Executive summary

SpaceX's mobile network is constrained by cell sites, not by spectrum and not by satellites. That conclusion comes from sizing the network end to end rather than arguing about megahertz, and it inverts the framing most coverage of the 4 August 2026 earnings call has taken.

1. Capacity growth on Starlink is reuse growth, and nothing else. The V3 satellite carries about ten times the capacity of V2 and has 10.7 times the beams. Per-beam throughput actually fell slightly over that generation, from roughly 500 to 488 megabits per second.

2. Reuse from orbit is capped three ways. By beam count rather than antenna size, by a fixed solar power budget divided across however many beams are formed, and by an aggregate emission limit that forces satellites to switch beams off as the constellation grows. SpaceX described the third of these to the Federal Communications Commission (FCC) in its own filings.

3. The 2,048-beam figure is the broadband payload, not the mobile one. The 65 megahertz is reused per cell on the ground and per beam from a mobile payload of roughly 768 beams at 2 gigahertz. The two are not interchangeable, and most coverage of this story treats them as one number. The ground layer delivers roughly two and a half times the capacity of the entire national direct-to-device (D2D) constellation, which is where the commercially relevant reuse happens.

4. The constellation is the cheap part. A 20 million subscriber network, roughly twice the current Xfinity Mobile base, needs about 290 satellites at an estimated 900 million dollars.

5. The site layer is not. That same business needs 37,000 to 62,000 cell sites at 1.9 to 3.1 billion dollars. Roughly 70 to 80 percent of the capital and effectively all of the execution risk sit in the terrestrial layer, and none of it is a space problem.

6. Sixty-five megahertz is sufficient for that business. Twenty million subscribers at 30 dollars a month is near 7.2 billion dollars a year of revenue against a build of 2.8 to 4.0 billion. Spectrum is not the binding constraint at any plausible commercial scale.

7. Sizing the same network against national traffic share gives much larger numbers, and they are a ceiling rather than a target. A quarter of all United States mobile traffic would need roughly 2,400 satellites and 308,000 to 513,000 cells, at 22.7 to 33.0 billion dollars. But a quarter of national traffic is about 165 million subscribers, close to 29 percent of every wireless connection in the country, and no operator is a plausible candidate at that scale. The 20 million subscriber business in findings 4 to 6 is 3 percent of national traffic. The two scenarios differ by a factor of 8, which is why traffic share is the wrong denominator for a business case and appears in this report only to establish the capacity envelope.

7. What the thin holding cannot do is serve heavy users. The median American consumes 14.5 gigabytes a month while the top decile exceeds 52 and accounts for close to 40 percent of all traffic. Twenty-five megahertz of downlink serves the first group well and the second not at all, which makes this a mid-market network by construction rather than by choice.

8. The fastest route to the site layer is somebody else's poles. Cable strand, or the independent small-cell platform spun out of Crown Castle in May 2026, supply pre-permitted mounting points with power and fiber already at them.