least technical bottleneck — the power grid — took about 40% off an Australian data centre operator's market value.
SpaceX's "Super Intelligence" unit brought in $7.56 billion in the third quarter, up 830% year over year and now 51.9% of total revenue — the first time AI has out-earned Starlink and rockets combined. Group revenue was $14.58 billion, up 161%, and GAAP net income was $4.4 billion against a $1.05 billion loss a year earlier, the Financial Times reported. Adjusted EBITDA reached $8.81 billion, up 306%, and operating income was $2.79 billion. The other two segments are still growing, just slower: connectivity brought in $5.25 billion, up 69%, and the space launch business $1.77 billion, up 6%.
The AI number is not one product. SpaceX attributed the jump to newly activated cloud service agreements, the Cursor acquisition closing into the quarter, and Grok adoption in both enterprise and consumer tiers. Cursor closed on August 14 at a $60 billion implied equity valuation and a $42.5 billion GAAP purchase price, and contributed $608 million of revenue from the closing date to September 30. Chief financial officer Bret Johnsen told analysts on the call that even stripping Cursor out, super-intelligence revenue grew 192% sequentially — Yahoo Finance carried the call quotes.
What is behind that growth is compute, and the compute bill is the part worth reading twice. Capital expenditure was $19.59 billion in the quarter, $16.44 billion of it inside the AI segment. Nominal capacity stood at 1.7 gigawatts on September 30, up from 1.4 gigawatts at the end of June and 0.6 gigawatts a year earlier. Musk told the call he expects to exit 2026 at 2.5 to 3 gigawatts and 13 gigawatts cumulatively by the end of 2027; the company guided fourth-quarter capex to $35–40 billion. It also holds $51.3 billion in cash plus $38.6 billion in marketable securities, and says new capacity pays back in under a year — which is the entire argument for spending at this rate.
The line that actually moved the stock was the revenue target. Musk put annual recurring revenue at $500 billion to $1 trillion by the end of 2027, well above the $100 billion ARR goal the company still expects to hit by December. Shares rose in after-hours trading from a $1.95 trillion close. He also addressed the Tesla question directly, calling a merger "a good idea" and confirming both boards have formed special committees of independent directors — while cautioning that treating the deal as certain "would be overstating it," since Tesla needs 50% of shareholders to vote yes.
Strip out the enthusiasm and the quarter has an obvious shape: the AI unit is now the company. That is a strategic win and a financing problem at the same time. SpaceX earned $8.81 billion in adjusted EBITDA and spent $19.59 billion in capex, with the next quarter guided to roughly double the spend. Those are not numbers that fund themselves from operations, which is why the company has been shopping the debt markets hard — we covered that in September, when it emerged SpaceX was seeking $40 billion led by Apollo to buy Nvidia chips. The profit beat is the headline; the funding gap is the story. Anyone valuing SpaceX as a rocket company with an AI hobby is now valuing it as an AI neocloud with a launch pad, and neoclouds are priced on capacity utilisation, not on earnings.
In Melbourne, a data centre could not get its power, and NextDC lost about 40% of its market value — as much as 44% in early trade — after cutting fiscal 2027 guidance. Reuters reported the operator as the biggest loser on the ASX 200, with roughly A$3.1 billion of market value gone and the shares at their lowest level since January. Revenue guidance fell to A$475–500 million from A$615–640 million; underlying EBITDA fell to A$245–265 million from A$385–410 million — a cut of about 36% at the midpoint.
The cause is not demand. Electricity distributor Jemena suspended infrastructure upgrades at NextDC's M3 Melbourne campus after safety concerns, restricted site access and the absence of council consent pending further community consultation, and does not expect to resume work before February 2027. NextDC says about A$140 million of fiscal 2027 revenue and EBITDA is deferred rather than lost, with the added capacity now expected in fiscal 2028. The company cut capital expenditure to A$4.5–5.0 billion from A$5.25–5.75 billion, put all new developments and land purchases on hold, and will decide on the next M3 stages only after Victoria's November 28 state election. RBC analyst Tom Beadle told Reuters that community concerns around data centre developments "present a real risk to the timing of delivery of contracted capacity" — which is the polite version of what just happened to a company with signed customers and no working substation.
Both stories are about the same constraint arriving from different directions. SpaceX is racing to add gigawatts and treating power, cooling and siting as the bottleneck rather than chips; NextDC just discovered that in a dense city, the bottleneck is a distributor, a council and a room full of neighbours. The AI trade spent two years pricing compute as a semiconductor problem. This week it got priced as a land-use problem, and Reuters noted the selloff landed on a market already soured on AI infrastructure after another operator withdrew its data centre IPO.
What to watch: SpaceX's fourth-quarter capex print against that $35–40 billion guide, and whether NextDC's pause becomes a template — several listed operators in Europe and Asia are sitting on the same combination of contracted capacity and contested grid connections.
Should a data centre operator disclose community-consent risk the way it discloses customer concentration? Write to us via our contact page.
