The SpaceX IPO: Is It a Space or an AI Company? (Part 3: A Close Reading)
This is Part 3 of a four-part series on the SpaceX IPO. Part 1 valued SpaceX’s operating businesses at approximately $1T against an IPO target of $1.75T to $2.0T. Part 2 examined what the $1T premium on top of the valuation ostensibly buys.
We now turn our attention to the prospectus itself.
Part 3: A Close Reading of the S-1
Disclosure Choices
The S-1 quantifies certain operational metrics but leaves a striking number of peer-standard disclosures absent. This is not unusual behavior, but investors should price the omissions in. For a company controlled by an operator with a long history of selective disclosure, the absences are worth examining individually because each one says something about which numbers SpaceX preferred not to make public. The most material gaps are cataloged below. Where possible, we note what is missing, what comparable peers disclose, and the likely direction of impact if SpaceX disclosed the same thing.
The disclosure gaps roughly fall into two clusters: AI-segment commercial commitments (Items 1-3) and traditional satcom unit economics (the rest). Investors looking to test the AI growth thesis will find the first cluster more material.
Named customers above 10% of revenue. Rocket Lab discloses a Government Customer at 28% of revenue and MDA Corporation at 23% in FY2024. CoreWeave discloses Microsoft at 67% of FY2025 revenue. While SpaceX does disclose that 20% of its FY2025 revenue came from U.S. federal agencies in aggregate, there are no other disclosures of customer concentration, beyond qualitative commentary. Since peers do disclose customer concentration and even revenue splits, it’s possible that SPCX’s customer concentration is unflattering to its growth story.
Segment-level remaining performance obligations. CoreWeave discloses $60.7B of RPO at FY2025 year-end, grown four times in a single year. SpaceX, with the $45B nominal Anthropic contract and NASA/NSSL commitments, likely has comparable or larger RPO. Disclosure would force the customer-concentration disclosure that the S-1 currently avoids. The likely direction is unflattering for the same reason as Item 1.
AI segment power capacity, RPO, and customer commitments. SpaceX discloses the Anthropic deal (reported elsewhere as ~300MW) at a headline level but does not provide CRWV-style breakdowns: total contracted power, committed-contract-percentage-of-revenue, or customer-by-name contract values. CRWV publishes all three, which allows investors to back into capacity utilization. With AI segment capex running at $30B annualized against $3.2B in revenue, these proxies would tell investors whether the build-out is keeping pace with bookings (CRWV-style) or running ahead of bookings (speculative build).
Starlink subscriber mix. Iridium and EchoStar disclose subscriber composition (commercial versus government, for instance). SpaceX discloses only a single Starlink subscriber number. The mix among consumer broadband, enterprise, government Starshield, and Mobile is unclear. The direction is genuinely ambiguous: an enterprise-heavy mix would flatter ARPU and margin; a consumer-heavy mix would suggest the $66 ARPU is increasingly residential and increasingly price-sensitive.
Starlink churn. Iridium and EchoStar disclose net subscriber change each period. SpaceX discloses only ending balance. For a business with rapidly declining ARPU, churn disclosure would clarify whether the subscriber growth is healthy or whether the gross adds are partly offsetting departing customers. The likely direction is unflattering.
Subscriber acquisition cost and terminal subsidy. EchoStar discloses subscriber acquisition costs of $195 million in FY2024, implying approximately $240 per gross add. Starlink kits are subsidized in many international markets, but the magnitude is invisible in the S-1. The likely direction is unflattering: aggressive international subsidies imply long payback periods at the $66 ARPU level.
Geographic revenue split. ASC 606 typically requires this disclosure for public companies. We know Starlink operates in 164 countries. A U.S.-heavy mix would flatter ARPU and margin; an international-heavy mix would expose FX risk and lower ARPU. The absence of disclosure may also reflect sensitivity at the country level, since Starlink has unique exposures in markets including Russia, Iran, and China.
Starshield revenue line item. Starshield is the explicitly defense-focused product but is not broken out from Connectivity revenue. Classified-work caveats provide some explanation for non-disclosure, but the product itself is openly discussed. If small, disclosure would deflate the implied DoD relationship while reassuring investors that consumer Connectivity is the engine. If large, disclosure would dominate the government-revenue concentration story.
Service vs. equipment gross margin split. Iridium discloses service and equipment margins separately. SpaceX provides Connectivity EBITDA margin (64%) on a blended basis. Peer cable and satcom service businesses run 70%+ service margins. While equipment is likely a drag, given terminal subsidization, disclosure would clarify whether unit economics improve as the subscriber base matures.
Starship unit economics and breakeven case. The S-1 says SpaceX’s growth strategy is “highly dependent on the successful development and scaling of Starship” but provides no information related to target launch and production costs. RKLB at least provides metrics that allow you to back into these numbers. However, since Starship is still under R&D, the absence may reflect genuine engineering uncertainty.
Fairness-opinion process for related-party transactions. Major related-party transactions (see next section) were all entered into pre-IPO. The S-1 does not describe an independent board review or fairness-opinion process. Disclosure would either confirm there was no such process (strongly unflattering) or describe one that exists (which would presumably already be disclosed if so). The likely direction is unflattering.
The Competitor the S-1 Does Not Foreground
The S-1’s competitive risk-factor language is generic, mentioning Amazon only in passing when in fact Amazon has assembled a vertically integrated stack that maps directly to SpaceX: Blue Origin provides launch services, Project Kuiper provides LEO connectivity, Amazon Web Services provides compute infrastructure. Amazon’s strategic investment in Anthropic provides frontier AI exposure. The four pieces correspond to SpaceX’s three segments, plus its largest AI customer relationship.
Amazon’s trajectory here is visible in their SEC disclosures even though this activity is not reported as a separate segment. Prepaid satellite launch deposits stood at $6.9 billion at December 31, 2025. And though only a small fraction is likely for Kuiper, AWS property and equipment additions totaled $96.5 billion in FY2025, exceeding SpaceX’s total consolidated revenue by five times and dwarfing SpaceX’s AI segment capex. Blue Origin holds 27 New Glenn launches contracted for Kuiper, representing $2.7 billion of direct contracted value, plus $4.7 billion of additional Kuiper launches via ULA’s Vulcan rocket and a $2.4 billion award under NSSL Phase 3 Lane 2. Together these disclosures describe a competitor with the capital, the launch contracts, and the strategic intent to challenge Starlink across the same markets. On a five to ten year horizon, the competitive landscape likely resolves toward a duopoly between SpaceX and Amazon, not a SpaceX monopoly. The S-1 does not engage with this directly.
Related-Party Arrangements
The S-1 catalogs a rich list of related-party transactions:
xAI is Musk’s AI startup, founded in 2023 and headquartered in San Francisco. It built Grok, a large language model that competes with ChatGPT and Claude. In March 2025, it was valued at $80B when it acquired X (formerly Twitter), valued at $33B, in an all-stock deal. In February 2026, SpaceX acquired xAI in an all-stock transaction at a $250 billion implied valuation. For both transactions, Musk sat on both sides of the table. The S-1 does not disclose any independent fairness-opinion process, which is unusual for a related party M&A (in contrast to Tesla’s 2016 acquisition of SolarCity), especially one that completed just a few months before the S-1 filing and now anchors a meaningful part of the IPO valuation. One resulting event to note is a $4.94B recorded loss on the merger.
Valor Equity Partners is a private equity firm founded by Antonio Gracias, a SpaceX board member and 7.3% Class A shareholder. Across three separate lease agreements, Valor has agreed to provide xAI with compute equipment in exchange approximately $20.2B in aggregate cash payments over the life of the leases. SpaceX guarantees the payment obligations. Of the aggregate, $885M was paid in FY2025 and $857M was paid in just January and February of 2026, a run rate above $400M per month. The arrangement commits roughly 5% of SpaceX’s current operating cash flow on an ongoing basis to a director’s firm.
Tesla sold SpaceX $144M of goods and services in FY2025, a thirty-six-fold increase from $4M in FY2024. xAI separately bought $506M from Tesla in FY2025, but just $34M in January and February 2026 (annualized, $205M). The S-1 describes these as priced at terms “no less favorable to SpaceX than those generally available to unaffiliated third parties under similar circumstances” but does not describe an arm’s-length benchmarking process, what goods and services are being bought, nor whether they are contractually recurring. In other words, SpaceX/xAI buys from Tesla at material scale, but the spend is lumpy and purpose unclear, which means investors cannot evaluate whether the relationship is being managed in shareholders’ favor.
Macrohard and Terafab are joint ventures with Tesla announced in March 2026. Macrohard is positioned as an agentic AI platform intended to compete with Microsoft’s enterprise software business. Terafab is a chip manufacturing initiative with Tesla and Intel. The S-1 states that “we and Tesla have not finalized a variety of details relating to our collaboration, including, but not limited to, financial terms, intellectual property rights, and the ultimate term of our collaboration.” SpaceX has entered into significant strategic collaborations with a public company controlled by its CEO, before agreeing to the economic terms. The structural protection for public-shareholder interests in this arrangement is limited.
Anysphere (operating as Cursor) is an AI coding assistant startup with no Musk affiliation. In April 2026, SpaceX entered into an agreement with Cursor, wherein SpaceX provides Cursor with GPU capacity (in exchange for data/IP) and SpaceX has a call option to acquire Cursor at a $60 billion implied equity value. If SpaceX terminates the option agreement, Cursor receives a $1.5 billion termination fee plus an $8.5 billion deferred services fee, totaling $10 billion of contingent exposure for SpaceX. The Cursor commitment is smaller in absolute terms than the Musk-affiliated commitments above, but it adds to a pattern of large pre-IPO commitments.
In aggregate, there are at least $35B of pre-IPO commitments to Musk-affiliated entities ($4.94B of recorded loss on the xAI absorption, $20.2B of guaranteed payment obligations to a director’s firm, and $10B of contingent Cursor exposure) against $30B-34B of FY2026E consolidated revenue. In other words, the pre-IPO related-party commitments amount to roughly one year of consolidated revenue, with the overwhelming majority of them due to AI, rather than Space or Connectivity segments.
Each commitment is individually defensible, but the aggregate transfers downside risk to public shareholders while preserving upside for Musk and his affiliates. The corporate opportunity waiver (discussed next) makes this an ongoing concern rather than a one-time inventory, as future opportunities can continue to be directed to Musk-affiliated entities without legal duty to SpaceX.
Governance Items
Three features of the governance package materially affect public-shareholder rights.
Musk holds 85.1% of voting power pre-IPO through Class B Common Stock, which carries 10 votes per share. This modestly dilutes after the IPO, but the concentration still sits at the high end of dual-class structures (Meta is approximately 56% post-IPO; Snap was approximately 89% at its IPO). Combined with the Texas reincorporation (see #3 below), the SpaceX dual-class is more management-friendly than Delaware-based dual-class peers.
The SpaceX charter includes a corporate opportunity waiver, which explicitly permits Musk and his affiliates to direct future business opportunities (acquisitions, technologies, customers) to other ventures without offering them to SpaceX first. Most public companies impose at least a duty of disclosure on such opportunities. SpaceX does not, which is exactly what allows the related-party pattern to continue post-IPO.
SpaceX reincorporated in Texas in 2024 as a coordinated structural choice that allows it to use mandatory arbitration provisions in its IPO. These provisions require investors to resolve federal securities claims through private arbitration rather than in court. The practical effect is to eliminate public-company investors’ class-action securities lawsuits and to remove the public-disclosure deterrent that litigation tends to produce through discovery and settlement. The SEC reversed its long-standing policy in September 2025 to allow mandatory arbitration provisions in IPO charters. While Delaware law prohibits these clauses, Texas allows them; SpaceX’s 2024 reincorporation in Texas positioned it to use the newly available structure as soon as the SEC policy permitted. The combined effect of the reincorporation and the arbitration clause is that the standard post-IPO enforcement pathways available to public investors at peer companies are materially weaker at SpaceX.
We could not find a comparable dual-class mega-cap IPO that had all of these governance features simultaneously. For example, Meta and Alphabet have dual-class supervoting but remain Delaware-incorporated and do not have mandatory arbitration. Tesla has the corporate opportunity waiver but Delaware courts retain jurisdiction over disputes. The combined effect at SpaceX is that the standard mechanisms by which public shareholders constrain controlling-shareholder behavior are all materially weaker than at peer public companies.
Strategic Event Timing
The AI segment is doing most of the heavy-lifting in our FY2026 projection (see Part 1). For a company whose origin and brand are anchored in space and connectivity, having the AI segment carry the growth story is already unusual. What makes the position even more unusual is the timing of the events that built the AI segment to its current form:
February 2026: xAI merger completed
March 2026: Macrohard (agentic AI) and Terafab (AI chip) joint ventures announced, terms not finalized
April 2026: Cursor agreement signed
May 2026: Anthropic compute commitment announced
Each event contributes to the prospectus narrative, but are too recent for audited annual financial statements.
A skeptical read would be that SpaceX is a space and connectivity company that timed four large AI-related events into the months immediately before its IPO, and is now relying on that segment’s projected growth rate to anchor its IPO valuation. The thesis connecting the Launch and Connectivity segments to AI rests primarily on future deployments of AI compute satellites, for which SpaceX has yet to disclose any prototypes, let alone commercial-scale operation. Aside from that single linkage, which is itself aspirational, the AI segment is operationally distinct from the rest of SpaceX. This admits the possibility that the company is taking advantage of the current AI capital-spending cycle to attach an AI premium that the launch and connectivity businesses alone would not support, with the timing of the AI events chosen to maximize the contribution to the IPO multiple. The SpaceX and xAI merger, in particular, could have been consummated at any point over the prior two years, since Musk controlled both.
On the other hand, it is also possible that the concentration of AI-related events is a temporary structural opportunity. Hyperscalers cannot build power and compute capacity fast enough to meet demand from frontier AI labs, and a window has opened for other operators. SpaceX/xAI had been building towards this since xAI’s launch (July 2023) and the Colossus build-out (2024-2025), so the timing of the public-disclosure events may reflect when Anthropic and Cursor were ready to commit, not sequencing engineered by SpaceX. Anthropic was reportedly shopping for gigawatt-scale capacity across providers and chose SpaceX’s timeline. Cursor’s option structure (exercisable in a 30-day window after the IPO) reflects Cursor wanting to see SpaceX’s public valuation before committing to a sale. In this more charitable reading, the IPO timing reflects the capital needs of continuing the AI build-out, not an effort to extract a higher valuation while the AI cycle is hot.
Coming in Part 4…
Regardless of fundamentals or disclosure quality, SPCX will not trade like an ordinary company in the near term. In many ways, the company has yet to figure out whether it is a space company or whether it is an AI company. In Part 4, we talk about how despite these uncertainties and dynamics, prospective investors can think about whether to enter, and when to exit, in a more principled way.
We put together this four-part analysis in just a day using Revelata’s deepKPI. You can do this kind of work yourself by signing up for our free tier. No credit card required.
 | A guest post by
|