Broadband from orbit

Starlink doubled in a year while both geostationary services shrank

For the first time, every company selling consumer broadband from orbit in the United States files its subscriber count with the SEC, so the three series can be put side by side. They point in opposite directions (Exhibit 1).

Starlink reached 12.0 million subscribers worldwide on 30 June 2026. A year earlier it had 6.0 million, so by SpaceX’s own figure the base grew 101.2% in twelve months. The series in the prospectus shows a pattern of doubling: 2.3 million at the end of 2023, 4.4 million at the end of 2024 and 8.9 million at the end of 2025. The geostationary operators report the mirror image. HughesNet, EchoStar’s broadband service, fell from 1,004,000 subscribers in December 2023 to 622,000 in June 2026, a loss of 38%, and 59,000 of those went in the second quarter of 2026 alone. Viasat’s American fixed broadband base went from about 189,000 in March 2025 to about 115,000 in June 2026, down 39% in fifteen months.

The two series are not on the same base. Starlink’s figure is global, because SpaceX discloses no American count, while HughesNet’s and Viasat’s are American only. The direction does not depend on the base: one technology doubled in a year, while the other lost more than a third of its customers in two and a half years. Together the two geostationary services had about 737,000 subscribers at the end of June 2026, 6.1% of Starlink’s global count, or one subscriber for every sixteen of Starlink’s.

Revenue grew faster than the price fell

The money followed the subscribers, though not quite in step. SpaceX’s connectivity segment took in $4,291 million of revenue in the second quarter of 2026, up 65.8% from $2,588 million a year earlier. Consumers paid $2,485 million of that and enterprise and government customers $1,806 million. The segment made $1,656 million of income from operations, which works out at a margin of close to 40%.

The average subscriber paid less. Revenue per subscriber was $66 a month in the quarter, against $85 a year earlier, a fall of 22.4%. SpaceX puts it down “primarily” to “international expansion and the addition of lower priced service plans”. The annual series runs the same way: $99 in 2023, $91 in 2024 and $81 in 2025. The company tells investors to expect more of it “as the portion of our subscriber base outside North America continues to grow” (Exhibit 3). For comparison, Viasat’s American customers paid $111 a month on average at 30 June 2026, nearly 70% more than the average Starlink subscriber worldwide.

Filed SpaceX 10-Q, quarter to 30 Jun 2026, segment revenue and key metrics

American retail prices, meanwhile, went up. From June 2026 billing, Starlink’s residential plans in the United States cost $55 a month for 100 Mbps, $85 for 200 Mbps and $130 for the top tier, according to SatelliteInternet.com and 9to5Google. In January they had been $50, $80 and $120. 9to5Google, following Ars Technica, also reported in June that Starlink had stopped selling residential hardware outright and now rents it. The average is falling because of where the new subscribers are, while the American price rises. Amazon Leo had published no consumer price and no commercial launch date by 7 October 2026.

One geostationary operator is in Chapter 11

The decline in geostationary broadband has reached the courts. On 2 August 2026 Hughes Satellite Systems Corporation and certain subsidiaries, including Hughes Communications and Hughes Network Systems, filed voluntary petitions under Chapter 11. EchoStar, the parent, did not file, and its two Chapter 11 cases are running in opposite directions. Its DISH DBS subsidiary, which runs the satellite television business, emerged from its own prepackaged case on 1 October 2026, two months after the broadband subsidiary went in.

Viasat, the other geostationary consumer provider, is in the awkward position of adding capacity to a shrinking base. ViaSat-3 F2, launched in November 2025, was expected to enter service in September 2026, and F3, launched in April 2026, by late summer. Its FY2026 10-K warns that government support programmes “provide, or will likely provide, support to terrestrial service providers under terms and conditions that are not available to satellite-based service providers”. BEAD tests that claim.

BEAD gives satellite a fifth of its locations and a twentieth of its money

BEAD became technology-neutral in June 2025, and by the end of August 2026 every state and territory had an approved plan. Satellite won a large share of the locations and a small share of the money, and in July, before all the plans were approved, NTIA moved to take part of the locations back.

The change came in NTIA’s Restructuring Policy Notice of 6 June 2025, which “eliminates the ‘Fiber Preference’” and ordered a fresh “Benefit of the Bargain” round for every eligible location. A priority project, under the notice, is one delivering at least 100 Mbps down and 20 Mbps up. For satellite it created a new instrument, the “LEO Capacity Subgrant”, with four conditions attached.

• Capacity. The provider has to reserve capacity for every location in the project area.
• Term. The period of performance is ten years.
• Payment. Up to half of the grant can be paid early, once service is certified across the project area or subscription milestones are met.
• Equipment. Customer equipment comes at no cost as part of a standard installation.

A condition that matters more than those four sits in the low-cost plan every subgrantee must offer. That plan has to deliver 100/20 Mbps with “latency performance of no more than 100 milliseconds”, and physics alone rules geostationary satellites out. The round trip through one is at least 477.5 ms, nearly five times the cap, and even O3b mPOWER in medium orbit starts at 107.6 ms (part 1). When Viasat says the terms are not available to satellite providers, it is right about its own orbit and wrong about low orbit.

On Broadband Breakfast’s tally of 28 August 2026 the approved plans cover 3.79 million locations and $18.19 billion of deployment spending. Fibre takes 62.9% of the locations, satellite 21.9%, fixed wireless 13.3% and cable 1.9% (Exhibit 4). The money is split differently. SpaceX and Amazon are in line for a combined $1 billion, which works out at about 5.5% of the deployment total: a fifth of the addresses for a twentieth of the cash, because satellite is cheap per address. ACLP at New York Law School counted 900,939 satellite locations and about $1,025 million in the final proposals, about $1,140 per satellite location against about $4,800 for the programme as a whole (Exhibit 5).

In July NTIA told states to check every satellite-awarded location against the updated FCC map and to remove any that the map shows as already served, or as not existing at all. StateScoop reported that NTIA would revoke approval of those locations unless a state could refute the new data. ACLP estimates the cut at 35% of satellite locations, about $354 million, or 42% and about $423 million if unlicensed fixed wireless counts as service. StateScoop’s own tally of the 47 states that had proposed satellite locations found an average expected cut of 33.77%, ranging from 0.9% in Rhode Island to 75.4% in Mississippi. On ACLP’s base, that leaves satellite with between $602 million and $671 million.

Two later steps complicate the picture. In August NTIA said, as Broadband Breakfast reported, that states cannot unilaterally alter BEAD contracts they have already signed, which is hard to square with a review that asks them to remove locations. On 3 September a Supplemental Deployment Policy Notice, reported by Light Reading, let states put part of the roughly $21 billion of non-deployment savings into a further deployment round for locations still unserved, without saying how much. Nothing had settled by 7 October 2026, when NTIA had still published no post-review satellite totals.

Satellite won about a fifth of BEAD’s locations for about a twentieth of its money, at roughly a quarter of the average subsidy per location, and a third or more of those locations are expected to go in the July review. Against the business it subsidises, the award is small: about $1.0 billion over ten years, before the cut, is less than half of the $2,485 million that Starlink’s consumers paid SpaceX in the second quarter of 2026 alone. In American broadband from orbit the subscriber pays and the subsidy is a footnote.

The subsidy record runs back to an award SpaceX lost

BEAD is the second federal programme to put satellite broadband on its books, and the first did not end well for SpaceX. In the 2020 Rural Digital Opportunity Fund auction it won “$885,509,638.40 in 10-year support” to serve 642,925 locations in 35 states, which is about $1,377 a location. The Wireline Competition Bureau declined to authorise the award on 10 August 2022, and the Commission affirmed that decision in FCC 23-105, released on 12 December 2023, with Commissioners Carr and Simington dissenting. In November 2024 Carr, by then the incoming chairman, said it was “very unlikely” the FCC would revisit it, and it has not (Exhibit 6).

Two comparisons put the subsidy in proportion. The award SpaceX lost in 2022 was 86% of the whole BEAD satellite award to every provider, on ACLP’s count, so the company has already been denied nearly as much public money as the current programme offers every satellite provider put together. SpaceX’s prospectus mentions neither RDOF nor BEAD, which in our view says how much they matter to the business.

Terms used

• ARPU. Average revenue per user. For Starlink, SpaceX defines it as service revenue divided by the average number of subscribers and by the months in the period.
• BEAD. The Broadband Equity, Access, and Deployment programme run by NTIA, under which states award subgrants to build broadband to unserved and underserved locations.
• Low-cost option. The cheaper plan every BEAD subgrantee must offer eligible households. Since June 2025 it has had to deliver 100/20 Mbps with latency of no more than 100 ms; the subgrantee sets the price.
• RDOF. The FCC’s Rural Digital Opportunity Fund, a 2020 reverse auction for ten years of support to serve rural locations.