Rent or own: what spectrum costs

What does it cost to stop being a tenant, who is doing it, and what happens to the landlord when the tenant becomes a competitor?

A lease is not a small version of a licence

A lease is fast, cheap and reversible, which is a large part of why both operators hold one. It is also derivative: it is written against someone else’s licence, in someone else’s markets, for a term that cannot outlast theirs. A licence is the opposite on every count. Expensive and slow, but yours: national in scope if you buy it that way, renewable, financeable, and sellable. For a service that has to work everywhere, that difference does not colour the business model; it decides it.

What the tenancy looks like in the record

What the distinction looks like in practice is visible in the FCC’s market-based licence file, filtered to active records and to licensee-type entities. Neither operator holds a licence for the spectrum its consumer service runs on: every one of AST & Science’s records is a lease, and all but two of SpaceX’s are as well. The bands, the markets, the landlords and the dates are all on the record.

Filed FCC market-based licence file, snapshot 13 Sep 2026, filtered to active records and to licensee-type entities

SpaceX’s expiry date in that last row is out of step with the rest of the G block. Filter the same file to every active G-block record and a pattern appears: T-Mobile License LLC, Nextel West, Trace-Tek, SDF and Tecore all show an expiry of 3 March 2036, most of them re-granted during March 2026, and only SpaceX’s 173 lease authorizations still carry 3 March 2026, the pre-renewal term. We cannot see the file jackets, so we will not tell you what it means; it may be nothing more than administrative lag behind a renewal, and we report the discrepancy as it appears in the file without having confirmed its cause. But it illustrates the structural point exactly, and it is checkable in an afternoon: a lease hangs from a licence, and when the licence moved, the leases did not move with it.

What owning costs in one comparable unit

Price per MHz per person covered, dollars per MHz-POP, is the only way to compare a 2020 auction with a 2025 private deal. The national-equivalent figure divides the total by the bandwidth and then by the US population of 340,110,988, which puts every transaction on one axis whatever its actual licence footprint. The chart and the table below place four American auctions and the two EchoStar disposals on that axis.

Auction totals are gross winning bids. The Auction 113 row is marked not comparable because it sold leftover licences in selected markets on a different basis, and it is excluded from the chart above.

Two things fall out of that table. Paired spectrum went at almost exactly a dollar per MHz-POP, and the unpaired, uplink-only block went at half that. The largest private spectrum purchase in this market, meanwhile, cleared within 4% of the C-band auction, a useful check the next time someone calls a spectrum position mispriced.

Now put AST’s tenancy against those benchmarks. Buying its 24 MHz nationwide would cost roughly $5.4B to $7.2B at the prices above, and low band has historically traded at a premium to mid band, so treat that as a floor rather than an estimate. That is the sum the company would have had to find before a single satellite was built.

It did not have it, and the filings say so without interpretation. AST closed FY2025 with $2,335.7M of cash while spending $1,064.7M a year on the constellation, a rate of spend set out in part 2. Buying was never an option, and renting was the only way the business existed at all.

The tenant is buying the room next door

SpaceX’s agreements with EchoStar are on the public record in EchoStar’s own filings rather than in press coverage, and the detail in them goes well beyond the headline number. Four items define the position SpaceX has contracted to hold, and their sequence matters as much as the total.

• On 7 Sep 2025, an agreement to buy 50 MHz (2000–2020, 2180–2200, 1915–1920 and 1995–2000 MHz) for $17B, transferred in two steps: first into Spectrum Business Trust 2025-1, then to SpaceX. SpaceX also lends the trust roughly $2B to service EchoStar’s notes in the meantime.
• On 5 Nov 2025, that agreement was amended to add up to 15 MHz at 1695–1710 MHz for $2,616,737,853, all in SpaceX stock at $212 a share, taking the total to $19,616,737,853.
• Regulatory approval has been granted. The FCC approved both EchoStar transactions on 12 May 2026. The part that matters: its waivers let SpaceX use the spectrum “flexibly for terrestrial, space-based, and hybrid network architectures”, subject to performance obligations (FCC; see part 5).
• The second step, the one that puts the licences in SpaceX’s hands, is expected on or about 30 November 2027.

Filed EchoStar Forms 8-K, 8 Sep 2025 · 6 Nov 2025 · 3 Aug 2026

The 1915–1920 / 1995–2000 MHz pair is the H block, the channel immediately adjacent to the G block SpaceX rents from T-Mobile today. Whatever the reasoning behind the choice, the effect on the record is that the spectrum being bought sits directly beside the spectrum the service runs on now.

The first step, meanwhile, has already happened. The FCC snapshot of 13 September 2026 shows Spectrum Business Trust 2025-1 holding 847 active market licences, while Space Exploration Holdings still shows the same 173 leases. The structure is executing in public, on schedule, a year ahead of the closing.

So the arithmetic of the pivot is this: 10 MHz leased from a carrier today, 65 MHz owned outright in late 2027. Six and a half times the bandwidth, and no landlord.

You do not need a position of that size to supplement someone else’s network; you need it to be one. Trade reporting of SpaceX’s Q2 2026 earnings call has executives naming AT&T, T-Mobile and Verizon and expecting to take their customers; we have not verified those remarks independently, and the argument does not rest on them. It does not need to, because the filings say it more reliably than a quote does.

Width is not the same as coverage, and neither is the same as walls

Three constraints survive the purchase. Ownership changes the terms on which SpaceX holds spectrum; it does not change the frequency that spectrum sits at, the geography a lease can reach, or the position of the operators who never needed to rent in the first place. Each of the three is a matter of record rather than a matter of strategy.

Frequency

Everything SpaceX has contracted to own sits at 1.7–2.2 GHz, and everything AST leases sits at 700 MHz. Owning 65 MHz buys width; it does not buy propagation. As part 3 set out, the low-band lease is the one that diffracts around terrain and gets through a wall, and the incumbents’ public position, that satellite direct-to-device does not serve indoors, is an argument about frequency and link budget, not about satellite count.

Geography

A lease is granted market by market: 654 areas for AST, 174 for SpaceX. The footprint is whatever the landlord holds and chose to lend, which is why the lease lists read like a patchwork of rural counties. A nationwide licence has no holes in it, and that is most of the difference between “supplemental coverage” and “coverage”.

The incumbents

Globalstar and Iridium appear nowhere in the lease records because they own mobile-satellite spectrum outright, but narrowly, on the order of 10–25 MHz each, which is why their businesses are messaging and wholesale rather than broadband. That leaves three positions available in this market: owned and narrow, rented and wide, or bought and wide at roughly $2B per 5 MHz pair. Each one produces a different company, which is why the rent-or-own question is not a financing detail.