I came across a post on X this morning by Aakash Gupta that cut through the noise around SpaceX’s mobile ambitions with unusual clarity.
SpaceX spent $19.6 billion on spectrum everyone said was too small to matter. 65 MHz—a fraction of what any one of the big three carriers controls. Then Gwynne Shotwell explained where the cell towers go, and AT&T, Verizon, and T-Mobile fell 2 to 4% in an afternoon.
The plan skips towers entirely. Starlink has 12 million customers with dishes bolted to their roofs, and Shotwell wants to attach a small cellular base station to the same mount. Every rooftop becomes a cell site.
That breaks the carrier cost model in a specific way. The radio was always the cheap part of a cell site. The money goes to land leases, permits, and trenching fiber backhaul to feed each antenna, which is why small cell buildouts have crawled for a decade.
Starlink’s version deletes all three. The site is a customer’s roof. The backhaul is the dish sitting next to the base station, already talking to satellites. And instead of paying rent to host the equipment, the host pays SpaceX $120 a month for internet.
Carriers pay landlords to carry their network. SpaceX’s landlords pay SpaceX.
The satellites cover everything the rooftops miss. Next-gen direct-to-cell satellites launch in 2027, and Shotwell says the upgraded service will be 100x better than the version that already texts your phone in dead zones today.
The strangest seat in the room belongs to T-Mobile. They leased SpaceX the 5 MHz that proved direct to cell works, marketed it as T-Satellite, and taught their own customers to trust the signal. In 2027 the partner files in as the fourth carrier.
What struck me most was the inversion itself. We’ve spent decades treating cell towers as fixed, expensive, hard-to-site infrastructure. The real costs were never the radios. They were the land, the permits, the fiber runs, and the ongoing rent checks. SpaceX is proposing to treat the customer’s roof as the site, the existing dish as the backhaul, and the monthly internet bill as the economic relationship—flipping who pays whom.
It is the kind of elegant deletion that looks obvious only after someone points it out. Most industries get stuck defending the expensive pieces of their stack long after those pieces stop being necessary. Here the expensive pieces are simply removed from the equation.
I don’t know yet whether the density will work in dense urban cores, how indoor coverage will actually feel, or how the regulators will treat residential rooftops as licensed transmitters. Those questions remain open. But the cost-model insight feels durable. Once you see that the landlords can be made to pay the network owner instead of the other way around, the old architecture starts to look like a historical artifact.
Worth sitting with for a while.
