The One Variable That Swings Starlink's ROIC From 0% to 27%
I’ve spent the past week building a unit economics model for LEO satellite broadband -- specifically Starlink. The model works mechanically, but the output ranges from capital-destructive (0% ROIC) to genuinely exceptional (27% ROIC). The entire spread comes down to two assumptions I can’t easily resolve from public data.
What the model looks like
The cost stack is built bottom-up from analyst estimates and public filings: manufacturing cost in the hundreds of thousands per satellite, $15-25M internal launch cost per mission (modeled -- SpaceX doesn’t disclose this), 23-29 satellites per Falcon 9 launch. That gives a fully loaded annual cost per satellite of roughly $250K to $500K depending on the scenario, and an annual fleet replacement burden of $2.4B to $4.6B for a 10,000-satellite constellation.
On the revenue side: ~10 million active subscribers at a blended ~$92/month ARPU implies roughly $11B in annual revenue, which cross-checks reasonably against Reuters reporting that SpaceX generated $15-16B total in 2025 with Starlink as the primary driver.
Credit where it’s due -- Quilty Space has done very impressive work and is hosting another webinar in April. Tim Farrar on competitive dynamics, Moffett on Bloomberg Television, Evercore’s work, the Kook Report on ASTS, and Kerri Cahoy’s work at MIT have all shaped the thinking.
The punchline
The model’s ROIC is almost entirely a function of satellite replacement cadence. At a 4-year satellite life with 35% EBITDA margins, Starlink earns roughly 0% on invested capital -- the replacement burden eats the entire profit. At 6 years with 48% margins, ROIC is around 11%. At 6 years with 61% margins, it’s 27%.
The swing variable is not manufacturing cost. It’s not launch cost (probably). It’s how fast you have to replace the fleet. The difference between a 4-year and 6-year satellite life is a ~$2.2B annual swing in replacement spending. That flows straight to returns.
The two questions
1. What is the actual observed average operating life of V2 Mini satellites?
FCC filings reference a 5-7 year life. Moffett referenced 4 years on Bloomberg. Eutelsat management has cited 7 for their constellation. The range between 4 and 6 years matters more than any other assumption in the model.
2. What is the true fully burdened internal launch cost per Starlink mission?
SpaceX’s published commercial price for a Falcon 9 launch is $74M. Most external models use $15-25M for internal Starlink missions, reflecting reusability economics. But how much booster amortization, refurbishment, and overhead should be allocated to Starlink vs. commercial customers? Moving from $17M to $30M roughly halves the base-case ROIC. But none of this accounts for Starship, which could carry several times the capacity of a Falcon 9.
If you can help
If you have a view on either question, I’d appreciate hearing from you. DMs are open.