High-altitude platforms

A platform at 20 km has a tower’s link budget and a region’s footprint

A platform at 20 km flies at a twenty-fourth of the height of Starlink’s main shells at 480 km, and that one ratio is the argument for putting a base station there.

Directly below a platform at 20 km a signal arrives in 0.067 ms, as near to instant as makes no difference to a phone. At 2 GHz the free-space loss over those 20 km is 124.5 dB; over the 550 km to a low-orbit satellite overhead it is 153.3 dB. The platform starts 28.8 dB ahead, about 760 times the power, and that margin lets an unmodified phone work with it in a terrestrial mobile band at terrestrial power.

Reach is what the height costs. At 10° minimum elevation a platform serves a disc 216 km across, about 36,600 km², a region rather than a country. Covering the Earth once would take at least 16,865 platforms, against 85 satellites at 480 km (part 1). A platform is a regional asset, and the size of the region depends on how low over the horizon its users can look (Exhibit 1).

WRC-23 settled the international rules

The ITU has a name for a platform used as a mobile base station, HIBS, and WRC-23 in Dubai wrote its rules. They are now in the Radio Regulations (Exhibit 2).

The Radio Regulations define a high-altitude platform station as one “located on an object at an altitude of 20 to 50 km and at a specified, nominal, fixed point relative to the Earth”. The three HIBS resolutions narrow that window and lower it, to platforms “operating between 18 km and 25 km”, and Resolution 221 notes that transmissions “may occur at altitudes down to 18 km”. The two definitions overlap rather than coincide, which matters later. Four conditions attach to every identification.

• Optional. Each identification is optional for administrations and “does not establish priority”.
• Unprotected. A platform cannot “claim protection from existing primary services”.
• Power-capped. Each resolution caps the power-flux density a platform may put on another country’s territory, for instance −114 dB(W/(m²·MHz)) to protect mobile stations at 694 to 960 MHz.
• Enforceable. Notification requires “a firm, objective, actionable, measurable and enforceable commitment” to remove unacceptable interference immediately.

Nothing more is coming from the ITU. The WRC-27 agenda adopted at WRC-23 has nineteen items, 1.1 to 1.19, and none concerns high-altitude platforms. The preliminary WRC-31 agenda has none either; its item 2.5 is about phones used aboard aircraft, a different matter. The international rulebook is settled until at least 2031, and the work has moved to national licensing and to 3GPP.

The United States has written no rules, and the one time it looked at the question it said no. The FCC asked in 2021 whether to admit high-altitude platforms to the 70/80/90 GHz bands. In FCC 24-16 of January 2024 it declined, finding “highly divergent claims” about integrating the platforms into those bands, “with limited actual data to support such action”. The footnotes say more than the finding. They record that the only advocate to file a detailed technical study, Loon, “shuttered in early 2021”, and quote T-Mobile, which told the Commission there was “little current documented interest in those services and, unlike 5G, no deployment”.

To 7 October 2026 there is no FCC rulemaking and no US Table footnote implementing the WRC-23 identifications. Japan is moving quickly the other way. Its Information and Communications Council delivered a partial report on technical requirements for the platforms on 14 November 2025. MIC consulted on a draft ordinance from December 2025 to January 2026, and the Radio Regulatory Council found the drafts appropriate on 3 February 2026. The published documents give neither the bands nor the date the rules take effect, so Japan is writing rules whose content is not yet public.

The standard treats a platform as a base station that flies

3GPP never wrote a satellite specification for the platforms and did not need to. TS 38.104, the base-station radio specification, says in Release 19 that seventeen terrestrial bands, from n1 to n90, “can be applied for HAPS operation” (Exhibit 3). The SoftBank-led item that added them notes that “HAPS can be supported using terrestrial network (TN) functionality available in Rel-17”. To the standard, a platform is a terrestrial base station with an altitude, in terrestrial spectrum, and it needs none of the satellite bands in part 5. The 6G study, TR 38.914, lists it among the non-terrestrial components 6G must support.

In September 2026 operator field evidence entered the 3GPP record, and it concerned interference rather than coverage. RP-262210, approved at RAN#113 in Madrid, comes from China Telecom, SoftBank and CICT. Its justification says that where a terrestrial network and a platform share a low band such as n5 at 800 MHz in the same area, “severe co-channel interference (CCI) occurs”. The cause is the platform’s height. The same thing that lets it reach a phone means its downlink “creates a massive line-of-sight (LOS) interference region” that degrades the terrestrial cells beneath it. The remedy the item specifies, due by March 2027, is to split the platform across two bands. Its downlink goes in mid-band, such as 2.1 GHz, and its uplink stays low on bands such as n89 at 800 MHz, because “closing the uplink budget at 2.1GHz is extremely challenging”.

The same height that lets a platform reach an unmodified phone by line of sight puts its signal into every cell for a hundred kilometres around. The only operator field evidence in the standards record is about that problem rather than about coverage, and the remedy splits the platform across two bands. In our view a carrier cannot drop a platform into the low band its towers already use and expect it to behave; it needs a frequency plan first.

No programme has reached a commercial service

The platforms that have flown longest belong to companies that have never sold a service, and every commercial date on the record is still in the future (Exhibits 4 and 5).

The flight records point the same way (Exhibit 6). The longest belongs to Loon, which closed. The longest by a company still flying belongs to AALTO, which has sold no service. The nearest thing to a service on the record is SoftBank’s trial, set out below.

No operator has published a cost or a revenue

No operator has published what a platform costs to build or to fly, or what a service earns, and the two data points that exist are a closure and an absorption. Loon’s chief executive said on its closure that “we haven’t found a way to get the costs low enough to build a long-term, sustainable business”. HAPSMobile lost ¥448 million in the year to March 2023, on ¥37 million of assets, before SoftBank absorbed it on 1 October 2023.

One transaction price exists. AeroVironment’s 10-K records that in March 2022 it “sold its 7 % equity interest in HAPSMobile to SoftBank, for 808,008,000 yen ($ 6,497,000)”; the same note calls the stake “approximately 7 %”. At 7% the price values the whole company at about ¥11.5 billion, or about $93 million (our calculation). On the funding side, HAPS JAPAN, led by NTT Docomo and Space Compass with Mizuho and the Development Bank of Japan, committed $100 million to AALTO in June 2024, subject to approvals. SoftBank has not disclosed its stake in Sceye.

Filed AeroVironment Form 10-K for FY2024, filed 27 Jun 2024, note on the investment in HAPSMobile; implied value calculated by Blue Prysm

Reported DCD, 24 Jul 2023 (HAPSMobile loss and assets) · HAPS JAPAN and AALTO release, 3 Jun 2024 · SoftBank, 2 Sep 2026

That leaves footprint against footprint as the only comparison available with low orbit. One platform serves 9,200 to 36,600 km² continuously, depending on whether its users look down to 20° or 10° above the horizon. The companies claim more: one platform replaces 200 terrestrial towers, says Stratospheric Platforms (2020), or covers the area of about 500, says Sceye (2026). The structural difference is larger. A constellation has to reach global scale before it serves any region continuously, while a platform serves one region from its first day and never more. In our view that makes it an operator’s layer, in the operator’s own spectrum, for disaster recovery, islands and mountains, rather than a broadband network that competes with Starlink or Amazon Leo. Whether it is a cheap layer or an expensive one cannot yet be said, because no published figure allows a cost per gigabyte or per square kilometre to be set beside either.

Terms used

• HAPS and HIBS. A high-altitude platform station is any radio station on a platform at 20 to 50 km. HIBS is the ITU’s term for one used as a mobile base station, under conditions written for 18 to 25 km.
• Power-flux density. The power arriving per square metre at the ground, which is the measure the Radio Regulations use to cap interference into a neighbouring country.
• Supplementary uplink. A band used only for the uplink, paired with a downlink in another band. The device can then transmit low, where its signal travels further, and receive high, where there is more capacity.
• Station-keeping. Holding a platform over a fixed point against the wind. The radius it holds decides how stable the cells beneath it are.