Why anyone is doing this, and what it costs

Each business model has a different financial shape

Straight from SEC filings, not from decks.

1 Globalstar tags only PaymentsToAcquireOtherPropertyPlantAndEquipment, so this is not comparable with the other two columns. Its FY2025 net cash used in investing activities was $550.4M, which is the better proxy for what it is spending — see part 6.

Filed SEC XBRL company facts (data.sec.gov), retrieved 17 Sep 2026. Revenue, capital expenditure, cash, R&D and net income read from each company’s filed 10-K and 10-Q tags.

Read across those columns and three different companies appear, not three stages of the same one. Iridium is the profitable incumbent: $871.7M of revenue in FY2025, the only positive net income on the page, and a mature constellation maintained on $100.3M of capital expenditure. Globalstar is a wholesale business of another kind again: $273.0M of revenue, near break-even, R&D of only $6.1M, and spectrum it already owns rather than rents. AST SpaceMobile is neither of those things: it is a construction project, spending $1,064.7M in a single year against $70.9M of revenue as filed in its FY2025 Form 10-K, and funded by a cash balance of $2,335.7M roughly the size of two years of that spend.

AST’s capital expenditure is the line that moves fastest: $174.1M in 2024, $1,064.7M in 2025, and $859.2M in the first half of 2026 alone.

That is a company mid-build, and the build is the strategy rather than a preliminary to it. The rest of AST’s column follows from that one fact: the loss is what construction on this scale produces, the $28.1M of R&D is dwarfed by the hardware spend beside it, and the $2,335.7M of cash matters chiefly as a measure of how long the build can continue at the current rate. Revenue of $70.9M set against capital expenditure of $1,064.7M is not a margin problem. It is a company that has not yet arrived at the point where margin is the relevant question.

The contrast that matters here is therefore not one of size. Iridium and Globalstar are being run; AST is being built. That difference decides which questions are worth asking of each of them: margin, retention and renewal for the two operating businesses, and, for the third, how much construction is left and what pays for it.

One name is missing from the table, and its absence is structural rather than accidental. SpaceX is private and files nothing, so the direct-to-cell economics of the largest constellation in the market are invisible: no revenue line to compare, no capital expenditure to set beside AST’s, no cash balance against which to measure a burn rate. The temptation is to fill that space with an estimate, and it should be resisted, because the estimate would be the only unfiled number in an argument built entirely on filings. Any market model that treats all four operators as comparably knowable is wrong at the foundation; the version that holds carries three measured companies and one acknowledged blank.

Why rent the spectrum instead of buying it

Part 1 showed that AST and SpaceX hold leases rather than licences, and that neither owns the spectrum its consumer service depends on. Taken alone that reads as a weakness; taken against the capital expenditure line above, it reads as the only route available. Three reasons hold the arrangement in place, and all three are financial rather than technical.

The first is crowding out. American 700 MHz licences of the sort AST leases were sold at auction for sums that would consume the balance sheet of a company already spending $1B a year on hardware, which makes buying spectrum and building satellites alternatives rather than two lines in the same budget. A lease keeps the capital in the constellation, which is where the commercial case sits.

The second is optionality. A lease is reversible and a licence purchase is a commitment, and for a service that has not yet proven consumer demand, the ability to reverse the decision is worth real money. Renting turns an irreversible bet on a market into a recurring cost that can be stopped.

The third reason belongs to the other side of the table, and it explains why the carriers agree to any of this. The carrier lends spectrum it already owns, adds coverage it cannot economically build, and bills its own subscriber, while the satellite operator gets reach without acquiring customers one at a time. Each side supplies the part of the business it is better placed to supply, which is why these arrangements are structured as wholesale rather than retail. It is also, precisely, the constraint that comes with them: the satellite operator’s revenue depends on a partner who owns both the spectrum and the customer relationship.

Globalstar shows the same logic working from the opposite end. It owns its spectrum outright and still earns most of its revenue wholesaling capacity to a large consumer partner rather than selling retail itself, which is part of why its accounts show substantial revenue and almost no margin. Owning the spectrum removes the landlord; it does not, by itself, change who owns the customer. Both it and Iridium have since agreed to be acquired, by Amazon and Rocket Lab respectively; part 6 has the terms.

The date on the calendar

The tenancy is not open-ended, and the dates attached to it are unusually exact. All 1,197 of AST’s lease authorizations were granted on one day, 21 April 2026, and every one of them expires on one day, 9 May 2028. A footprint assembled in a single administrative act comes up for renewal in a single administrative act, which puts the whole of the American business on one calendar entry.

Whatever gets built on that footprint has roughly two years to prove itself before commercial terms are renegotiated, and the counterparties in those negotiations are the same carriers whose customers the service serves.

Filed FCC market licence file, snapshot 13 Sep 2026. 1,197 active AST lease records, all granted 21 Apr 2026, all expiring 9 May 2028.

For anyone modelling this business, that date is an input rather than a footnote. It bounds the window in which usage, pricing and retention evidence has to be generated, and anything meant to improve the renewal terms has to exist before it. The alternative is to stop being a tenant altogether, which carries a price of its own: part 4 works out what it would cost to own the spectrum instead.