What the S-1 Doesn't Tell You: The Bundling Problem at $1.75T

In our previous posts we laid out SpaceX's S-1 data and corrected the record on terminology — it's not a "conglomerate discount," it's a bundling problem. Investors can't cherry-pick Starlink. They must buy the entire package: Starlink, launch, xAI, Mars optionality, governance risk, all wrapped into one security at an all-in price.

This post goes one step further: what does the S-1 reveal about the structure of the valuation, and what questions remain unanswered one week before the roadshow?

The Starlink Number — Even When You're Generous

The previous post established that Starlink generates $11.4B in revenue (61% of total) and posted $7.17B in Segment Adjusted EBITDA with ~60% margins. The question everyone asks is: what is Starlink worth on its own?

Revenue multiples get messy because Starlink's margin profile is unusual. A 5x revenue multiple implies ~8x EBITDA — which is actually very conservative for a business with 50%+ growth, 60%+ EBITDA margins, and recurring revenue at scale. So let's use EBITDA:

Even at 40x EBITDA — which is richer than what any pure-play satellite or broadband operator trades at — Starlink accounts for 16.4% of a $1.75T valuation. At any reasonable multiple, it's the dominant piece of revenue, but it does not explain the valuation.

This is the real point: you do not need to argue that Starlink is "cheap" inside SpaceX. The argument is simpler and stronger. Starlink is a world-class asset, but even at a generous valuation for Starlink, the remaining ~$1.68T (at 10x EBITDA) to ~$1.46T (at 40x EBITDA) has to be justified by launch, xAI, Mars, compute infrastructure, and optionality.

The $20.7B Capex Problem

SpaceX spent $20.7B on capital expenditures in 2025. Its revenue was $18.7B.

The company spent more on capex than it earned in revenue.

This is not unusual for a capital-intensive growth business in heavy investment mode. But the S-1 breaks down where that money went, and the breakdown matters:

• AI segment consumed $12.7B of the $20.7B capex. The Colossus data center, GPU clusters, and xAI infrastructure buildout dominate the spend.
• Launch and satellite manufacturing consumed the rest, including the production of Starlink satellites and rocket hardware.

This is a consolidated capital allocation problem: the company's total capex ($20.7B) exceeds its total revenue ($18.7B), with AI consuming the majority. Starlink's operating cash flow effectively subsidizes the buildout.

The S-1 does not disclose a standalone Starlink capex figure. Segment reporting may allocate shared infrastructure costs differently. But the headline number is stark: capex > revenue, and AI is consuming the majority of it.

Anthropic: $1.25B/Month — Headline Value, Not Backlog Value

The S-1 discloses that Anthropic agreed to pay SpaceX $1.25B/month through May 2029 for compute capacity at the Colossus data center. That's a ~$45B headline number if the contract runs its full term to May 2029.

But the contract has termination rights. Reuters and Business Insider reported that Musk's own words suggest the commitment is closer to a rolling 180-day lease than a guaranteed multi-year obligation. Either party can walk away with notice.

This does not make the deal worthless. A committed customer is a real asset. But it should not be valued as $45B of backlog. A conservative valuation should haircut it heavily until there is evidence of actual run-rate continuation beyond the initial lease window.
To be fair: the S-1 itself flags this as a "commitment" rather than revenue, which is the right accounting treatment. But investors should resist the temptation to fold the $1.25B/month into forward revenue projections.

The Cursor Option: Structural Exposure Without a Liability

SpaceX entered into a $10B collaboration with Cursor (the AI-native code editor) that includes an option to acquire Cursor at a $60B implied valuation. The structure is noteworthy:

• $1.5B termination fee under the option agreement
• $8.5B deferred services fee under the compute agreement
• These fees are payable in cash, or in Class A common stock if the IPO has not closed

The maximum contractual exposure is large. But it should not be treated as a current liability or an automatic breakup check without reading the triggering conditions. The legal mechanics depend on specific clauses in the agreements that are not fully disclosed in the S-1.

What this does tell us: SpaceX is building an AI developer toolchain ecosystem. Whether Cursor ends up inside the consolidated company or not, the capital deployed to secure this relationship is real. It is part of the AI infrastructure bet.

The Musk Compensation Package: 1.3B Shares for a Mars Colony

This deserves its own section because it is unprecedented. The S-1 discloses performance grants totaling approximately 1.3 billion Class B shares to Elon Musk. Each tranche has dual triggers:

1. A market cap milestone (from $500B to $7.5T in $500B steps)
2. A Mars colony milestone (1 million people living on Mars)

There are at least 15 tranches plus a secondary award with 12 additional tranches. The milestones escalate from $500B market cap + 1M Mars colonists to $7.5T + 1M Mars colonists.

To be precise: no one knows if 1 million people can live on Mars given current physics, engineering, and economics. The requirement is physically and economically non-trivial. But the grants are already approved and filed. If triggered, they would dilute all other shareholders dramatically.

This is not just a compensation story. It is a capital structure story. The 1.3B shares represent a contingent claim on the company that is not reflected in current share count math, and it creates asymmetric upside for Musk relative to public shareholders.

Governance Red Lines

The S-1 confirms several governance provisions that triggered a public rebuke from major institutional fiduciaries:

• Dual-class structure: Musk holds 42% equity but 85% of voting rights
• Related-party transactions: Tesla invested ~$2B in xAI, which converted to SpaceX stock (Tesla now holds ~19M shares). xAI bought $506M of Megapack batteries from Tesla in 2025. SpaceX paid The Boring Company $1M for tunnel construction.
• Aircraft usage: Musk uses SpaceX-owned aircraft for Tesla CEO travel, billed back to Tesla ($1M in 2023, $2M in 2025)
• The 1.3B share grant discussed above

CalPERS and the NYC Comptroller issued a joint public statement — unusual at this scale — expressing concern over the governance structure. For context: SpaceX raised ~$10B+ from early VCs (DFJ, Founders Fund, Craft Ventures, Adapt Ventures) at valuations that will be a fraction of the IPO price. Public investors at $1.75T are entering at the outer edge of that curve.

The AI Spend: $6B Loss, $12.7B Capex, $10B+ Burn Run Rate

The AI segment (xAI + X) generated $3.2B in revenue but lost $1.24B at the EBITDA line. Total losses were $6B in 2025, with another $2.5B in Q1 2026 alone.

The capex allocation is telling: $12.7B of $20.7B total capex went to AI. The Colossus data center houses 220,000 GPUs. xAI is on pace to burn ~$10B in 2026.

This is a real bet on frontier AI infrastructure. The question is not whether AI is important — it is. The question is whether the market is pricing in $10B/year in sustained AI losses at a $1.75T valuation, with no clear path to profitability for the AI segment.
SpaceX's AI business is competing with Google, Microsoft, Meta, and Amazon — all of which have deeper pockets and more diversified revenue bases. The consolidated nature of the company means Starlink's cash flow effectively subsidizes AI buildout. Public investors would be funding a consolidated capital plan in which Starlink cash flow may help absorb xAI's losses and infrastructure buildout.

What's Missing From the S-1

The S-1 is a window, not a full picture. Several things remain opaque:

1. Starlink unit economics: subscriber growth rate, churn, CAC, LTV — these are critical for a $11.4B recurring revenue business but largely absent
2. Launch segment standalone economics: The segment is listed as losing money, but it is also the vertical integration that gives Starlink its cost advantage. Launch is strategically indispensable, but it should not be treated as a standalone cash cow on the basis of Adjusted EBITDA alone.
3. Customer concentration: US government revenue accounts for 31.6% of total revenue ($5.9B / $18.7B). Contract duration and renewal terms are not disclosed.
4. The IPO price range: Still blank. Will be set during the roadshow (week of June 8) based on institutional demand.
5. Secondary market pricing: Private platforms showed $420–$674/share in late 2025, implying $800B–$1.3T. The IPO price could be significantly higher or lower than private tender prices.

What This Means for Different Investors

Public investors at $1.75T are buying a bundled package they cannot unbundle. They get Starlink (the best part) together with xAI losses, governance risk, and Mars optionality. This is not a "discount" — it's a bundled exposure at a premium price you cannot negotiate down by segment.

Institutional investors have roadshow access and can size positions accordingly. The 30% retail allocation ($22.5B) means retail gets a bigger piece than typical mega-cap IPOs, but pricing will be set before retail has access.

Pre-IPO secondary investors who bought at $421–$674/share are sitting on 2–4x paper gains at a $1.75T IPO. But lockups are 180 days, and first-day pop depends on demand — precedent IPOs range from Alibaba (+38%) to Snowflake (+45%).

Long-term holders who believe in the integrated thesis (space + AI + compute infrastructure) may find $1.75T acceptable. But the S-1 makes clear that this is a bet on the package, not on any single asset. Starlink justifies maybe 4–16% of the valuation. Everything else has to justify the rest.

The Bottom Line

The SpaceX S-1 is a remarkable document for what it reveals and what it omits. It confirms that:

• Starlink is the cash engine: $11.4B revenue, 60%+ margins, growing fast
• AI is the cash burner: $6B+ in losses, $12.7B in capex, on pace for $10B+ burn
• Capex exceeds revenue: $20.7B spent vs. $18.7B earned
• Governance risk is real: dual-class voting, related-party transactions, unprecedented compensation structure
• The $1.75T valuation requires believing that launch + AI infrastructure + Mars + optionality are worth ~$1.46T+

If you believe Starlink is one of the most strategically valuable communications assets in the world, that is a defensible view. But you are not buying Starlink at $1.75T. You are buying the entire SpaceX package, with all its strengths, risks, and questions. The S-1 does not tell you the price is fair. It tells you what you are actually buying.

And what you are buying is a bet on the future — space, AI, and Musk's ambition — priced at 94x revenue, with GAAP losses of $6B+, and a governance structure that concentrates extraordinary control in one person.

Whether that's a good bet depends entirely on your time horizon and your tolerance for risk. The S-1 gives you the data. The rest is judgment.