On 21 October 2026, SpaceX told investors that its artificial intelligence and cloud compute business, the Super Intelligence segment, had out-earned Starlink for the first time. The numbers: US$7.56 billion of quarterly revenue against US$5.248 billion for Connectivity, the segment that carries Starlink. Every outlet that covered the result led with that line. Almost none of them printed the number that qualifies it. Super Intelligence's operating income for the same quarter was US$431 million. Connectivity earned US$2.367 billion: more than five times as much, on revenue equal to about two-thirds of Super Intelligence's own total.
Five days later, on 26 October, SpaceX signed what it describes as the largest committed GPU financing arrangement on record. The arrangement is roughly US$40 billion, structured as a sale-and-leaseback with an entity managed by Apollo Global Management. Apollo buys Nvidia's next-generation Vera Rubin chips and leases them back to SpaceX for around five years a tranche. Apollo's own description of the money is one word: terrestrial. The compute SpaceX is spending forty billion dollars to secure sits on the ground, not in orbit.
SpaceX's artificial intelligence revenue overtook its connectivity revenue this quarter. That does not make it a pivot. The forty billion dollars went where the earnings call did not spend much time: down, not up.
What happened
SpaceX's third quarter 2026 result, taken whole, was strong by any ordinary measure. Total revenue reached US$14.575 billion, up 161 per cent on the same quarter a year earlier. Net income was US$4.4 billion, reversing a US$1.1 billion loss twelve months before. Adjusted earnings before interest, tax, depreciation and amortisation rose 306 per cent to US$8.809 billion. Quarterly capital expenditure reached approximately US$19.6 billion, with roughly US$16.4 billion of that directed at the Super Intelligence segment alone. [1][2]
The result splits across three reported segments. Super Intelligence, the AI and cloud compute business, produced US$7.56 billion of revenue, about 52 per cent of the total, and US$431 million of operating income. Connectivity, which carries Starlink's 13.6 million consumer subscribers, produced US$5.248 billion of revenue and US$2.367 billion of operating income. A third segment, Space, covering launch operations, produced US$1.767 billion, up a modest 6 per cent year on year, and went almost unmentioned in this quarter's coverage. [2] SpaceX attributes Super Intelligence's growth to new cloud-services agreements, the acquisition of Cursor, and wider adoption of Grok among enterprise and consumer customers. [1]
Then came the financing. On 26 October, SpaceX and the Apollo-managed entity signed a binding sale-and-leaseback commitment. Apollo buys the hardware; SpaceX leases it back for terms of roughly five to five and a quarter years a tranche. The deal is expected to close by the end of 2026, with leases beginning in the first half of 2027. SpaceX's chief financial officer, Bret Johnsen, is quoted citing payback periods of less than one year on the infrastructure this buys. [3][4]
The number that qualifies the headline
Read the two figures from the earnings call next to each other. A different story appears than the one the headlines carried. Super Intelligence produced 52 per cent of SpaceX's revenue. It produced roughly 15 per cent of the combined operating income its two largest segments produced between them. Connectivity earned about five and a half times as much from revenue equal to roughly two-thirds of Super Intelligence's own, not two-thirds of total company revenue. The two readings are easy to blur, and only one of them is arithmetically correct. "SpaceX's AI business has overtaken Starlink" is true of the top line and false of the bottom one. Any comparison that treats the crossover as evidence of a strategic pivot needs both figures in the same sentence to mean anything. Super Intelligence's adjusted earnings figure, inflated by US$1.2 billion of share-based compensation tied to the Cursor acquisition, widens that gap further once the accounting adjustment is set aside.
In Space Mafia I wrote about this exact migration before this quarter's numbers existed to test it. SpaceX's third-generation Starlink satellites carry their own compute. FCC filings and investor briefings suggested more than two thousand of them would be capable of commercial AI inference and training by December 2025. Elon Musk, quoted in Space Mafia, put the ambition plainly: "Starlink is no longer just connectivity; it is compute at the edge of the atmosphere. The economics are too compelling to ignore." That was the orbital version of the pivot: compute migrating off the ground and into the constellation itself.
This month supplied a second version of the same ambition, often paired with the Apollo financing as if the two were evidence of the same trend. At the International Astronautical Congress in early October, SpaceX unveiled "Starmind," a visualisation of a proposed orbital data-centre constellation of roughly one million satellites, about a hundred times the size of the current Starlink fleet. SpaceX's own senior vice president for Starlink engineering, Michael Nicolls, described it on the record: "a very notional concept, and thus, it's probably wrong, and will change a lot." [5]
One of those two things has a signed five-year lease behind it. The other is a video. The forty billion dollars went to the thing with the lease. Apollo's own language about what it is financing settles the question without needing to be interpreted: a terrestrial compute buildout. [3][4] SpaceX added a lower-margin line of business on top of a higher-margin one, and financed the infrastructure for it on the ground, where power, cooling and fibre already exist, rather than wait on an orbital data centre that its own engineering leadership calls probably wrong. This is a different test than the regulatory-escape reading this series usually gives the Kardashev Pivot, where commercial operators treat Earth's jurisdictions as something to out-run. Does a revenue crossover between two business lines count as a pivot at all, when the newer line earns a sixth as much per dollar as the one it is supposedly succeeding? On the evidence of one quarter, it does not. The operating-income split may look different by next quarter, when the same crossover is reported again.
The economics behind that choice are not exotic. A Vera Rubin-class GPU cluster needs continuous high-density power, active cooling and fibre bandwidth measured in terabits. All of that exists today at terrestrial data-centre sites. None of it exists yet in orbit at any comparable scale. Thermal rejection in a vacuum and radiation-hardened electronics are engineering constraints Space Mafia catalogued in detail before this year's financing round existed to confirm them. A sale-and-leaseback on ground-based hardware is a bet that can be revised next year, when the next GPU generation ships. A constellation of a million satellites is not. The chief financial officer's own payback-period framing only works because the hardware sits somewhere its use can be measured and replaced on an ordinary commercial clock, not a launch manifest.
There is a second finding worth naming for what it says about how this story travelled. CNBC reported preliminary financing talks on 7 October. [6] Bloomberg, TechCrunch and CNBC itself then reported the signed, binding commitment on 26 October: a different event, nineteen days later. [3][7] Treating the two as one makes the financing look like it moved faster than it did, which matters in an industry where the gap between an early conversation and a signed commitment is usually the whole story. This is a clean instance of the evidentiary habit V.E.R.A. Saturday examines in detail: two real, dated events from the same outlet, reported nineteen days apart, where the earlier one's date attached to the later one's claim produces a citation that sounds right and is not. A visualisation, a set of early talks and a signed lease are three different categories of fact, and this month produced one of each.
What this means for New Zealand organisations
New Zealand's own anchor in this week's cycle sits a long way from an earnings call. On 23 October, Rocket Lab's Electron lifted off from Launch Complex 1 on the Māhia Peninsula at 8:11pm. It carried three HAWK satellites to a 520 kilometre orbit for the operator HawkEye 360. It was the 98th Electron launch overall, the 19th of 2026, and the third dedicated HawkEye 360 mission, completing a fifteen-satellite constellation. HawkEye 360 states its own purposes for that constellation directly: maritime tracking, detection of GNSS jamming, radar monitoring, communications mapping and space defence. Those are the operator's own stated terms, reportable as such. [8][9]
The Federal Communications Commission was scheduled to vote on 29 October on expanding the Supplemental Coverage from Space framework by a further 482 megahertz. A separate vote covered an auction of dedicated mid-band spectrum for direct-to-device satellite services. Three major American carriers jointly petitioned to strip part of the proposal on technical grounds. The outcome was not available at the time of writing. [10]
For organisations working through their obligations under the Government Bodies and Space Infrastructure Act, none of this week's findings changes the compliance picture directly. The Act's transitional period closed on 29 July 2026. Its mechanics are unchanged by a United States earnings report or a GPU financing deal: who counts as a space operator, what the notification thresholds are, and what the Hidden Space Operator category captures. This week changes the shape of the dependency New Zealand organisations are managing. SpaceX's own disclosure shows the connectivity business carrying the launch and AI businesses in profitability terms. Connectivity is the segment New Zealand's most exposed sectors actually rely on day to day. A board assessing exposure to a single satellite operator should ask which segment of that operator's business is funding the one it depends on. The question is not only how large the operator's total balance sheet has grown.
Ask four questions before the next operator update lands, not after:
- Which segment of the operator's business is actually profitable, not just large.
- Is the infrastructure behind a claim a signed commitment or a visualisation.
- Where does the hardware physically sit, and under whose jurisdiction.
- What happens to the service if the financing structure, not the hardware itself, changes hands.
New Zealand's operational dependencies in this domain have not moved since they were last checked. Close to 99 per cent of the country's international data traffic still transits a small number of subsea cables landing at Auckland. There is no large-country alternative-route redundancy. SouthPAN, the joint Australia and New Zealand satellite-based GNSS augmentation system, is led on the New Zealand side by Toitū Te Whenua LINZ. It depends on contracted satellite transponder capacity and is due to move to a new satellite from 2027. A dedicated AWS Asia Pacific (New Zealand) region has operated in-country since September 2025. It narrows the offshore cloud concentration this series has flagged before.
What to watch
Two threads carry into November. First, the FCC's spectrum decision will shape how much additional capacity SpaceX's connectivity business can draw on for direct-to-device services, whichever way it falls. That business is the one actually funding the infrastructure its AI arm depends on. Second, Rocket Lab's own manifest puts Electron two missions from its hundredth flight, a count that cannot slip the way a launch date can. [8]
Reported growth of 830 per cent year on year suggests SpaceX's AI segment will keep growing. Watch whether the operating-income gap narrows as that segment matures, and whether the next capital commitment of this size, when it comes, is described as serving orbit or the ground. Apollo chose its word carefully this time. The next financier may not have to.
Three numbers matter out of this week, not seven. Fifty-two per cent of SpaceX's revenue came from artificial intelligence this quarter. Roughly fifteen per cent of its segment profit did. And forty billion dollars just went to hardware sitting on the ground, financed by a lender who said so in writing. None of those numbers need the word pivot. None of them depend on a constellation that does not exist yet.
The ground-based infrastructure this financing buys is not entirely proprietary. Hyperscale data centres, including the kind this debt will fill with Vera Rubin-class GPUs, increasingly follow open hardware standards published through the Open Compute Project. Those are shared specifications for racks, power delivery and cooling that let operators source hardware from more than one vendor rather than lock into a single integrated design. The project began inside one hyperscaler and now counts member organisations across cloud providers, chip makers and network operators internationally. It will not appear in SpaceX's financing filings, because an open hardware standard is a supply-chain detail rather than a disclosure requirement. It is part of why a ground-based GPU fleet can be swapped and replaced in a way an orbital constellation of custom satellites cannot.
The same ground-versus-orbit question recurs on the defence side of this industry. Starshield, the hardened military variant of the Starlink constellation, is used under contract by the United States Space Force and allied partners. It sits on the same corporate infrastructure whose AI segment this quarter out-earned its connectivity segment on paper, while depending on that connectivity segment for most of its profit. Allied users tasking commercial satellite imagery, including the kind HawkEye 360 collects from Māhia, increasingly rely on AI-assisted analysis pipelines built on infrastructure that neither they nor their suppliers directly control. A capability built on a single commercial provider's compute is a sovereignty question independent of whether the provider's AI business is growing. It is financed through arrangements structured for that provider's own balance sheet rather than for allied continuity of service.
If your organisation depends on a satellite operator for anything, from imagery to connectivity to the sensors it carries, when did you last check which part of that operator's business is actually paying the bills?
This is one of seven weekly series in The Hamberger Report. Subscribe on LinkedIn and the next one arrives in your feed.
The views expressed in this article are entirely my own, informed by more than 30 years of professional experience in architecture, security, and technology leadership in New Zealand. I write as director of Te Pono Limited; the views are personal and do not represent the position of any client, any government agency, or the New Zealand government. My commentary on legislation and policy is analytical, drawing on publicly available sources and my professional expertise in architecture, security, and AI governance, and it is politically neutral.
Andreas Hamberger is a New Zealand leader in Architecture & Security and Associate Member of the Institute of Directors. Space Mafia examines the sovereignty implications of orbital compute infrastructure.
This article was produced with AI assistance under my direction. Research, drafting and images pass through a pipeline I built and govern: automated gates for source verification, forbidden language and political neutrality, and my own review before anything is published. The tools include Claude, Gemini and Openart. The frameworks, arguments and editorial judgements are mine and are the same discipline I apply to the AI systems I audit for clients. AI accelerated the work; the thinking, and the responsibility for it, are mine.
[1] itechguides.com. "SpaceX Had a Monster Quarter in Q3 2026, and Starlink Was Not the Whole Story." 21 October 2026. https://www.itechguides.com/spacex-had-a-monster-quarter-in-q3-2026-and-starlink-was-not-the-whole-story/
[2] StockTitan. "SpaceX Q3 2026 Earnings: AI Revenue Passes Starlink." 21 October 2026. https://www.stocktitan.net/articles/spacex-q3-2026-earnings-ai-revenue-passes-starlink
[3] TechCrunch. "SpaceX Locks in $40B GPU Financing Deal With Apollo." 26 October 2026. https://techcrunch.com/2026/10/26/spacex-locks-in-40b-gpu-financing-deal-with-apollo/
[4] Bloomberg. "SpaceX Agrees to $40 Billion Chip Financing Deal With Apollo." 26 October 2026. https://www.bloomberg.com/news/articles/2026-10-26/spacex-agrees-to-40-billion-chip-financing-deal-with-apollo
[5] Yahoo Tech. "A Bigger Starlink: SpaceX Provides a Look at 'Starmind.'" October 2026. https://tech.yahoo.com/science/articles/bigger-starlink-spacex-provides-look-212039193.html
[6] CNBC. "SpaceX, Nvidia Chips, and Apollo: Financing Talks Underway." 7 October 2026. https://www.cnbc.com/2026/10/07/spacex-nvidia-chips-apollo-financing.html
[7] CNBC. "SpaceX Closes $40 Billion Debt Deal for AI Chips." 26 October 2026. https://www.cnbc.com/2026/10/26/spacex-closes-40-billion-debt-deal-for-ai-chips-elon-musk-.html
[8] Quiver Quantitative. "Rocket Lab Launches Third Dedicated HawkEye 360 Mission on Record 98th Electron Flight." 23 October 2026. https://www.quiverquant.com/news/Rocket+Lab+Launches+Third+Dedicated+HawkEye+360+Mission+on+Record+98th+Electron+Flight
[9] Rocket Lab. "HawkEye 360 Mission." Mission page. https://rocketlabcorp.com/missions/launches/hawkeye-360/
[10] TechTimes. "AT&T, T-Mobile, Verizon Demand FCC Strip 290 MHz From Satellite Vote Before Wednesday." 27 October 2026. https://www.techtimes.com/articles/329536/20261027/t-t-mobile-verizon-demand-fcc-strip-290-mhz-satellite-vote-before-wednesday.htm