Business

Know the Business

Bottom line. SpaceX is three businesses stapled together: a cash-generative LEO broadband utility (Connectivity), a sub-scale launch business spending its margin on Starship, and a deeply loss-making frontier AI business (xAI/Grok) that consumes more capex than the other two combined. The IPO target of ~$1.75 trillion (~94x trailing sales, ~266x trailing EBITDA) puts the burden of proof on AI and Starship, not Starlink.

1. How This Business Actually Works

The engine is a one-way flywheel: reusable launch lowers cost-to-orbit, which fills the sky with Starlink satellites, which collect monthly subscription cash, which (along with debt and equity raised against the future option value) funds Starship and the AI build. Each segment has fundamentally different unit economics — and the consolidated income statement obscures all three.

The revenue model in plain English.

• Launch services sells Falcon 9 / Falcon Heavy seats per kilogram to commercial and government customers. Reusability turned the rocket into an aircraft: at 34 reflights per booster, the marginal cost of an additional launch is propellant, second-stage hardware, range fees, and refurb. NASA cites Falcon 9 at roughly $2,700 per kilogram to LEO vs the $18,500/kg historical industry average — an 85% cut. That cost curve is the source of every advantage downstream.
• Internal launch is "free." When a Falcon rocket carries Starlink satellites, no inter-segment revenue is recorded; the launch cost is capitalized into satellite PP&E and depreciated through Connectivity's COGS. That makes reported Space revenue an under-statement of true launch economic activity, and Connectivity's reported margin a fair representation of true unit economics — depreciation already includes the launch.
• Starlink sells $66–$81/month broadband subscriptions plus a (typically subsidized) ~$200–$500 terminal. Once a satellite is in orbit and a region is "lit," the marginal cost of an additional subscriber is terminal manufacturing, payment processing, and ground-network operating cost. Network capex is largely amortized as you add subs.
• Government and national security sells multi-year fixed-price contracts (NSSL Phase 3 Lane 2: $13.7B / 54 missions, 2025–32) plus crewed and cargo missions to NASA. Reliability gates this revenue — one failure can halt the manifest.
• xAI / Grok sells advertising on X, Grok and X Premium subscriptions (6.3M paid as of Mar-2026, of which 1.9M are SuperGrok), API access, data licensing, and increasingly compute capacity. The economic engine is cost per token, which is a function of compute capex amortization, power, and model efficiency.

The cost stack is doing something unusual. Cost of revenue grew 18% in FY25 against 33% revenue growth — operating leverage in the established businesses is real. But R&D grew 150% (to $8.6B), capex grew 86% (to $20.8B), and depreciation grew 75% (to $6.7B). The company is not optimizing for current-period profit. Management explicitly says so in the S-1: they prioritize "execution speed, capacity expansion, and technological leadership over near-term margin optimization." Read every quarter with that lens; treat any single year of GAAP income as a residual of capital-allocation choices.

The chart is the entire story of the consolidated business. Operating cash flow scales steadily (+50% over two years). Capex scales violently (+371% over two years). The gap — funded by $19B of stock issuance and $16B of new debt in FY25 alone — is what AI and Starship are absorbing. Whether that consumes shareholder value or creates it is the real investment question.

2. The Playing Field

No listed company is a true peer, and that single fact is the most useful thing about the peer set. The five US-listed companies the S-1 itself names — RKLB (launch), VSAT (GEO broadband), IRDM (LEO mobile satellite services), GSAT (direct-to-cell with Apple SOS), and ASTS (direct-to-cell pure-play) — each cover one of SPCX's product lines, none cover more than one, and none operate at anywhere near SPCX's scale.

What the peer set actually reveals.

• In launch, SpaceX is not competing — it is the market. Per the S-1, SpaceX put more than 80% of the world's mass to orbit in 2023–25 and completed 11 of 12 NSSL missions in 2025. RKLB is the only listed challenger and its $41B market cap is for Neutron (a Falcon 9 competitor that has not yet flown) plus a small-lift business that is one-fifteenth of SPCX's launch revenue. ULA, Blue Origin, Arianespace are private.
• In LEO broadband, every other LEO operator is years behind. Amazon Kuiper has not reached commercial scale; Eutelsat OneWeb is a fraction of Starlink's subscriber base; Telesat Lightspeed and Blue Origin's TeraWave are still being built. The closest listed analogue is IRDM (mature LEO operator, 51% EBITDA margin, 4x EV/Sales). Starlink Connectivity at 63% Segment EBITDA margin and 50% YoY growth would command higher multiples than IRDM if disclosed standalone — what level the market actually assigns is the open question.
• In direct-to-cell, SpaceX is a late entrant but the spectrum-richest one. GSAT and ASTS were first to commercial; SpaceX's planned $19.6B EchoStar spectrum purchase (close expected Nov 2027) reframes the competitive landscape into one where spectrum, not satellite engineering, is the bottleneck — and SpaceX is buying the bottleneck.
• In AI, no listed company is comparable. OpenAI and Anthropic are private. The hyperscalers are conglomerates. The closest economic analogue is xAI's own most recent private round (which the market will need to triangulate from secondary trades and the Cursor compute agreement that the S-1 prices at an implied Cursor equity value of $60.0B).

3. Is This Business Cyclical?

SpaceX is structurally secular, not cyclical, but it has three cycle exposures that look nothing like a typical industrial. Macro recession barely matters. What matters are (1) the engineering cycle (Starship dev → V3 satellites → AI compute satellites), (2) the regulatory cycle (FAA launch licenses, FCC spectrum dockets, NSSL recompetes), and (3) the capital-markets cycle (because the company is running at negative $14B FCF and depends on equity and debt issuance to fund the AI build).

The most underappreciated point is that Starlink ARPU is already declining — $91 (FY24) → $81 (FY25) → $66 in the most recent quarter, a 28% drop in 15 months. Subscribers grew 105% YoY in the same quarter, so revenue and Segment EBITDA still surged. But subscribers cannot keep doubling forever, and once growth normalizes the ARPU trajectory matters enormously. Management frames this as "international mix + lower-priced plans," which is true; the question is whether the developing-world ARPU is profitable at scale or merely subsidized to lock in spectrum and footprint.

4. The Metrics That Actually Matter

Forget P/E. SPCX reported a $4.9B GAAP net loss on $18.7B revenue in FY25; on $1.75T equity that is meaningless. The metrics that move value here are operational, segment-level, and unit-economic.

Why these and not the usual ratios: ROE is meaningless for a company that just absorbed $19B of equity issuance and is amortizing $32B of accumulated deficit; gross margin is mostly a function of how much launch cost is capitalized into Connectivity satellites; revenue growth alone hides the segment mix shift that is the actual story.

5. What Is This Business Worth?

This is genuinely a sum-of-the-parts business and should be valued as one. The three segments have different economic engines, different growth profiles, different capital needs, and require completely different valuation lenses. Forcing a single multiple onto consolidated numbers will systematically misprice it.

Why the consolidated multiple is misleading. At an IPO target of approximately $1.75T equity, plus ~$1.9B of net debt, the implied EV-to-sales is ~94x and EV-to-EBITDA is ~266x. Both numbers are roughly meaningless. Strip out AI and apply a Starlink-appropriate multiple to Connectivity alone — say 25–40x Segment EBITDA — and you arrive at $180–290B for Connectivity. Apply a launch-services multiple to the non-Starship part of Space (10–15x on a normalized $3.5–4.5B EBITDA once Starship R&D rolls off) and add $35–70B. That leaves anywhere from $1.4T to $1.55T of the IPO target valuation that must be assigned to (a) AI / xAI, (b) Starship as a real option, and (c) the orbital AI compute thesis. Whether $1.4T+ is the right number for those three optionalities together is the actual valuation debate.

What would support a premium. (1) Starship commercial operation in 2H 2026 on time, with V3 satellites scaling Starlink capacity 20x per launch; (2) AI Segment EBITDA turning positive within 36 months on a clear path to $10B+ of run-rate cash; (3) regulatory close on the EchoStar spectrum unlocking US 5G satellite-to-handset at scale.

What would argue for a discount. (1) Starship V3 slip into 2027 or later; (2) AI capex continuing to outgrow AI revenue (FY25 capex/revenue is 4x); (3) Amazon Kuiper or another LEO competitor reaching commercial scale and forcing Starlink ARPU compression in mature geos; (4) regulatory pushback on D2C spectrum or NSSL recompete losses.

6. What I'd Tell a Young Analyst

Treat SpaceX as three different companies stapled together, not as "the rocket company that also does Wi-Fi and AI." The single biggest mistake new analysts make on this name is benchmarking it against defense primes (wrong margin profile, wrong capital intensity) or against pure SaaS (no satellite capex). Build your model bottom-up by segment, project capex separately by segment, and never let a consolidated number stand alone in a thesis.

Watch four things in every filing. (1) Starlink Subscribers and ARPU mix — the subscription cash machine that funds everything else. (2) Space Segment R&D — the level of Starship spend tells you when the cost curve flips. (3) AI segment capex and operating loss — the magnitude tells you how long the company can fund itself before requiring more equity. (4) Stock-based compensation — at $1.9B in FY25 (up 148% YoY) it is becoming a material economic cost, especially against a $1.75T cap.

Where the market read may be off. First, Starlink alone supports a defensible enterprise value of $250–300B+ once Connectivity ARPU stabilizes and Mobile/D2C add incremental margin — a frame that may be underweighted in consolidated multiples. Second, orbital AI compute is being treated as a base-case asset rather than a long-dated option; the thesis depends on Starship commercialization, V3 satellite economics, and a 2028+ deployment timeline, none of which are derisked. Own the moated core on cash earnings, treat Starship as a real option, and underwrite AI separately.

What would change the thesis. A failed Starship V3 program (kills the cost curve and the AI satellite thesis). A 30%+ ARPU collapse in mature Starlink geos with no offsetting volume (kills the cash engine). A material loss of NSSL share to ULA or Blue Origin (signals the reliability moat is closing). On the upside, a positive AI Segment EBITDA print within 24 months would compress the implied AI discount in the SOTP.