The single most important number in the SpaceX investment thesis is Starlink's EBITDA margin. At 70 percent, the $11.8 billion 2025 revenue line produces $8.3 billion of EBITDA — enough to justify, on its own, an enterprise value of $250 billion at 30x. At 50 percent — which is the margin profile of a mature global telco — the same revenue produces $5.9 billion of EBITDA and a $150 billion enterprise value. At 35 percent — which is where competitive pressure from Amazon Leo and Eutelsat OneWeb could push it — the math collapses to $4.1 billion of EBITDA and a sub-$80 billion enterprise value.
The difference between these scenarios is approximately $170 billion of SpaceX SOTP. It is the largest single dispersion in our model. And it is determined by unit economics that no published institutional research has modeled at the level of granularity required to test the margin assumption directly.
This article does that work. We model Starlink at the unit level: per-satellite CapEx by generation, launch cost per satellite at internal transfer pricing, the 5-year amortization curve, ARPU by customer segment (residential / maritime / aviation / mobile / Starshield), churn dynamics, $/Mbps economics versus fiber and GEO incumbents, and the Direct-to-Cell sensitivity tree. The output is a stress test. The 70 percent margin survives some scenarios. It does not survive all of them. The question for institutional investors is which scenarios to weight.
Key Takeaways
- Per-satellite CapEx (model estimate): V1.5 ~$0.5M, V2-mini ~$0.8M, V3 ~$1.3M. Production cost has scaled roughly linearly with capability per satellite.
- Launch cost per satellite (internal transfer): Falcon 9 carrying 23 V2-mini at ~$15-20M total mission cost = ~$650-870K/sat. Starship V3 deployment at ~$40-70M per mission carrying ~30-60 sats = ~$700K-2.3M/sat.
- 5-year amortization model: Total per-satellite landed-in-orbit cost amortizes to approximately $0.30-0.45 per subscriber per month at current subscriber base.
- ARPU by segment: Residential US $120/mo blended global ~$85/mo; Maritime ~$250-5,000/mo (annual ARPU ~$34K); Aviation ~$25K/mo (annual ARPU ~$300K); Mobile/Roam premium; Starshield government rates classified but estimated 2-4x residential equivalent.
- Churn: Estimated ~2% monthly residential US, ~3-4% emerging markets, near-zero maritime/aviation given switching costs.
- $/Mbps: Starlink residential at ~$0.40-0.60/Mbps; fiber at $0.10-0.30/Mbps in served areas; legacy GEO at $5-15/Mbps. Starlink is structurally between fiber and GEO economically.
- Direct-to-Cell: T-Satellite text live July 23, 2025; data live October 1, 2025; voice in beta. Take rate at ~5% of T-Mobile US base could add ~$500M-1B annual revenue at $10-15/sub blended.
- Our base case 2027 margin: 60-65% EBITDA, below Quilty's 70% but above Sacra's 63%. The compression to 60% reflects Amazon Leo competitive pressure beginning to land in 2027.
- Embedded in our hub SOTP at Starlink segment $370B base / $300B bear / $500B bull.
1. The Constellation Today
Before modeling unit economics, ground the operational state.
As of May 8, 2026 (Jonathan McDowell aggregation):
- Active satellites: ~10,358
- Total in orbit (incl. recently launched, pre-commissioning): 10,374
- Total launched lifetime: 11,955
- Deorbited/failed: ~1,580 (~13% of lifetime launches — primarily early V0.9 and V1 satellites at end-of-life)
- FCC authorized ceiling: 15,000 satellites (Jan 2026 ruling adding 7,500 Gen2 spacecraft to the existing 7,500 first-tranche authorization)
- Subscribers (Feb 2026 SpaceX disclosure): 10 million
- Countries served: 110+
Generational composition (model estimate):
| Generation | Approximate count | Capability per sat | Deployment vehicle |
|---|---|---|---|
| V0.9 (60 prototypes) | ~0 (all deorbited) | First commercial; early hardware | Falcon 9 |
| V1 / V1.5 | ~3,500 (most still operational, aging) | ~20 Gbps backhaul, ~30 user beams | Falcon 9 (~60/launch) |
| V2-mini | ~6,200 | ~3-4x V1.5 capacity per sat | Falcon 9 (~23/launch) |
| V3 (initial deployment) | ~650 | >1 Tbps per sat (FCC filings); larger physical form | Starship-class (initial Falcon 9 trial bundles) |
The V3 deployment ramp is the single largest variable in 2026-2028 constellation economics. V3 satellites are physically larger and require either dedicated Falcon 9 missions with reduced satellite counts per launch or — preferentially — Starship deployment to amortize their CapEx efficiently. The Block 3 / V3 Starship transition (IFT-12 onward) is the unlock; until Starship Starlink deployment cadence reaches 10-15 missions per year (target: 2027-2028), V3 ramp is constrained by Falcon 9 mission economics.
2. Per-Satellite CapEx: Model Build
SpaceX does not publicly disclose per-satellite production costs. The numbers in published institutional research range widely — Sacra and Quilty estimate $250K-1M depending on generation; press reporting has cited Elon Musk's targets at "well under $500K" for early generations. We build our own model from disclosed inputs.
V1.5 — $0.5 million estimated unit cost
V1.5 was the workhorse of the constellation from 2021 through early 2023. SpaceX disclosed manufacturing was conducted in Redmond, Washington at production rates approaching 6 satellites per day. At a stated production cost of "well under $500K" per Musk's public commentary, and given the heavy vertical integration (phased array antennas, Hall thrusters, propulsion built in-house), we estimate ~$0.5 million per V1.5 satellite at full-rate production. This figure represents bill-of-materials plus direct labor, excluding facility amortization.
V2-mini — $0.8 million estimated unit cost
V2-mini introduced significantly more capability per satellite — laser links between satellites (eliminating the gateway-hop architecture of V1.5), upgraded phased arrays, more capable Hall thrusters, and approximately 4x the capacity per satellite. The unit cost increase reflects:
- Inter-satellite optical link hardware (~$100-200K incremental per satellite)
- Larger phased array (~$100-150K incremental)
- Increased propellant for orbital insertion at lower deployment altitude
- More capable processors and software-defined radio
Net: ~$0.8 million per V2-mini, with continuing learning curve compression toward $0.6-0.7M as production scales.
V3 — $1.3 million estimated unit cost
V3 is a fundamentally different satellite. Capacity per spacecraft increases to >1 Tbps per FCC filings, requiring substantially more capable phased arrays, larger solar arrays, more propellant, and Direct-to-Cell-capable antennas integrated for the spectrum partnerships (T-Mobile, etc.). Physical mass increases approximately 3-4x versus V2-mini.
The unit cost increase is roughly proportional: ~$1.3 million per V3 satellite at initial production, declining toward $0.9-1.1M as volume scales.
Constellation CapEx through 2028 (model):
| Generation | Approximate count by 2028 | Per-unit cost | Cumulative CapEx |
|---|---|---|---|
| V1.5 (legacy + replacement) | 3,500 | $0.5M | $1.75B |
| V2-mini | 7,000 | $0.7M (blended) | $4.9B |
| V3 | 3,500 | $1.1M (blended) | $3.85B |
| Total constellation CapEx through 2028 | ~14,000 | — | ~$10.5B |
This compares to publicly cited cumulative Starlink CapEx estimates of $10-15B through 2024-2025 (per Sacra, Payload, and press reporting). Our model is consistent with these estimates.
3. Launch Cost Per Satellite: The Internal Transfer Math
The most under-appreciated element of Starlink unit economics is launch cost at internal transfer pricing. Every Starlink launch uses SpaceX's own Falcon 9 boosters, refurbished and reflown, at cost. The cost basis for this internal transfer is fundamentally different from the commercial Falcon 9 price.
Falcon 9 Starlink dedicated missions:
A dedicated Falcon 9 Starlink launch carries 21-23 V2-mini satellites or 50-60 V1.5 satellites. SpaceX's commercial price for a Falcon 9 mission is $65-70 million. The internal cost is materially lower:
- Booster amortization (10+ flight reuse): $5-10M per flight against $30-40M booster manufacturing cost
- Second stage (expendable): $8-12M
- Propellant + range services: $2-3M
- Refurbishment + integration labor: $2-3M
- Approximate per-flight internal cost: $15-20 million
At 23 V2-mini satellites per flight: ~$650-870K per satellite for launch services. Including the $0.7-0.8M unit cost of the satellite itself: landed-in-orbit cost per V2-mini satellite is approximately $1.4-1.7M.
Starship Starlink V3 missions (target operational economics 2027-2028):
Starship can carry significantly larger V3 satellite payloads — initial planning is approximately 30-60 V3 satellites per launch depending on satellite final form factor. At Starship marginal cost of $40-70M per launch (our probable 2027-2028 model, lower than the Musk-marketed $20-30M):
- 45 V3 satellites per mission × $1.1M unit cost = $49.5M satellite CapEx
- Plus $55M launch cost = $104.5M per mission total
- Per V3 in orbit: ~$2.3M landed-in-orbit cost
This is higher than V2-mini per-satellite-in-orbit cost — but the per-Gbps cost is dramatically lower. A V3 satellite with >1 Tbps capacity at $2.3M is approximately $2.3/Gbps, compared to V2-mini at approximately $13/Gbps. V3 deployment economics are 5-6x more capacity-efficient per dollar landed.
This is the structural unlock that makes the marketed Starlink margin expansion case credible.
4. The Amortization Curve
Starlink satellites are designed for approximately 5-year operational life (FCC filings cite 5-7 years for V1.5; V2-mini and V3 are targeting comparable lifecycles with active deorbit at end-of-life). The amortization model for unit economics:
Per-satellite landed-in-orbit cost amortized over 5-year life:
| Generation | Landed-in-orbit cost | 5-year amortization | Monthly per-sat amortization |
|---|---|---|---|
| V1.5 | $1.0M (at V1.5 launch cost ~$0.5M/sat) | $200K/year | ~$16,700/month |
| V2-mini | $1.5M | $300K/year | ~$25,000/month |
| V3 | $2.3M | $460K/year | ~$38,300/month |
Capacity-weighted amortization per subscriber (model):
The constellation today serves 10 million subscribers across approximately 10,358 active satellites. Capacity per satellite varies by generation. We model the capacity-weighted amortization per subscriber per month:
- Average satellite serves approximately 1,000 active subscribers (10M / 10K satellites, with capacity-weighted distribution favoring V2-mini and V3 over V1.5)
- Weighted-average monthly amortization per satellite: ~$24,500 (blended)
- Monthly amortization cost per subscriber: ~$24.50 / 1,000 = $0.30-0.45 per subscriber per month
This is a strikingly low absolute number. The reason is the leverage of subscriber count over fixed orbital infrastructure: every additional subscriber added on top of the existing constellation adds essentially zero satellite CapEx (capacity utilization scales subscribers per satellite within design margins).
The amortization burden compresses further as V3 satellites replace V1.5 — each V3 supports approximately 3-4x the subscribers of a V1.5 at roughly 2.3x the amortization cost. Per-subscriber amortization at full V3 deployment falls to approximately $0.20-0.30/month.
For a $85 blended global ARPU subscriber, satellite amortization is 0.4 percent of revenue. Even at the operating cost line (terminal manufacturing, ground station operations, customer service, R&D), the satellite CapEx is not the binding margin constraint.
This is why Starlink margins are so high. The constellation is a fixed asset that amortizes against a rapidly growing subscriber base. Every incremental subscriber generates near-full ARPU contribution to gross margin.
5. ARPU by Segment: The Real Granularity
The blended Starlink ARPU figure obscures the actual revenue mix. We break it into five operating segments.
Residential
- US standard residential: $120/month (post-2024 adjustment; was $99-110 earlier)
- US "Roam" plan: $150-165/month
- Blended global residential: ~$85/month as international markets in lower-income geographies grow at price points $40-65/month
- Subscriber composition: Approximately 6-7 million of the 10M total (60-70% of base)
- Annual ARPU: $1,020 blended global
- Annual revenue contribution at 6.5M subs: $6.6 billion
- Margin profile: ~65-70% gross margin after customer service, terminal subsidies (terminals are sold near cost), and gateway costs
Maritime
- Pricing: $250/month base for 50 GB priority through $5,000/month for 5 TB high-throughput
- Annual ARPU: ~$34,000 blended across vessel types (yachts, cargo, cruise, fishing)
- Subscriber composition: Approximately 30-50K maritime subscribers (estimates; SpaceX does not publicly disclose maritime subscriber counts)
- Annual revenue contribution: $1.0-1.7 billion
- Margin profile: >80% gross margin — premium pricing with low marginal cost
- Growth dynamic: Among the fastest-growing premium segments; near-zero churn given switching costs
Aviation
- Pricing: ~$25,000/month per aircraft for commercial operators
- Customer base: United, Hawaiian Airlines, Qatar Airways agreements; smaller business jet operators
- Annual ARPU per aircraft: ~$300,000
- Subscriber composition: Approximately 3,000-6,000 aircraft as of early 2026 (combining commercial and business aviation)
- Annual revenue contribution: $0.9-1.8 billion
- Margin profile: >80% gross margin — premium tier
- Growth dynamic: Fastest-growing premium segment as fleet retrofits accelerate through 2026-2027
Business / Enterprise / Mobile
- Pricing: Variable, generally $250-2,000/month depending on bandwidth and SLA
- Customer base: Remote enterprise sites, government agencies (excluding Starshield), nomadic users on premium Roam plans
- Annual ARPU per account: $3,000-24,000
- Subscriber composition: Approximately 200-400K accounts
- Annual revenue contribution: $0.6-1.5 billion
- Margin profile: ~75% gross margin
Starshield / Government
- Pricing: Classified; estimated 2-4x equivalent residential ARPU on per-bit basis
- Customer base: US DoD, NRO (Starshield architecture), Ukraine via DoD contract, allied government communications
- Annual revenue: ~$3 billion in 2025 (Quilty Space estimate)
- Margin profile: Estimated >70% EBITDA margin consistent with defense communications comps
- Growth dynamic: Anchored by $1.8B NRO contract (2021), $537M Ukraine through 2027, and reported $2B Golden Dome award (November 2025)
Total revenue triangulation
| Segment | Estimated 2025 revenue | % of total |
|---|---|---|
| Residential | $6.6B | ~56% |
| Maritime | $1.3B | ~11% |
| Aviation | $1.4B | ~12% |
| Business/Enterprise/Mobile | $1.0B | ~9% |
| Starshield/Government | $3.0B | ~25% |
| Hardware (terminals) | $1.3B | ~11% |
| Total (with cross-segment net) | ~$11.8B | 100% |
(Hardware revenue is largely cost-recovery and is netted against segment costs in EBITDA analysis; we show it separately because it appears in revenue but not materially in margin.)
This composition matters: Starlink is not a residential broadband company. It is a premium-tier infrastructure operator (maritime + aviation + government = ~48 percent of 2025 revenue) that happens to have a large residential subscriber base. The premium tiers carry significantly higher margins than the residential blend, and the premium tiers are where Starlink's competitive moat is most defensible against Amazon Leo and OneWeb.
6. Churn Dynamics: What Survives Competition
Churn — monthly subscriber loss as a percentage of base — determines whether the unit economics maintain their structural profile as competition arrives. Our model:
| Segment | Estimated monthly churn | Annual implied churn | Notes |
|---|---|---|---|
| US Residential | ~2% | ~22% | Pre-Amazon Leo; could rise to 3-4% post-Amazon Leo beta |
| Emerging-market residential | ~3-4% | ~36% | Price-sensitive; higher churn risk |
| Maritime | <0.5% | <6% | Near-zero given switching costs; equipment installed on vessels |
| Aviation | <0.3% | <3.5% | Multi-year fleet contracts; near-zero churn |
| Business/Enterprise | ~1% | ~12% | Mid-range; varies by use case |
| Starshield/Government | ~0% (renewal-based) | n/a | Contract-driven; renewal risk at multi-year intervals |
The institutional reading:
Premium segments (maritime, aviation, government) are structurally low-churn. They represent approximately 48 percent of 2025 revenue and approximately 60 percent of EBITDA contribution. These revenue lines are highly defensible against competition.
Residential is the segment exposed to Amazon Leo's eventual market entry. US residential churn could rise from ~2% monthly to ~3-4% as Amazon Leo enters competitively in 2027-2028, particularly in markets where Starlink price advantage versus terrestrial alternatives is thinner. Emerging-market residential is more exposed to pricing pressure than absolute substitution (Amazon Leo will not deploy aggressively in tier-3 markets until 2028-2029 at earliest).
Net subscriber economics depend on whether gross adds continue to materially exceed churn. At current pace (~19,000 daily net adds in early 2026), Starlink is adding approximately 6.9 million subscribers annually against a base of 10 million. Even at 3% monthly residential churn (vs current 2%), gross adds would only need to fall to approximately 12,000/day to maintain net growth. The bull case is sustained growth through 2028; the bear case is net growth approaching zero by 2028 with churn pressure offsetting slowing gross adds.
7. $/Mbps Economics: The Structural Comparison
The cleanest unit-economics frame for satellite internet is dollar cost per megabit per second of capacity. Three frames matter.
Starlink residential US ($120/month, ~200-300 Mbps median speeds): ~$0.40-0.60/Mbps
Fiber (US municipal / metro markets, $50-80/month, 500-1000 Mbps): ~$0.05-0.16/Mbps
Geosynchronous satellite (HughesNet/Viasat, $70-150/month, 25-100 Mbps): ~$0.70-6.00/Mbps
Amazon Leo (target pricing not finalized, but reported $50-80/month for residential): ~$0.10-0.32/Mbps if speeds reach 250 Mbps
OneWeb (enterprise/maritime focused, $400-2000/month equivalent enterprise tier): ~$0.20-2.00/Mbps
The institutional reading:
Starlink is structurally between fiber and legacy GEO satellite economics. Where fiber is deployed and competitive, fiber wins on $/Mbps. Where fiber is not deployed (rural, maritime, aviation, mobile), Starlink wins on $/Mbps versus all satellite alternatives.
The Amazon Leo question is whether Amazon can deploy at scale fast enough to compete with Starlink on cadence-served markets (where fiber is not present). Amazon has the launch capacity (Project Kuiper has signed launches across ULA, Arianespace, and SpaceX itself), the capital depth, and the consumer brand. What Amazon does not have is the four-year head start Starlink has on subscriber acquisition, the operational maturity of orbital deployment cadence, or the vertical integration on launch costs that makes Starlink's marginal cost structure defensible.
For 2026-2028 modeling, we assume Starlink's $/Mbps advantage versus Amazon Leo erodes by 20-30 percent as Amazon scales but remains structurally favorable through 2028. Beyond 2028, the competitive dynamic becomes harder to model.
8. Direct-to-Cell: Take Rate × ARPU Sensitivity
Direct-to-Cell is the most consequential new revenue line in Starlink's 2026-2028 trajectory. It deserves its own unit-economics treatment.
Operational state (May 2026):
- Direct-to-Cell satellites launched: 650+ (per Starlink)
- T-Satellite (US, via T-Mobile partnership) commercial launch: July 23, 2025 (text messaging)
- Data services rollout: October 1, 2025 (initial dozens of phones)
- Voice service: beta late 2025, broader rollout targeted 2026-2027
- International coverage: 22 countries live as of early 2026
Revenue model:
D2C revenue does not flow to Starlink as a per-subscriber subscription. It flows as wholesale revenue from mobile network operators (MNOs) like T-Mobile, Optus (Australia), Rogers (Canada), and others — typically structured as wholesale capacity payments or revenue-share arrangements on subscriber tiers that include D2C connectivity.
Take-rate sensitivity model:
Assume T-Mobile US has approximately 130 million postpaid subscribers as of mid-2026. T-Mobile's D2C value proposition is positioned as a premium feature, particularly for subscribers who travel outside terrestrial coverage. Take rate scenarios:
| Take rate | T-Mobile US D2C subscribers | Implied wholesale per-sub revenue to Starlink | Starlink annual D2C revenue (T-Mobile US only) |
|---|---|---|---|
| 2% take rate | 2.6M subs | $5/mo (low-end MNO share) | $156M |
| 5% take rate | 6.5M subs | $10/mo (mid MNO share) | $780M |
| 10% take rate | 13M subs | $15/mo (premium tier) | $2.34B |
| 20% take rate | 26M subs | $15/mo | $4.68B |
Adding the other 21 international MNO partners (Optus, Rogers, Salt in Switzerland, KDDI in Japan, etc.), the total addressable wholesale revenue at 5% blended take rate is approximately $1.2-1.8 billion annually by 2027-2028 — material to Starlink's $20B 2026 projected revenue but not transformational at this take rate.
The bull case scenario is 10-20% take rate by 2028, particularly as voice service rolls out and creates an "always-connected" value proposition. At 10% blended take rate across global MNO partners, D2C wholesale revenue could reach $3-5 billion annually, contributing approximately 15-20 percent of Starlink's projected 2028 revenue.
The institutional reading:
D2C is real revenue with a real near-term contribution. But the institutional question is what take rate to model. We assume 5-7% blended take rate by 2028, contributing approximately $1.5-2.5 billion in annual revenue. This is upside relative to the Quilty Space 2026 revenue projection of $20 billion (which already includes a modest D2C contribution).
The bear case for D2C is that consumer take rate proves disappointing at modest premium pricing. Most mobile-network subscribers do not regularly travel outside terrestrial coverage; the value proposition has natural ceiling. We assign 30-40% probability to D2C revenue underperforming the 5% take-rate assumption.
9. Regional Capacity vs Demand
One under-appreciated risk in Starlink unit economics is regional capacity allocation. The constellation has finite capacity per geographic cell, and high-density markets can become capacity-constrained.
Demand concentration:
- North America: Approximately 35-40 percent of Starlink subscribers; highest density per square mile
- Europe: Approximately 15-20 percent
- Latin America (especially Brazil, Mexico): Approximately 10-15 percent
- Asia-Pacific: Approximately 10-15 percent
- Africa: Approximately 5-10 percent (rapidly growing)
Capacity allocation:
Starlink capacity per geographic cell scales with satellite count over that cell and bandwidth allocation per beam. In high-density markets, oversubscription can result in:
- Speed throttling during peak hours
- Wait lists for new subscribers (active in selected US suburban markets)
- Reduced "priority" tier availability
The V3 transition is the unlock for capacity expansion. V3 satellites with >1 Tbps per-spacecraft capacity provide approximately 4-5x the per-cell capacity of V2-mini. As V3 deployment ramps through 2026-2028, capacity-constrained markets ease, and unit economics improve through reduced marginal-capacity investment.
For investor modeling, the regional capacity dimension matters because:
- Oversubscription in high-density markets creates churn risk if subscribers experience degraded service
- Capacity additions are timed to V3 deployment cadence, which depends on Starship Block 3 operational success
- Markets like Brazil and India could be early V3-capacity targets given their rapidly growing subscriber bases
10. The Unit Economics Stress Test
Bringing all dimensions together: what scenarios survive, and which break the 60-70% EBITDA margin?
Stress test 1: Amazon Leo enters competitively in 2027
- Assumption: Amazon Leo launches commercial residential service at ~$50/month for 250 Mbps in 25-50 major US/EU markets by late 2027
- Starlink response: Reduce US residential pricing 15-20% to ~$95/month; blended global ARPU drops from $85 to ~$70
- Subscriber impact: Lower gross adds (slowed by competitive pricing); modestly higher churn (2.5-3% monthly)
- Revenue impact: 2028 residential revenue grows but slower; total Starlink 2028 revenue approximately $18-19B vs $25B+ in the optimistic case
- Margin impact: EBITDA margin compresses to approximately 58-62% (from 70% Quilty baseline)
- EBITDA result: ~$11-12B in 2028 vs $14B+ in the optimistic case
Net SOTP impact: Starlink segment value compresses from $500B (bull) to $320-360B at 25-30x EBITDA. Approximately $140-180B of fair-value compression.
Stress test 2: Starship V3 deployment delayed by 18 months
- Assumption: Starship Block 3 development extends; V3 satellite deployment cadence in 2027-2028 is half of base-case rate
- Starlink response: Continued reliance on V2-mini deployment for capacity (Falcon 9 missions); slower V3 ramp
- Capacity impact: Regional oversubscription persists in high-density markets through 2028; some growth deferred
- Margin impact: Modest compression of approximately 200-300 bps as capacity-constrained markets become churn risks
- Net SOTP impact: $40-60B of segment-value compression. Material but bounded.
Stress test 3: Direct-to-Cell take rate disappoints
- Assumption: Global D2C blended take rate plateaus at 2-3% by 2028 rather than the 5-7% base case
- Revenue impact: D2C contribution of approximately $400-700M annually vs base-case $1.5-2.5B
- Margin impact: Minimal — D2C is a small contributor at this take-rate range
- Net SOTP impact: $5-15B of segment-value reduction. Modest.
Stress test 4: Spectrum or regulatory disruption
- Assumption: ITU spectrum coordination dispute, major national-level service ban (Russia/China extension to additional markets), or US FCC license modification
- Impact: Highly variable; could range from minor (market-specific capacity reduction) to material (multi-country service interruption)
- Net SOTP impact: Difficult to model precisely; bounded at perhaps $30-80B of fair-value impact in moderate scenarios
Stress test composite
- Single stress test (Amazon Leo): -$140-180B
- Two simultaneous stress tests (Amazon Leo + Starship delay): -$180-240B
- Three simultaneous stress tests: -$220-300B
- All four stresses concurrent: -$280-400B
The stress test composite explains why our Starlink segment bear case is $300B (down from $500B bull): we are pricing in approximately one moderate stress event (Amazon Leo competitive pressure) plus partial weighting of secondary risks.
The institutional question is which stress events to weight at material probability. Our base case assumes (a) Amazon Leo enters in 2027 but with limited competitive impact on premium segments, and (b) Starship V3 deployment proceeds on a reasonable timeline. Both of these are bull-case-adjacent assumptions.
11. What We're Watching
Six leading indicators for 2026-2028:
- Starlink subscriber pace. Daily net adds were approximately 19,000 in early 2026. Sustained pace through Q3 2026 would confirm residential growth durability ahead of Amazon Leo competitive entry.
- V3 deployment cadence. Starship-class V3 deployments are the critical capacity unlock. First Block 3 deployments are tied to IFT-12 (May 19, 2026 window) success and follow-on operational cadence.
- Amazon Leo beta launches. Amazon has reported targeting commercial beta launches in 2026-2027. The geographic markets and pricing tier of first commercial offering will materially inform Starlink's competitive response.
- D2C take rate disclosure. T-Mobile and other MNO partners report subscriber metrics quarterly. T-Satellite take rate visibility will emerge through 2026-2027.
- Starshield revenue trajectory. Any new NRO or Golden Dome contract disclosures would materially affect the government-segment revenue line within Starlink unit economics.
- ITU and FCC regulatory actions. Spectrum coordination disputes, market access decisions, and FCC modifications to the 15,000-satellite ceiling all affect the long-term constellation economics.
12. The Bottom Line
Starlink's unit economics are defensible at the 60-70 percent EBITDA margin in 2025 because:
- Per-subscriber satellite amortization is approximately $0.30-0.45/month — a near-trivial portion of $85 blended ARPU
- Internal-transfer launch pricing on Falcon 9 dramatically reduces deployment costs versus what any competitor pays on commercial launch services
- The premium tiers (maritime, aviation, government) are structurally low-churn and contribute disproportionately to absolute EBITDA
- Vertical integration across launch, satellite manufacturing, terminal manufacturing, and ground operations captures economic value that no competitor can replicate
The unit economics are exposed in 2027-2028 if:
- Amazon Leo enters competitively at material scale and forces residential pricing compression
- Starship V3 deployment delays persist and constrain capacity expansion
- Direct-to-Cell take rate proves materially below the 5-7% base case
- Regulatory pressure in major markets compresses revenue or operational economics
We model the base case at 60-65 percent EBITDA margin by 2028 (down from Quilty's 70 percent 2026 estimate, up from Sacra's 63 percent 2025 estimate). At 2028 revenue of $24-28 billion, this implies $14-18 billion of segment EBITDA. At 22-28x EBITDA, this implies a segment value of $310-500 billion — consistent with our hub SOTP range.
For institutional readers, Starlink unit economics survive most realistic scenarios. The 70 percent EBITDA margin compresses to 60 percent in our base case, but the segment value range remains substantial. The bear case ($300B Starlink segment value) requires multiple simultaneous adverse scenarios; we assign approximately 20 percent probability.
The bull case ($500B Starlink segment) requires Amazon Leo to underperform, Direct-to-Cell to outperform, and Starship V3 deployment to proceed on schedule. We assign approximately 25 percent probability.
Read this article in context with the hub SOTP analysis, the valuation and secondary markets analysis for the broader SOTP framework, the Starship economics piece for the V3 deployment unlock detail, and the existing Starlink business overview for the editorial context on growth and competitive landscape.
Sources & Methodology
Primary disclosure sources:
- Starlink Direct-to-Cell service brochure
- FCC Gen2 7,500-satellite authorization
- T-Mobile T-Satellite data ready announcement
- SpaceX 10 million subscribers announcement (Feb 2026 via Broadband Breakfast)
Constellation tracking:
Financial estimates:
- Sacra SpaceX coverage
- Quilty Space Starlink Financial Overview 2025 2H / 2026 Forecast
- Payload — Estimating SpaceX's 2024 Revenue
- SpaceNews — Starlink set to hit $11.8B revenue in 2025
- Advanced Television — SpaceX tracking to $20B revenue in 2026
Methodology notes:
Per-satellite CapEx estimates ($0.5M / $0.8M / $1.3M for V1.5 / V2-mini / V3) are model-derived from disclosed inputs and publicly cited bill-of-materials commentary. SpaceX does not disclose per-satellite production costs. Launch cost per satellite (internal transfer) is model-derived from estimated booster amortization, second-stage costs, and propellant + range operations costs; these reflect best estimates and may differ from SpaceX's internal accounting. ARPU figures combine publicly listed Starlink pricing with third-party analyst estimates for international blended ARPU. Take-rate sensitivity for Direct-to-Cell is illustrative and reflects the absence of disclosed take-rate data from T-Mobile or other MNO partners. Churn estimates are model-derived from comparable LEO satcom operator data and limited publicly disclosed Starlink churn commentary.
This article will be updated when material new operational metrics are published, when Amazon Leo enters competitive service, or when SpaceX/Starlink discloses more granular unit economics through the Project Apex S-1. Recency stamp: May 13, 2026.
Frequently Asked Questions
Is the 70% EBITDA margin on Starlink real? We think the 70% margin Quilty Space estimates for 2026 is plausible but at the upper end of defensible range. Sacra estimates 63% for 2025. Our base case for 2028 is 60-65%, reflecting modest compression from Amazon Leo competitive pressure. The premium tier mix (maritime + aviation + government = ~48% of 2025 revenue) supports a structurally higher margin than a residential-only operator would carry.
What is the most important unit economics variable? ARPU compression risk in the residential segment. Amazon Leo's eventual competitive entry creates the largest single risk to Starlink's margin structure. Premium segments (maritime, aviation, government) are structurally insulated and continue to drive margin even if residential ARPU compresses.
How does Starship V3 deployment change the unit economics? V3 satellites have approximately 4-5x the per-spacecraft capacity of V2-mini. Per-Gbps cost in orbit drops from approximately $13/Gbps (V2-mini) to approximately $2.3/Gbps (V3). This is the structural unlock that supports continued margin expansion as the constellation transitions to V3-dominant capacity through 2027-2028.
Are Direct-to-Cell economics meaningful? At 5-7% blended take rate across MNO partners by 2028, D2C contributes approximately $1.5-2.5 billion in annual revenue — material but not transformational. At 10-20% take rate (bull case), D2C could contribute $3-5 billion annually, becoming a meaningful component of total Starlink revenue.
How does Starlink's churn compare to terrestrial broadband? US terrestrial broadband churn is approximately 1.5-2% monthly. Starlink US residential churn is estimated at approximately 2% monthly currently, slightly elevated versus terrestrial alternatives. Premium segments (maritime, aviation) are well below 1% monthly. Emerging markets are higher at 3-4%.
Can Amazon Leo realistically compete on Starlink's economics? Amazon Leo has substantial advantages (capital, brand, AWS infrastructure) but faces structural disadvantages on launch costs (must pay ULA, SpaceX, or Arianespace for deployment) and four-year head-start gap on subscriber acquisition. We expect Amazon Leo to compete effectively in residential markets where pricing pressure can be applied, but to struggle in premium segments where Starlink's operational moat is most defensible.
Is the V3 satellite count of 15,000 a hard ceiling? The FCC's January 2026 ruling authorized an additional 7,500 Gen2 satellites, bringing the total authorized constellation to 15,000. SpaceX has applied for additional authorizations beyond this; future FCC and ITU approvals are required for further constellation expansion. The 15,000-satellite ceiling is sufficient for capacity needs through approximately 2030 based on our subscriber projections.
Disclosure: This article is for informational and educational purposes only and does not constitute investment advice. SpaceX is a private company; all financial figures presented here are estimates derived from third-party analyst reports, public regulatory filings, and our model assumptions unless explicitly cited to a primary SpaceX or government disclosure. Past performance does not predict future returns. Authors and SpaceOdysseyHub have no economic position in any secondary-market vehicle, pre-IPO fund, or publicly-traded security referenced. Consult a licensed financial advisor before making capital-allocation decisions in private or pre-IPO securities.
