The prospectus tells you what SpaceX is building. The capital architecture tells you why it can’t be unwound.
SpaceX’s S-1 registration statement, filed May 20, 2026, describes a company targeting a $1.75 trillion valuation and a $75 billion capital raise that would surpass Saudi Aramco’s 2019 IPO as the largest in history. The financial disclosures get the headlines. The capital architecture buried inside the filing, three interlocking mechanisms that create structural lock-in for counterparties, investors, and regulators, is the more durable story.
The Compute Lock-In: $105B in Contracted AI Infrastructure
The S-1 discloses compute arrangements that turn SpaceX into an infrastructure counterparty for the two highest-valued private AI companies on the planet.
This hub’s May 23 coverage of the same filing documented the headline figures: a $45 billion Anthropic compute arrangement through the Colossus data center cluster, and a $60 billion Cursor option for additional capacity. Combined, that’s $105 billion in contracted AI compute obligations, structured as multi-year commitments that create mutual dependency.
The lock-in mechanism is straightforward. Anthropic’s $45 billion commitment at Colossus scale isn’t a purchase order that can be moved to AWS or Azure next quarter. It’s an infrastructure dependency, training runs optimized for specific hardware configurations, latency paths tuned to Colossus topology, operational teams scaled to a particular cluster’s characteristics. The switching costs compound over time. Every month that passes with Anthropic training on Colossus hardware makes the alternative, migrating petabytes of checkpoints and reoptimizing training infrastructure, more expensive than continuing.
That’s not a vendor relationship. It’s a structural position.
Vertical Integration: Starlink to SpaceXAI to Enterprise Compute
The S-1 presents SpaceX as three businesses in a trench coat.
Who This Affects
Starlink, with 10.3 million subscribers across 164 countries and approximately $11.4 billion in 2025 revenue, is the cash engine. The xAI merger, now branded as SpaceXAI, is the compute division. The launch business is the orbital infrastructure layer. The vertical integration thesis is that these three businesses create capabilities none of them could achieve independently: satellite-delivered low-latency AI inference, orbital data centers with global coverage, and a distribution network that doesn’t depend on terrestrial fiber optics.
Per Fortune’s coverage of the filing, SpaceX’s prospectus presents a total addressable market estimate of $28.5 trillion for this combined strategy. That figure is issuer-generated, not independently verified. But the vertical integration is real regardless of whether the TAM is credible: SpaceX controls the launch vehicles, the orbital infrastructure, the consumer distribution network, and now the AI compute layer. No other company on earth has that stack.
The enterprise implications are concrete. If SpaceXAI achieves the compute density and latency guarantees that the prospectus envisions, it becomes the first credible non-hyperscaler compute option for frontier AI workloads. That changes procurement conversations for every enterprise currently locked into AWS, Azure, or GCP. It doesn’t change them today. It changes them in the 18-month window following a successful IPO that gives SpaceX the capital to build what’s currently in the vision section.
85.1% Voting Control: The Governance Architecture
The S-1 discloses a dual-class share structure. Per Fortune’s governance analysis, Elon Musk holds 12.3% of Class A shares and 93.6% of Class B shares, a combination that gives him 85.1% of total voting power. Class A shares carry one vote each. Class B shares carry ten.
Public investors buying at IPO acquire an economic stake in a $1.75 trillion company with no meaningful ability to influence its direction. That’s not unusual for founder-led tech companies, Meta has a similar structure, but the concentration is extreme. Musk doesn’t just control SpaceX. He controls Tesla, xAI (now merged), X, Neuralink, and The Boring Company. The cross-entity transaction disclosures in the S-1, approximately $506 million in Tesla transactions in 2025 alone, demonstrate that 85.1% voting control in a multi-entity empire creates governance surface area that minority shareholders can’t address through standard mechanisms.
For institutional investors evaluating IPO participation, the governance analysis is binary: you’re comfortable with 85.1% founder control and zero ability to contest related-party transactions, or you’re not. There’s no middle ground at this voting concentration.
What to Watch
What This Architecture Means for AI Market Structure
The SpaceX S-1 describes a company that, if successfully capitalized through its IPO, would occupy a unique structural position in AI infrastructure. It would be simultaneously:
A compute provider to two of the three most valuable AI companies (Anthropic, Cursor), with multi-year contracts that create mutual lock-in. A vertically integrated infrastructure operator controlling orbital delivery, consumer distribution, and AI training capacity. A public company where minority shareholders have no governance recourse against decisions that serve the founder’s broader corporate network.
That combination doesn’t exist anywhere else in the current market. The closest analogue is the hyperscaler model, where AWS or Azure serves as both platform and competitor, but hyperscalers have diversified governance and public market accountability mechanisms that SpaceX’s structure explicitly forecloses.
Watch the June pricing round for institutional demand signals. Watch the first quarterly earnings call for how the company reports compute revenue versus Starlink revenue. And watch whether any of SpaceX’s $105 billion in compute counterparties, Anthropic or Cursor, seeks to diversify away from a provider controlled by a single individual with cross-entity conflicts. That diversification decision, if it comes, would be the first structural test of SpaceX’s AI capital architecture.