SpaceX priced the largest IPO in history on June 11, 2026, raising roughly $75 billion at a valuation near $1.77 trillion, and the prospectus behind that debut does not describe the company most investors thought they were buying. For two decades, the market understood SpaceX as a launch and satellite business. The S-1 describes a vertically coordinated infrastructure system spanning connectivity, space, and AI, in which two economically productive legacy segments carry one deeply loss-making AI business.
For most of the SaaS era, investors operated on a clean hierarchy. Software was the premium layer, with high margins and low capital intensity, while physical infrastructure was the commodity underneath it. The SpaceX filing pressures that hierarchy in a way worth taking seriously, because the segment that prints the cash is the one moving photons through space, and the segment burning the cash is the AI lab.
The cash engine is connectivity, and the loss is concentrated in AI
Start with what the filing documents. SpaceX reported $18.674 billion in total revenue for 2025, up from roughly $14.1 billion in 2024, with a consolidated GAAP loss from operations of $2.589 billion and adjusted EBITDA of positive $6.584 billion.
Inside the consolidated numbers, the segments diverge sharply. The Connectivity segment, primarily Starlink, generated $11.387 billion in revenue in 2025, about 61 percent of company revenue, on income from operations of $4.423 billion and segment adjusted EBITDA of $7.168 billion. It is the only segment that prints consistent GAAP operating profit. The Space segment, the launch business, produced $4.086 billion in revenue. It is important to be precise here: Space ran a GAAP operating loss of $657 million in 2025, weighed down by roughly $3 billion in Starship research and development, while still generating positive adjusted EBITDA of $653 million. The AI segment, which houses xAI after a February 2026 merger, generated $3.201 billion in revenue against a $6.355 billion operating loss.
Connectivity prints the profit. Space is strategically vital but capital-hungry. AI is the bet. All three sit inside one balance sheet, and the consolidated loss is almost entirely the AI segment.
The structure is the story. The consolidated loss is driven by AI. Strip the AI segment out, and the Space and Connectivity segments together produced roughly $7.8 billion in adjusted EBITDA in 2025. A satellite internet business with more than 10.3 million subscribers as of March 2026 is generating the operating profit that anchors the entire enterprise.
One technical correction worth stating plainly, because it gets reported wrong constantly. The gap between positive adjusted EBITDA and the GAAP net loss is not explained by AI capital spending directly. It is reconciled by depreciation and amortization, stock-based compensation, interest, and taxes. AI capital expenditure, which ran to roughly $12.7 billion for the segment, affects cash flow and then flows through the income statement later as depreciation. The distinction matters because conflating capex with the EBITDA-to-net-loss bridge is exactly the kind of imprecision a skeptical CFO uses to dismiss an entire argument.
“Starlink is financing the AI” is close, but say it precisely
It is tempting to write that Starlink cash is paying for xAI. The filing does not document a direct transfer of Starlink revenue into the AI segment, and I am not going to assert one. The defensible formulation is narrower and more interesting.
Starlink’s profitability reduces, but plainly does not eliminate, the combined enterprise’s dependence on external capital. It gives the company greater capacity to absorb AI losses and, just as importantly, improves the terms on which it can raise the enormous additional capital the strategy requires. The evidence that external capital remains essential is unambiguous: SpaceX just raised $75 billion from public markets, carries roughly $29 billion in long-term debt, and disclosed a $20 billion short-term bridge loan due within six months of the listing. A profitable connectivity business changes the financing math for the whole system. It does not make the system self-funding.
The portfolio behaves like a system, and the filings show the wiring
The related-party disclosures across Tesla and SpaceX make the coordination concrete. The wiring is documented, not inferred.
Tesla committed a $2 billion investment in xAI in January 2026 through preferred stock. After SpaceX completed the xAI merger, that right converted into SpaceX Class A shares, issued in March 2026. As of May 1, 2026, Tesla held roughly 19 million SpaceX shares, less than 1 percent of the total.
xAI bought Tesla Megapacks to power its data centers, reaching roughly $1 billion in cumulative purchases since 2024, including a single $269 million purchase in April 2026.
SpaceX purchased Tesla vehicles, with Tesla recognizing $143.3 million in SpaceX-sourced revenue in 2025. Tesla has stated those vehicle sales used rates generally available to unaffiliated parties.
Notice what one entity supplies to the next. Tesla supplies energy storage and capital. SpaceX supplies launch, satellites, and a balance sheet. xAI supplies models and data-center demand. X supplies data and distribution. Each business is, at various moments, a supplier, a customer, a financier, or a distribution channel for the others.
Precision matters here, because the lazy version of this story calls these transactions cross-subsidies. The evidence does not support that word. Tesla recognized roughly $430 million in xAI Megapack revenue in 2025 against about $285 million in cost, which is a markup, not a subsidy. The accurate description is strategic portfolio coordination on apparently commercial terms, not below-market favors. That distinction is the difference between a defensible thesis and a lawsuit.
It also cuts the other way. The same filing shows how much the system depends on parties Musk does not control. In June 2026, SpaceX disclosed an agreement to provide Google with compute capacity supported by roughly 110,000 Nvidia GPUs, at about $920 million per month from October 2026 through June 2029, alongside a separate arrangement renting Colossus 1’s compute to Anthropic at roughly $1.25 billion per month. Those deals turn AI infrastructure into recurring revenue, which strengthens the story. But they also underscore the dependence: the system runs on Nvidia silicon it must buy, it sells capacity to competitors to defray the cost, and it just asked public shareholders for $75 billion. This is coordination, not autarky.
Why the software-era hierarchy is bending
For twenty years the playbook said own the application layer and let someone else own the data centers, the fiber, and the power. In the AI era that logic weakens, but the precise claim matters. Value is not simply draining out of software. Bargaining power is migrating toward scarce inputs, and in this cycle many of those inputs are physical: compute, energy, advanced manufacturing, connectivity, and distribution. Whoever controls a constrained input gains pricing power over everyone who needs it. Model quality, data, and distribution can still capture enormous value. What is changing is the leverage that flows to whoever owns the scarce layer.
Amazon is the cleanest comparison, with one honest caveat. AWS was the infrastructure business that funded Amazon’s ambitions for years, and Amazon now spends heavily to control its own chips, data centers, and power for AI. The hyperscalers broadly are integrating downward into silicon and electricity. The imperfection in the comparison is governance: AWS grew inside a single corporation, while Musk’s system spans separately incorporated and separately governed entities transacting with one another. That is what makes his version both more aggressive and more fraught.
Where CPAG’s lens adds something
The integration that creates strategic strength simultaneously degrades financial clarity, and that is an attribution problem before it is anything else. When a Megapack sale, a compute lease, and an equity conversion all flow between entities under one controlling shareholder, it becomes genuinely hard to determine which entity created the value, which bore the cost, and which deserves the economic credit. Bundled infrastructure can raise execution speed while lowering transparency, and the harder it is to attribute value and cost independently, the more weight falls on governance, disclosure, and verification to hold the structure accountable. That tension does not resolve itself with scale. It intensifies with it.
The strongest objections
A disciplined read has to hold the objections in view, and they are serious.
The first is governance and valuation clarity. Consolidation may improve financing capacity while destroying the ability to price the parts. Investors cannot separately value Starlink’s profitable economics and xAI’s losses now that they sit inside one controlled company that has opted out of independent-board requirements, with Musk holding the overwhelming majority of voting power.
The second is that public shareholders, not Starlink, are now funding the ambition. The $75 billion raise is the point. The combined enterprise did not finance the AI bet internally. It turned to public markets for another $75 billion.
The third is execution risk concentrated in one person and one rocket. The S-1 flags dependence on Musk and on Starship as top risk factors, and nearly every growth plan, from V3 satellites to orbital AI compute, depends on Starship commercializing on schedule. Interconnection distributes strength in good conditions and contagion in bad ones.
The broader implication
The interesting claim in this filing is not that Musk owns several large companies. It is that the companies increasingly function as one capital and infrastructure system, in which a profitable connectivity business improves the financing terms for everything else in the structure. The market may be valuing SpaceX less as a satellite company with an AI side bet and more as the physical operating layer for a much larger ambition.
If that read is right, the lesson for operators and investors is not “copy Musk.” Almost no one can assemble these pieces, and the same structure that creates the advantage also concentrates governance risk and obscures valuation. The lesson is about where leverage is migrating. In an era defined by scarce compute, energy, and connectivity, control over a constrained physical input is becoming a source of durable bargaining power that the application-layer playbook does not capture. The premium is not abandoning software. It is being pulled toward whoever owns the scarce input. The SpaceX filing is the clearest documented example yet of someone building the whole stack on purpose, and asking the public markets to fund the part that does not yet pay for itself.
Sources
- SpaceX S-1 / S-1A, SEC (filed May 20, 2026; amended June 1 and June 5, 2026) — primary filing for segment figures, capital expenditure, risk factors, related-party disclosures, and Google compute agreement.
- Reuters, “Musk’s SpaceX raises $75 billion in largest IPO ever,” June 11, 2026 — IPO pricing at $135 per share, approximately $75 billion raised, and approximately $1.77 trillion valuation.
- Reuters, “SpaceX by the numbers: six charts,” June 11, 2026 — Starlink customer count and revenue share.
- Via Satellite, “SpaceX’s IPO Filing Gives First Look Into Company’s Financials,” May 20, 2026 — segment breakdown, debt, and controlled-company status.
- Yahoo Finance / Reuters, “SpaceX files IPO prospectus,” June 2026 — Space segment detail and consolidated figures.
- TechCrunch, June 5, 2026 — Google compute agreement details, including compute capacity supported by approximately 110,000 Nvidia GPUs.
- CNBC, June 2026 — Anthropic Colossus 1 compute arrangement.
- 24/7 Wall St., “Tesla’s $2 Billion xAI Bet Just Got Folded Into SpaceX,” May 21, 2026 — Tesla preferred-stock conversion and SpaceX shareholding.
- Electrek, May 1 and June 4, 2026 — Tesla related-party disclosures, including xAI Megapack revenue, SpaceX revenue, cumulative Megapack purchases, and April 2026 purchase.
- Fortune, “Elon Musk’s SpaceX buys xAI,” Feb. 2, 2026 — Megapack sales and xAI investment context.
About the Author
Andrew Miller is the founder of Crown Point Advisory Group (CPAG), where he helps B2B SaaS founders eliminate executive debt and architect scalable leadership structures. Previously, he served as the founder and CEO of Cameyo through its acquisition by Google. He also writes The Middle Way in AI, a blog dedicated to cutting through the utopian and doomer extremes to provide grounded, pressure-tested analysis for operators, founders, investors, and policymakers navigating the transition.