As we read SpaceX’s filings, its AI compute business has three layers: data centers on the ground, capacity rented to Anthropic, and a plan to put compute in orbit. Separately, it has agreed to buy spectrum licences. In the third quarter, its AI segment, SI, brought in about 52% of revenue and took about 84% of capital expenditures, by our arithmetic on its 10-Q.
This explainer sets out each layer with the figures SpaceX has filed and Musk has said, and what each one depends on.
Business Pill · Vertical Integration
An evergreen Business Engineer explainer on vertical integration. It teaches the general idea only and does not discuss SpaceX.
The key insight: SpaceX’s AI segment brought in about 52% of third-quarter revenue and took about 84% of capital expenditures, by our arithmetic on its 10-Q; as we read its filings, the AI business has three layers: 1.7 GW of installed ground compute, capacity rented to Anthropic and a stated orbit goal of 100 GW a year.
SpaceX’s 10-Q for the quarter to 30 September reports nameplate compute of 1.7 GW, against 0.6 GW a year earlier. It defines nameplate compute as “the number of GPUs installed in our data centers at the end of the period multiplied by their respective all-in power”.
On the 20 October call, Elon Musk said: “I would be surprised if we brought on less than 13 GW of compute by the end of next year”. That is about 7.6 times the installed base at 30 September, by our arithmetic. He named “power, cooling, and electrical equipment infrastructure” as “somewhat of a limiting factor”.
The prospectus lists its data centers in the Greater Memphis Area and Southaven, Mississippi. Third-quarter SI capital expenditures were $16,444 million of $19,593 million across segments, about 84%, by our arithmetic.
SpaceX compute in gigawatts on a log scale: installed nameplate compute (10-Q), Musk’s 13 GW expectation for the end of 2027 (20 October call), and the prospectus goals for compute launched to space each year. Targets are not results; 7.6x is our arithmetic.
What Was Posted on X
.@dwarkesh_sp: โSo, data centers in space is really a regulatory play?โ
@elonmusk: โIt’s harder to scale on ground than in space.โ
โIn 36 months (but probably closer to 30 months), the most economically compelling place to put AI will be space.โ
A clip reposted at 16:03Z on 4 November from Musk’s interview with Dwarkesh Patel, published on 5 February 2026, in which Musk said it is harder to scale on the ground than in space.
Bloomberg’s post at 15:50Z on 4 November linking a newsletter on Texas deals with Musk and SpaceX. We have not read the newsletter.
Update, 4 Nov, 22:03Z: Goldman and Apollo begin marketing portions of SpaceX’s roughly $40 billion GPU financing, Bloomberg reports. Goldman Sachs and Apollo Global Management are in talks with Wall Street investors to start selling portions of a roughly $40 billion chip financing deal for SpaceX, Bloomberg reported at 21:10 UTC on 4 November, citing people familiar with the matter. According to Wall St Engine’s summary of the report, an Apollo-backed entity would buy approximately $47.3 billion of Nvidia GPUs and related equipment with roughly $40 billion of debt and a $7 billion lease prepayment from SpaceX. The $7 billion is about 15% of the equipment cost and about 43% of the $16,444 million of SI capital expenditure SpaceX reported for the third quarter, by our arithmetic. We set out the reported terms and SpaceX’s own 8-K in SpaceX to prepay $7B in $47.3B Nvidia chip deal, sources say; the deal sits in our AI debt financing tracker.
The Rental Business
SI revenue was $7,560 million of $14,575 million in the third quarter, about 52%, by our arithmetic. Its Solutions & Infrastructure line was $7,022 million, against $359 million a year earlier.
The prospectus names one such customer, Anthropic, which it says “has agreed to pay us $1.25 billion per month through May 2029” for compute that “includes approximately 325,000 NVIDIA GPUs”. That is about $5.27 per GPU-hour all-in, by our arithmetic, as we reported.
SpaceX says the structure lets it “monetize a portion of the compute capacity in our infrastructure, while still permitting reallocation of that capacity for our own internal initiatives if needed in the future”. Anthropic’s amended S-1 expects about $3.1 billion a month once all capacity is delivered, about 2.5 times the prospectus fee by our arithmetic, as we covered.
The Orbit Plan
On 30 January, SpaceX asked the FCC for authority to launch up to one million satellites as the “SpaceX Orbital Data Center system”, about 90 times its more than 11,100 Starlink satellites, by our arithmetic. The prospectus says: “We expect to begin deploying our orbital AI compute satellites as early as 2028”.
It sets the goal: “Our goal over time is to launch 100 gigawatts of compute to space each year”. It also names a lower bar, about 10 gigawatts a year, “which we believe would still enable a commercially attractive AI compute business with strong economic returns”.
Update, 4 Nov, 22:22Z: Google’s energy chief, now named, prices orbit at up to 100 times Earth. Amanda Peterson Corio, Google’s head of data center energy, said data centers in space could cost up to 100 times as much as those on Earth, The Information reported on 1 November. “Let’s explore all the terrestrial options before we go extraterrestrial,” she said at The Information’s WTF summit, the outlet reports. Re-sharing the storyon X at 22:00 UTC on 4 November, The Information wrote that SpaceX’s orbital data center ambitions are helping support its roughly $1.8 trillion valuation, and that Corio says the economics are far from viable today, “with enormous launch costs and maintenance challenges.”
A clip reposted on X on 4 November comes from Musk’s interview with Dwarkesh Patel, published on 5 February 2026. Patel suggested that space was “really a regulatory play”. Musk answered: “It’s harder to scale on the ground than it is to scale in space”.
He added that “any given solar panel can do about five times more power in space than on the ground” and predicted space “will be by far the cheapest place to put AI”, within “36 months or less”. Thirty-six months from that interview is February 2029, by our arithmetic.
The cost side is in our first piece: by our arithmetic on a Google research paper’s estimates, launching a kilowatt of satellite power at $3,600 a kilogram costs about $14,700 a year, 4.9 times the top of the US data center power-spend range it cites.
Separately: Spectrum
The 10-Q says the total consideration for EchoStar’s spectrum licences “is approximately $19.6 billion”. The licences were transferred to a trust, “where they will remain until the Spectrum Acquisition Closing”, and SpaceX had paid $856 million to the trust by 30 September. In a separate announcement on 4 November, Qualcomm said SpaceX selected its platforms for Starlink Mobile’s US terrestrial 5G network, and that “SpaceX plans for U.S. nationwide deployment rollout to begin in 2027, with tens of thousands of sites deployed over the multi-year agreement”.
The Texas Question
AP called the Texas governor race for Democrat Gina Hinojosa at 04:17Z on 4 November. Bloomberg posted on X that “As Democrats take control in Texas, deals that Elon Musk and SpaceX had struck with the state could be threatened”, linking a newsletter we have not read.
SpaceX’s data centers are listed in Tennessee and Mississippi; its Texas sites are Starbase, McGregor and Bastrop, where it expects to add the production of AI compute satellites. Techmeme summarised Politico as reporting that Musk spent $446M on the midterms, and FEC filings show he gave America PAC $216.0 million this cycle to 14 October.
As we read the filings, the ground layer is where the money is today: installed GPUs earn rental and modelrevenue now, while the orbit layer is a stated goal with a first deployment the prospectus puts as early as 2028.
As we read the prospectus, renting capacity and keeping the right to reallocate it lets the same installed GPUs serve paying customers and SpaceX’s own models.
SpaceX IPO prospectus, 12 June 2026
“Our goal over time is to launch 100 gigawatts of compute to space each year”
Three Implications
FOR AI LABS RENTING COMPUTE The prospectus says Anthropic agreed to pay $1.25 billion a month for compute that includes approximately 325,000 Nvidia GPUs, about $5.27 per GPU-hour all-in, by our arithmetic.
FOR THE GROUND BUILD Musk said on the 20 October call he would be surprised by less than 13 GW of compute by the end of next year, about 7.6 times the 1.7 GW installed at 30 September, by our arithmetic.
FOR THE SPECTRUM DEAL The 10-Q puts the total consideration for EchoStar’s spectrum licences at about $19.6 billion; the licences sit in a trust until the acquisition closes.
The framework says: “Companies that solve the power equation will dominate the AI era; those that don’t will watch their chips sit idle.”
As we read it, SpaceX’s filings describe two answers to that power equation: grid-connected data centers in Tennessee and Mississippi, and solar-powered compute in orbit. The framework predates these filings and does not discuss SpaceX.
The 13 GW figure is Musk’s spoken expectation and the orbit figures are stated goals; neither is a result. SpaceX has not reported an orbital compute satellite in service. Anthropic’s agreements can end on 90 days’ notice after the initial period. We have not read the Bloomberg newsletter or the SemiAnalysis report on SpaceX’s build, and we make no forecast about launch costs, Texas policy or SpaceX’s results.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
As we read the filings, SpaceX’s AI strategy has three layers: 1.7 GW of installed ground compute, which Musk said on 20 October he would be surprised to see stay below 13 GW by the end of 2027; capacity rented to Anthropic; and a stated goal of 100 GW a year in orbit. Separately, it has agreed to buy EchoStar spectrum licences for about $19.6 billion. In the third quarter, its AI segment was about 52% of revenue and 84% of capital expenditures, by our arithmetic.
A note on sourcing. SpaceX figures come from its IPO prospectus (12 June 2026), its 10-Q for the quarter to 30 September 2026 and the edited transcript of its 20 October call; orbit figures are stated goals. Musk’s quotes on space are from his interview published by Dwarkesh Patel on 5 February 2026. Shares, multiples and dates derived from these are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
