Three Companies in a Rocket’s Clothing: Reading the SpaceX S-1
The most anticipated public filing in years arrived this week. The number everyone wanted was left blank. What the document does say is more revealing.
The filing landed on a Wednesday afternoon. Three hundred and eight pages, and the one number everyone had been waiting twenty-four years to see was missing. Where the price per share should sit, the document reads “$ and $ per share.” Where the founder’s voting power should be, there is a blank space and a stray percent sign. The most valuable private company ever to open its books did so with the headline left empty.
That is normal. A company’s first registration statement, the S-1, is a draft. Bankers call the early version a red herring, after the warning printed in red down the side that says the thing is not yet complete. The price comes later, in an amended filing, after the company has spent a few weeks on the road showing the business to large investors and taking their temperature. So the absence of a price is not a scandal. It is procedure.
The interesting part is what they chose to put in.
For most of its life, SpaceX was a sealed container. We knew it launched rockets, we knew Starlink had a lot of subscribers, and we knew private investors kept handing Elon Musk money at higher and higher marks. We did not know what any of it earned, what it cost, or how the pieces fit together. An S-1 ends that. To sell stock to the public, a company has to show audited financial statements, name its risks, and disclose the deals it has cut with people connected to the founder. The document that does all of this is also the document a serious investor reads most carefully, because it is the first time the company is legally on the hook for what it says.
So I read it. Here is what I found, and why it matters for anyone trying to understand where private-market value is heading.
Open the books and you find three businesses, not one
The company calls itself SpaceX and lists under the ticker SPCX. Inside, it runs as three segments, and they could not be more different from one another.
The first is Space. This is the rocket business, the part most people picture: Falcon launches, Starship development, carrying satellites and astronauts for NASA and commercial customers. In 2025 it brought in about $4.1 billion in revenue and ran roughly at breakeven on an operating basis, posting a small loss from operations of around $657 million while pouring about $3 billion into developing Starship, the next-generation vehicle the whole growth plan depends on. A launch business that funds a multibillion-dollar research program out of its own pocket and still lands near breakeven is doing something rare. It is also not where the money is.
The second segment is Connectivity, which is almost entirely Starlink, the satellite internet service. As of the end of March, Starlink had roughly 10.3 million subscribers across 164 countries and territories, served by about 9,600 satellites in low orbit. In 2025 the segment generated about $11.4 billion in revenue and roughly $4.4 billion of operating income, with revenue growing close to 50 percent year over year and operating income more than doubling. Read that again. A satellite broadband business that did not exist commercially before 2020 now throws off billions in profit and is still compounding at a rate most software companies would envy. This is the engine. Strip away the mission language about consciousness among the stars, and Starlink is the reason this company can credibly talk about a trillion-dollar valuation.
The third segment is the one that changes the whole picture. It is called AI, and it is xAI, the company behind the Grok model and the social platform X. SpaceX acquired xAI in early 2026, and xAI had itself absorbed X the year before. In 2025 this segment produced about $3.2 billion in revenue against an operating loss of roughly $6.4 billion. Its capital spending for the year came to about $12.7 billion. Put those together and the AI business consumed close to $19 billion in 2025 between operating losses and the cost of building data centers.
Now hold the three side by side. Starlink earned about $4.4 billion from operations. The AI segment lost about $6.4 billion from operations. The loss from one segment is larger than the entire profit of the other. That single fact explains nearly everything about how this company looks on paper.
The consolidated numbers are mostly a story about xAI
When you add the segments together with corporate costs, SpaceX reported $18.7 billion in total revenue for 2025 and a net loss of about $4.9 billion. A casual reader sees a company growing fast and losing a fortune, and reaches for the usual conclusion about unprofitable rockets.
That conclusion is wrong, and the segment data is what corrects it.
Look at the year before. In 2024 the combined company reported revenue of about $14 billion and net income of roughly $791 million. It made money. Then in 2025 it swung to a $4.9 billion loss. What changed? Research and development expense more than doubled, from about $3.5 billion to about $8.6 billion. That increase is xAI training frontier models and racing to build compute. Layer on close to $1.9 billion of interest expense on borrowed money, and the profitable company of 2024 becomes the deeply unprofitable company of 2025, without the rocket-and-broadband core doing anything different.
There is an accounting wrinkle worth understanding here, because it shapes how you read every historical figure in the document. xAI and X were brought under the same control as SpaceX through transactions among entities Musk already dominated. When that happens, the rules require the company to combine the financials retrospectively, as if the three had always been one. So the 2023 and 2024 figures you see for “SpaceX” already include xAI and X, even though SpaceX did not own them at the time. The history has been rewritten to match the present structure. It is legitimate and required. It also means the clean, standalone SpaceX that existed before the AI deal is no longer visible in these statements. You have to reconstruct it yourself from the segments.
When you do, the standalone rocket-and-Starlink business looks like this: roughly $3.8 billion of combined segment operating income in 2025, growing quickly, carrying its own research bill. That is a real, profitable, expanding company. The thing dragging the consolidated result into the red is the AI segment that was stitched on a few months before the filing.
The balance sheet absorbed someone else’s debt
Here is where the picture turns from interesting to concerning, and where I would slow an LP down before they get swept up in the narrative.
In March 2026, weeks before this filing, SpaceX took on a $20 billion bridge loan. The proceeds did not go to build rockets or launch satellites. They went to repay debt that belonged to X and xAI: a stack of term loans and a tranche of senior secured notes carrying a 12.5 percent rate, the kind of expensive borrowing a struggling social media company takes on when cheaper money is unavailable. The clean, cash-generating space company put its name on $20 billion of borrowing to clean up the financing of Musk’s social and AI ventures.
Total debt and finance leases now sit around $23 billion. The accumulated deficit, the running tally of all the losses the company has booked over its life, stands at about $41 billion as of the end of March. The company also bought back nearly $4 billion of its own stock in connection with the xAI acquisition. None of this is fatal for a business generating real cash from Starlink. All of it is debt and dilution that the rocket business did not create and would not carry on its own.
The use of proceeds section tells you where the new money from the IPO is meant to go: expanding AI compute infrastructure, improving launch vehicles, scaling the satellite constellations, and general corporate purposes. Notice the order. Compute comes first. The company is telling you, in the polite language of a prospectus, that a meaningful share of what public investors put in will flow toward GPUs and data centers, not toward the launch business that made the name.
The capital appetite is the headline risk
If there is one number in this filing that should stop a reader cold, it is not the revenue or the loss. It is the capital spending.
Across all three segments, SpaceX spent about $20.7 billion on capital projects in 2025. Then in the first quarter of 2026 alone, it spent about $10.1 billion, and roughly $7.7 billion of that single quarter went to the AI segment. Annualize that pace and you are looking at a company that wants to spend on the order of $30 billion a year building AI infrastructure, on top of everything the space and connectivity businesses require.
This is a different kind of company than the one most people think they are buying. The romance is rockets and Mars. The reality on the page is a capital-intensity profile that looks like one of the large cloud computing companies, with a satellite network and a launch business attached. Whether that is brilliant or reckless depends entirely on a bet the company is making out loud: that owning the compute, building it faster and cheaper than anyone else, and eventually moving it to orbit where the Sun provides limitless power, will produce returns that justify the spend. The company says it expects to begin deploying AI compute satellites as early as 2028. That is a remarkable sentence to underwrite. It is also unproven in every dimension that matters.
Two deals that tell you how this company thinks
Buried in the business section are two arrangements that reveal more about the company’s strategy than the mission statement does.
The first involves a coding software company called Cursor, run by a firm named Anysphere. In April 2026, SpaceX agreed to supply Cursor with GPU compute and took an option to buy the company outright at an implied value of $60 billion, payable in SpaceX stock. An option, not a commitment. Here is the part that matters. If SpaceX walks away, or if Cursor terminates the deal because SpaceX breached it, Cursor collects a $1.5 billion termination fee plus an $8.5 billion deferred services fee. That is $10 billion of downside exposure attached to an acquisition the company has not even decided to make, of a target it would pay $60 billion for, before it has finished going public. The structure tells you that this is a company comfortable writing checks with more zeros than most national budgets, on optionality alone.
The second deal points in the other direction, and it should interest anyone holding a position in the broader AI landscape. In May 2026, SpaceX entered into cloud services agreements with Anthropic, the AI research company behind the Claude models and a direct competitor to xAI’s Grok. Under the agreements, the customer pays SpaceX $1.25 billion per month through May 2029 for access to compute capacity in SpaceX’s data centers. That is roughly $15 billion a year, up to something near $45 billion over the term, terminable by either side on 90 days’ notice.
Sit with the strangeness of that. xAI, which exists to build a frontier model that competes with Anthropic’s, is now selling Anthropic the compute it needs to train against xAI. SpaceX frames this as smart monetization of capacity it would otherwise leave idle, and as a business matter it is hard to argue with $15 billion a year of high-margin revenue from someone else’s training runs. As a strategic matter it tells you how desperate the demand for compute has become, and how thin the line now runs between collaborator and competitor in this industry. For anyone weighing the durability of an Anthropic position, the disclosure cuts both ways. It confirms Anthropic’s compute needs are enormous and growing. It also reveals that a large slice of those needs now depends on infrastructure controlled by the person most determined to beat them.
You are buying a passenger seat
Now the part that experienced investors read before anything else, and that retail buyers tend to skip.
SpaceX will have two classes of stock in public hands. Class A, the kind being sold, carries one vote per share. Class B, which Musk holds, carries ten votes per share, and the Class B holders get to elect a majority of the board outright. The filing states plainly that Musk will control the outcome of essentially every matter requiring a shareholder vote, and that the company will operate as a “controlled company” under Nasdaq’s rules. That status lets it skip several governance requirements that protect ordinary shareholders, such as having a majority-independent board.
The document is honest about what this means. It says, in the risk factors, that the structure will limit or preclude your ability to influence corporate matters or the election of directors. The bylaws go further, pushing many shareholder disputes into mandatory arbitration and restricting where and how investors can bring claims. The company does not plan to pay dividends. So your entire return depends on the share price rising, while you hold a security whose voting power rounds to nothing against the founder’s.
This is not unusual for a founder-led technology company, and plenty of them have rewarded shareholders handsomely despite it. The point for an LP is to name the trade clearly. You are not buying a say in the company. You are buying exposure to one person’s judgment, across a set of businesses he runs simultaneously, financed in part by debt that originated outside the business you thought you were buying.
The web around the founder
That last point deserves its own paragraph, because the related-party section of this filing is unusually dense.
Musk is the chief executive of SpaceX. He is also the chief executive of Tesla, and the filing describes several joint efforts between the two companies. There is Terafab, a chip manufacturing project with Tesla and Intel aiming to produce enormous quantities of compute. There is Macrohard, an agentic AI platform under development with Tesla. SpaceX uses Tesla battery systems in its operations. The founder is separately involved in Neuralink and The Boring Company. The directors include people whose investment firms hold large stakes and have done business with the company for years.
None of these relationships is hidden. They are disclosed at length, which is the system working as intended. But the cumulative effect is that buying SPCX means buying a node in a tightly connected industrial network, where capital, talent, contracts, and compute move among entities the same person controls. When those transactions are priced fairly, the network is a strength, letting the pieces reinforce one another. When they are not, minority shareholders have little recourse, given the voting structure described above. Reasonable people will weigh that differently. No one should ignore it.
So what is it worth, and what does this mean for us
The filing does not say what the company thinks it is worth. Press reports have circulated a valuation in the neighborhood of $1.75 trillion and a raise in the tens of billions, which would make it the largest public offering in history. Those figures are reporting, not disclosure, and they may move before the deal prices.
If something like that valuation holds, the company would be priced at roughly ninety times its trailing revenue while losing close to $5 billion a year. That multiple is incomprehensible for a normal business. The argument for it rests almost entirely on three things: Starlink’s profitable, fast-compounding growth; Starship as a step change that could collapse the cost of putting mass into orbit and unlock businesses that do not exist yet; and the wager that orbital AI compute becomes real this decade. The first is proven and visible in the numbers. The second is probable but unscheduled. The third is a hope with a date attached.
For a firm like ours, the relevance is not whether to buy the stock. It is what this filing signals about the world we invest in.
It resets the ceiling. When the most valuable private company finally prints its financials, every other private valuation gets measured against a real set of audited statements rather than a rumor. That is healthy. It also tightens the screws on the AI infrastructure trade. A company willing to spend $30 billion a year on compute, and another willing to pay $15 billion a year to rent it, are telling you that the cost of staying at the frontier has reached a scale only a handful of balance sheets on Earth can sustain. Early-stage founders building on top of that frontier should understand whose economics they are riding.
And for the Anthropic exposure specifically, the disclosure is worth holding in both hands. It confirms the demand. It also names the dependency.
The thing I keep returning to, having read these 308 pages, is how little the document is about rockets. The mission section talks about Mars and consciousness and the light of the stars, and it is genuine, and it is moving. Then you turn to the financial statements and the largest single line of spending is graphics processors. The company that taught the world rockets could be reusable is now, by the measure of where its capital goes, an artificial intelligence company that happens to own the best launch business and the best satellite network ever built.
Whether that combination produces the greatest value creation of our lifetime or the most expensive conglomerate discount in market history is the question the next few weeks will start to answer. The blank where the price should be is the company asking the market to decide.
This analysis is based on the preliminary Form S-1 as filed. Figures are drawn from the filing itself; the valuation and offering size discussed are from press reports and are not stated in the document. Nothing here is investment advice. Team Ignite Ventures and its affiliates may hold positions in private companies referenced, including primary and secondary exposure in the sector.