SpaceX Guides Q1 2027 Revenue to $33 Billion to $37 Billion as AI Compute Overtakes Rockets and Broadband as the Growth Engine
Third-quarter 2026 earnings call, October 20, 2026: Management lays out a 13 gigawatt compute target, a chip fab timeline and a Tesla merger caveat
SpaceX reported third-quarter revenue of $14.6 billion, up 161% year over year, and adjusted EBITDA of $8.8 billion, up 306%. Net income was $4.4 billion, against a loss of $1.1 billion a year ago. The more consequential news was the forward guide. CFO Bret Johnsen said the company now sees Q1 2027 revenue of $33 billion to $37 billion, more than double the quarter just reported. The call confirmed that SpaceX is now primarily a compute and connectivity business that also happens to own the world's dominant launch franchise.
Super Intelligence is now the largest segment, and the economics are unusually rich
The Super Intelligence segment generated $7.6 billion of revenue, up $5 billion sequentially, and passed Connectivity at $5.2 billion and Space at $1.8 billion. Segment adjusted EBITDA was $5 billion, up $3.8 billion from the prior quarter, driven by what Johnsen called higher-margin cloud service agreements. Solutions and Infrastructure revenue was $7 billion, including $608 million from Cursor, which closed roughly mid-quarter. Excluding Cursor, that line grew 192% sequentially on the activation of new cloud contracts.
The pricing disclosure is the key data point for investors. "At current pricing, new compute deployments are generating a payback of less than a year," Johnsen said, adding that pricing is tracking slightly above the high end of the $30 to $50 per watt range discussed last quarter. Elon Musk went further, saying "it's not impossible that we will have a six-month payback period for compute next year." He described demand as intense: "We are also, frankly, hounded relentlessly by every hyperscaler and by dozens of smaller companies to rent more of our compute." He illustrated the scarcity with a Monty Python cheese shop analogy, saying SpaceX's attempts to offload inference to other cloud providers met with GB300s, GB200s and TPUs all unavailable.
Musk's commercial stance is notable. SpaceX intends to rent compute at roughly spot rates rather than lock in long-term deals, aiming for an approximate 50-50 split between internal use and third-party rental, with the mix shifting month to month. His reasoning is that "the value of compute per watt will increase over time, not decrease," because better models extract more useful intelligence per watt. He cited H100s renting for more than when they were new, and said short-term contracts are renewing at higher dollars per watt. This is a strong bet on sustained scarcity, and it leaves revenue exposed if the supply shortage eases faster than management expects.
The build-out: 13 gigawatts by the end of 2027, with a stretch goal well above that
Nameplate compute capacity was 1.7 gigawatts at the end of Q3. Management expects nearly 2.2 gigawatts by the end of October and 2.5 to 3 gigawatts by year-end. Musk said he is "highly confident" of adding an incremental 10 gigawatts in 2027, for roughly 13 gigawatts in total. SpaceX is also planning power, cooling and infrastructure for an incremental 20 gigawatts, so that power does not constrain GPU deployment. He acknowledged the gap between plan and outcome: "I don't expect everything to go right," though he said a total above 20 gigawatts online in 2027 "used to be zero, it's not zero anymore."
The most concrete competitive argument was about turbine components. SpaceX has built advanced casting capability for rocket engines and is applying it to turbine blades and vanes. Musk said turbine suppliers told him those parts trail the rest of the turbine by 12 to 18 months. "That means that we can bring power online 12 to 18 months faster than anyone else," he said. He declined to describe the full power plan, calling it proprietary. Musk also said Colossus II in Memphis is on track to bring online roughly 200,000 GB300s per month through year-end. In the Q&A he said 20,000 GB300s per month, an inconsistency management did not reconcile on the call. A 440,000-unit Vera Rubin training cluster is planned for the end of Q2 2027.
The financial cost is large. Capital expenditures were $19.6 billion in Q3 and are guided to $35 billion to $40 billion in Q4. Johnsen's funding answer was qualitative: operating cash flow, multiple pools of debt capital supported by the company's investment-grade status, and partnerships with vendors and customers across the value chain. He gave no financing mix or total capital requirement for a 13 gigawatt build. SpaceX also guided Q4 gross margin back toward Q2 levels, after Q3 benefited from one-time items and customer mix.
The $500 billion to $1 trillion ARR claim
Musk said the compute trajectory implies annual recurring revenue of "somewhere between $500 billion and $1 trillion" by the end of 2027. The company's formal near-term target is $100 billion of ARR by the end of this year, defined as December revenue annualized. The Q1 guide of $33 billion to $37 billion implies an annualized run rate of about $130 billion to $150 billion, so the near-term path is at least consistent with the nearer target. The 2027 figure is a different matter. It is a stated aspiration, built on an assumed Grok bot ramp and a compute build that has not yet been financed or powered. Investors should treat it as a statement of ambition rather than guidance. Backlog of $63 billion, with about $42 billion expected to be recognized within a year, provides firmer support for the next several quarters.
Grok bot and Cursor: early, fast, and still small
Grok bot, the agentic assistant launched in August, had 800,000 paid weekly users earlier this month, more than doubling in a month. B2B is the fastest-growing category. Musk called it "the fastest growth of a product that I've ever seen in my life" and said per-user usage is rising over time. Enterprise customers with more than $1 million in run-rate revenue, including Cursor, rose 50% sequentially on a pro forma basis. SpaceX also announced a bundle of Grok, Grok bot, Cursor and X Premium on one plan. Advertising revenue grew 19%.
The Cursor deal carries a real accounting cost. Of the $60 billion equity value, $23 billion was attributed to retention and replacement awards, recognized as stock-based compensation over roughly three years. The segment's $431 million of operating income would have been $1.6 billion excluding $1.2 billion of Cursor-related stock compensation in the quarter. Because Q3 included only about half a quarter of ownership, the full-quarter charge will likely be larger in coming periods, a point the call did not address.
Starship: orbit achieved, double catch next
Flight 14 reached orbit for the first time and deployed operational Starlink V3 satellites. Musk said the roughly half-loaded vehicle delivered 26 terabits per second of bandwidth, equivalent to 10 Falcon 9 launches. Flight 15, expected next month, will attempt to catch both the booster and the ship. Musk put success at "more likely than not" and said Gwynne Shotwell expects the FAA license to cover return-to-launch-site and tower catch for both stages. Management targets a second-stage reflight and an on-orbit refueling attempt in early 2027, and Mobile V2 and Starmind satellites flying in 2027.
Cadence targets are aggressive. Musk expects weekly Starship flights in the first half of 2027 and twice weekly by the end of next year. He described the Louisiana spaceport as eventually hosting 20 or more launch towers. For Louisiana he gave an internal goal of a first launch in 2028 and said at-scale launching in 2029 is "highly likely." Gwynne Shotwell said that a few hundred V3 satellites are needed before users perceive a network improvement, which she expects by mid-2027.
Starlink: strong subscriber growth, with a one-time margin assist
Connectivity revenue was $5.2 billion, up 69% year over year and 22% sequentially. Consumer subscribers rose by 1.6 million to 13.6 million, and monthly ARPU rose 5% sequentially to $69 on favorable mix. Enterprise and government revenue grew 91% year over year. Gross margin was 55%, including roughly 3 points from one-time tariff refunds. Excluding that, adjusted EBITDA margin expanded about 2 points year over year. Starlink is contracted for roughly 12,000 aircraft, almost a quarter of the available fleet, and aviation production has quadrupled since the start of the year, with another doubling expected by year-end. Starlink Mobile exceeded 22 million monthly users, up 37% sequentially. Service using the newly acquired low-band spectrum and approved 15,000 Mobile V2 satellites is targeted for early 2028.
Musk's longer-term claims are large. He expects about 100x the current constellation's bandwidth within three years, from roughly 100,000 satellites. He sees demand for 1,000x more bandwidth in three to five years as AI agents proliferate, and "a clear path to several hundred billion dollars a year in connectivity revenue." He deferred the supporting detail to a future presentation. In response to a question on whether the satellite layer can match terrestrial incumbents, Shotwell said that 85% to 93% of mobile traffic runs over Wi-Fi and that S-band spectrum would carry the majority of satellite traffic, with low-band spectrum providing building penetration.
Space segment revenue of $1.8 billion was up 84% sequentially but only 6% year over year, which shows how much of the company's growth now sits outside launch. SpaceX flew 35 launches in the quarter and won three more Dragon crew missions, 18 national security launches and a lunar payload contract on Starship.
Terafab and the Tesla question
Asked by Morgan Stanley's Adam Jonas whether SpaceX has ordered EUV machines, Musk did not confirm. He gave internal targets with "big error bars": a functional test SRAM chip at the Austin research fab around May or June 2027, and volume production at the Gibbons Creek Terafab around Q3 2028, with Q3 or Q4 2028 as the plausible range. He said SpaceX has "significant potential breakthroughs in chip fabrication" but declined to describe them. The stated ambition is a terawatt of chips per year. A volume start inside roughly two years for a new fab operator would be unprecedented.
On a possible Tesla combination, Musk said both companies have formed special committees of independent directors, from which he is recused as a conflicted party. The committees have "not reached agreement on the terms of any transaction." He cautioned against assuming a deal is certain and noted that Tesla would need approval from 50% of all outstanding shares, "quite a high hurdle." He added that he thinks a combination "would be a good idea," citing overlap in AI software, chip design, Terafab, robotics and connectivity.