Business

Know the Business

SpaceX is three economically distinct companies sharing one balance sheet: a launch monopoly that operates mostly as internal infrastructure, a global LEO broadband ISP (Starlink) that is the actual cash engine at a 63% segment EBITDA margin, and a money-losing AI hyperscaler (xAI) absorbing 60% of FY2025 capex for 17% of revenue. At the ~$1.5T private mark, the buyer is paying mostly for the option value of Starlink scaling at telecom-like steady state and xAI surviving the GPU arms race. The market is almost certainly underestimating how good Starlink's standalone economics already are, and overestimating how quickly the xAI segment stops burning cash.

1. How This Business Actually Works

SpaceX runs a vertical reinvestment loop: Falcon 9 reusability collapsed the cost of putting mass into orbit, which made Starlink commercially viable, which generates the cash flow that funds Starship and xAI compute. Launch sells to outsiders at a premium price (because no one else can match its cadence) but spends most of its cadence flying its own payloads — only 43 of 165 FY2025 Falcon launches were for paying customers. Launch revenue looks small because most of the value the launch business creates is captured one floor up, inside Starlink.

The economics by segment look almost nothing alike. Connectivity is a subscription business: 10.3M Starlink Service Lines as of Q1 2026 paying $66/month on average, with a network whose marginal cost of an additional subscriber is near zero once the satellite is overhead. Space is per-flight launch revenue at premium pricing, but the productive cadence is allocated internally — the "profit" is captured by the segment that uses the launches, not the one that sells them. AI is currently a capital-vortex: 1.0 GW of GPU compute, ~117M monthly Grok users, no path yet to operating profitability, and every dollar of revenue costs roughly four dollars of capex.

The non-obvious mechanic: incremental profit comes from Starlink subscribers, not launches. Launch is the moat; Connectivity is where the moat compounds. Treat launch as the equivalent of Amazon's fulfillment network — the part of the company that sets per-unit economics for everything else. AI is the optionality bet that may or may not justify the current spend.

2. The Playing Field

There is no single peer for SpaceX. Each public comparable maps to one slice of the business, and several of them trade at multiples that already imply the SpaceX outcome they are racing toward.

Four takeaways. SpaceX's revenue is larger than its four NASDAQ peers combined, and only Lockheed Martin exceeds it — but LMT earns $7.7B of operating profit on $75B of revenue at a 10.2% gross margin, the opposite of SpaceX's model. RKLB and ASTS already trade on optionality (74x and 405x revenue); they are priced as miniature SpaceXes-in-waiting, which suggests the public market does pay up for the playbook SpaceX has already executed. IRDM is the only peer that demonstrates the steady-state math — 71.5% gross margin and 27.1% operating margin on a mature LEO satcom franchise — and is the closest hint at where Starlink margins land at maturity. Fourth, no listed peer captures the AI bet: there is no public xAI comparable, and putting Grok next to Iridium would be analytically lazy.

3. Is This Business Cyclical?

SpaceX is not a single cycle — each segment cycles on its own driver, and the three are weakly correlated. Space is exposed to defense-budget and NSSL allocation cycles plus FAA licensing throughput. Connectivity is largely secular (rural broadband demand, terrestrial gaps) with macro sensitivity only on premium-priced terminals. AI is in its first capex super-cycle with secular demand and chip/power-rationed supply.

The most important "cycle" observation: the ARPU compression from $86/mo to $66/mo over 12 months (Q1-25 to Q1-26) is not a downturn signal — it is a mix-shift signal. International markets carry lower price points (often $30–50/mo equivalent), and SpaceX is choosing to grow share over price. Subscribers more than doubled while ARPU fell ~23%, so blended Connectivity revenue still rose ~50%. An investor reading the ARPU line as pricing-power loss is reading the wrong cycle. The real cycle risk is capex pacing in the AI segment — at 397% of segment revenue, FY2025 AI capex would unwind quickly if private capital markets cooled.

4. The Metrics That Actually Matter

The standard A&D ratios — P/E, ROE, EV/EBITDA — barely engage with this business. SpaceX requires a different scorecard built from the few operational signals that explain value creation.

Two underappreciated points. Starlink's adjusted EBITDA margin expanded from 51% to 63% in one year as fixed satellite costs were spread across a near-doubling subscriber base — the most important number on the entire deck and the one most analysts skip. Second, the internal-to-customer launch ratio is the single best leading indicator of where management thinks the cash is — when SpaceX flies its own payloads it is essentially buying Starlink/xAI infrastructure at internal cost. A rising ratio means management is leaning further into Connectivity and AI; a falling ratio means customer launch pricing has gotten attractive enough to redirect cadence outward.

5. What Is This Business Worth?

The right valuation lens here is sum-of-the-parts (SOTP), not a single multiple. Consolidated SpaceX has -13.9% operating margin and -$4.9B net income — multiples on those numbers produce nonsense. The three segments have fundamentally different economic profiles, different peer sets, and different stages of maturity, and the consolidated picture hides every one of them.

The ranges sum to roughly $700B–$1.25T after the holding-company discount — bracketing but generally below the ~$1.5T private mark and well below the reported $1.75T IPO target. At today's marks, you are paying full price for Connectivity, fair value for Space, and an option premium for xAI. That is not necessarily wrong — Starlink could clear $20B in EBITDA within three years, Starship reusability could change orbital economics again, and Grok could reach ChatGPT-scale — but the buyer should know this is a bull-case bundle, not a margin-of-safety price.

6. What I'd Tell a Young Analyst

Anchor on three things and ignore most of the noise. First, watch Starlink subscribers × ARPU monthly — the only line that matters for whether the cash engine compounds. Sub adds slowing or ARPU stabilizing below $50/mo would force the entire thesis to be re-underwritten. Second, watch the internal-to-customer launch ratio quarterly — when it rises, management is reinvesting cadence into Starlink/xAI build (bullish for future Connectivity EBITDA, bearish for current Space revenue), and the directional signal usually leads the segment results by 9–12 months. Third, watch AI capex / AI revenue — it was 397% in FY2025. If that ratio is not moving meaningfully toward 100% by FY2027, xAI stops being an option and becomes a permanent capital drain, and the SOTP discount widens.

Three things the market is probably misreading. (1) Consolidated losses are a feature, not a bug — the $4.9B FY25 net loss is the result of voluntarily spending $20.7B of capex on the AI segment; the underlying ex-AI business is profitable and growing. (2) Starlink ARPU compression is mix, not erosion — international expansion mathematically pulls blended ARPU down even when each market's price holds. (3) The xAI merger was funded with equity issuance plus a $20B bridge loan, not cash — when you own SpaceX you own a diluted claim on the launch + connectivity stack plus a fresh claim on Grok. Pretending the AI segment is "free" because it was paid for in stock is the single most common analytical mistake on this name today.

What changes the thesis: a Falcon 9 or Starship grounding event (most likely thesis-killer in the near term); a Kuiper-led price war that compresses Starlink ARPU below the floor; loss of NSSL Phase 3 share to Vulcan or New Glenn; a regulatory action that limits Starlink Mobile spectrum or constrains xAI in the EU; or a private capital-markets shutdown that forces the IPO at a price-taker valuation. None are base case, but each is plausible enough to warrant tracking, and any one of them moves the SOTP by 20%+ on its own.