Two Ways to Build a Constellation
The satellite broadband race has produced two opposite answers to one question: who flies your satellites? SpaceX puts its Starlink birds on rockets it builds itself, at cost. Amazon, whose Leo network is the challenger, buys launches from nearly everyone else, including, awkwardly, its direct rival. The contrast says a lot about each company’s position in the launch market.
SpaceX: The Vertical Machine
The internal economics are favorable in a way no customer enjoys. SpaceX’s 2026 price sheet lists a Falcon 9 flight at $74 million for outside buyers, while Elon Musk told Aviation Week back in May 2020 that a reused booster’s marginal cost runs near $15 million in the best case, a figure the company has never updated publicly. Starlink effectively rides those internal margins. A Falcon 9 carries up to 29 V2 Mini satellites per run, and the June 2026 prospectus filed with regulators shows the cadence behind it: 134 Falcon launches in 2024 grew to 165 in 2025, with much of that capacity absorbed by Starlink and the costs capitalized into the satellites rather than billed as launch services. Boosters are qualified for 40 flights each. Starship widens the gap further: Flight 14 reached orbit on September 28, 2026 and deployed 26 V3 satellites, with the prospectus expecting up to 60 per Starship launch. The business this feeds is already large, a Connectivity segment that booked $11.387 billion of 2025 revenue and $4.423 billion of operating income.
Amazon: The Multi-Rider Customer
Amazon Leo, the former Project Kuiper, has no house rocket, so it contracts across the industry: United Launch Alliance’s Atlas V and Vulcan Centaur, Arianespace’s Ariane 6, Blue Origin’s New Glenn and SpaceX’s Falcon 9. The hardware is made at a high-volume plant in Kirkland, Washington, with more than 1,000 spacecraft built and roughly 400 production satellites now operating. The Gen1 architecture targets 3,236 satellites at altitudes of 590, 610 and 630 kilometers, served by three terminal models, a compact 100 Mbps standard unit and an enterprise Ultra antenna rated for gigabit speeds among them.
The Deadline Squeeze
Amazon’s catch is regulatory. Its initial Federal Communications Commission license requires half the Gen1 network, 1,618 satellites, in orbit by July 30, 2026, and the company has filed for extensions citing global launch vehicle shortages while it ramps cadence across its five providers through late 2026. Its commercial counter is bundling: disclosed on October 11, 2026, the plan ties Leo service into Prime subscriptions and the wider retail, streaming and AWS ecosystem to cut subscriber acquisition costs, a deliberate contrast with Starlink’s standalone subscription model. The bet is that distribution beats vertical integration. SpaceX’s bet is the reverse, and it is the one already flying a constellation at internal prices, with the prospectus numbers to show for it.
Sources
- How Much Does a Starlink Launch Cost SpaceX? (2026)
- Amazon Leo Plans Prime Subscription Bundling to Challenge Starlink in Consumer Satellite Broadband Market – SatNews
