🛰️ Sometime after mid-August, a helium airship will lift off from a hangar in Roswell, New Mexico, climb to around 20 kilometres, and start drifting west. About two weeks later it should be over Japan, where it will spend a week or so doing the job of a mobile phone tower. SoftBank calls this a pre-commercial service. What it actually is: a bet that the stratosphere is a better place to hang a base station than orbit.

What actually flies in August

SoftBank laid out the plan at an online briefing on July 23. The vehicle is a HAPS, short for High Altitude Platform Station, and the Japanese nickname for it translates as "flying base station." The concept is not complicated. Park an aircraft about 20 kilometres up, hang mobile network gear off it, and it covers ground the way a tower does, except that one vehicle can reach up to about 200 kilometres across. A conventional ground station covers a few kilometres.

The aircraft is not SoftBank's own. It comes from Sceye, a New Mexico company that builds lighter-than-air platforms, essentially steerable stratospheric airships. SoftBank invested in Sceye in June 2025 and, under an agreement signed on June 20 that year, took exclusive rights to offer HAPS services in Japan using Sceye's platform. SoftBank put the vehicle's length at roughly 65 metres.

It has to fly from the United States because nothing in Japan can house it. SoftBank was blunt about that at the briefing: there is no domestic hangar big enough for a 65-metre vehicle. Building one is a later goal. For now the airship crosses the Pacific under satellite control, and only once it reaches Japanese airspace does it link to a ground gateway and start carrying traffic.

That gateway is already up, on high ground in Kochi Prefecture, with an antenna that tracks the airship automatically as it circles and tilts overhead. Kochi faces the Nankai Trough, the offshore fault Japan expects to produce its next catastrophic earthquake, so the test scenario writes itself: a simulated disaster zone, cut off from the ground network, served from above. The final site will be chosen closer to the day based on weather, with Kansai, Kyushu and Tohoku also in the running.

Three things get measured. Whether the vehicle holds position in Japanese weather the way it has elsewhere. Whether voice calls, SMS and data from 20 kilometres up feel like ordinary mobile service. And whether the thing interferes with ground base stations, which matters more than it sounds. In a real disaster some towers keep running while others go dark from power cuts or severed backhaul, so the airship has to slot into a patchy network without trampling it.

Then it flies home. Total time over Japan: seven to ten days.

Why not just use satellites

Fair question, and SoftBank has an answer, partly because it already sells the alternative. SoftBank Starlink Direct launched on April 10, 2026, connecting ordinary phones straight to SpaceX satellites roughly 340 kilometres up. It reaches essentially the whole country. It also, for now, handles text and a narrow list of apps.

Masayuki Kamimura, senior director of SoftBank's Ubiquitous Network Planning Division, drew the line at capacity, and specifically at the uplink: the data travelling from your handset up into the network. Sending a drone a flight command takes almost nothing. Streaming that drone's video back to a monitoring centre in real time is a different problem, and it is the direction satellites handle worst. At 20 kilometres, SoftBank argues, a HAPS can deliver something close to normal terrestrial service in both directions, over a defined area.

Which is why the two are being designed as a sequence rather than a contest. The moment a disaster hits, Starlink Direct is already overhead and can carry messages immediately. Days later, once the airship has been moved into position, it layers voice and heavier data on top.

Everyone who has tried this before

The stratosphere has a graveyard, and SoftBank has been standing in it for years.

Alphabet's Loon was the loudest attempt: balloons carrying internet to Kenya, to Puerto Rico after Hurricane Maria, to Peru after an earthquake. In January 2021 Alphabet shut it down, saying it could not get costs low enough to build a lasting business. SoftBank's HAPSMobile unit had put $125 million into Loon in April 2019. After the shutdown SoftBank bought roughly 200 of Loon's HAPS patents, bringing its portfolio to around 500, and carried on.

The fixed-wing route has been rough too. Airbus subsidiary AALTO's Zephyr set a HAPS endurance record of 64 continuous days in 2022, and a consortium led by NTT Docomo and Space Compass committed $100 million to AALTO in June 2024 to bring the aircraft to Japan. Then, on April 28, 2025, a Zephyr 8 came down into the Indian Ocean on its 67th day aloft.

What changed the arithmetic for SoftBank happened this spring. Sceye's SE2 airship lifted off from Roswell on March 25, 2026, and flew for more than twelve days and over 6,400 miles before a planned, controlled termination in international waters off Brazil on April 6. Along the way it spent more than 88 hours holding position over selected spots, at times inside a radius of a single kilometre. Staying up is one problem; staying put, night after night on stored solar power, is the harder one, and that is what the flight was built to answer.

The awkward question SoftBank hasn't answered

Here the story gets less tidy.

SoftBank does not intend to launch these things after an earthquake. The plan is to keep them flying in ordinary times, earning revenue from construction sites with poor coverage, drone video, remote sensing, extra capacity over crowded events. Government agencies have already asked about specific uses, the company says. Then, when disaster strikes, an airship that is already aloft gets moved to where it is needed.

Move it, though, and the customers it was serving lose their coverage. For a business paying for a network link, "we relocated your infrastructure to another prefecture" is not an acceptable outage notice. SoftBank recognises the tension and is leaning toward flying HAPS where terrestrial coverage already exists, as supplementary capacity rather than sole coverage, so that a relocation degrades service instead of ending it. Whether that scales, and whether it eventually needs spare vehicles or contractual fallbacks, is unresolved.

Weather is the other snag. Typhoons will be dodged rather than endured, at least initially, which means the technology is least available during precisely one category of Japanese disaster. Earthquakes and wildfires, where the sky is usually calm, are the better fit.

And the operating window is still short. Seven to ten days now. Three to six months is under consideration for 2027 onward, depending on what August reveals about batteries, solar output and motor performance across the seasons.

American hardware, Japanese customer

There is something odd about the whole arrangement when you step back from it. The airship is American. The company that built it is American. The best-funded direct-to-phone satellite networks are American. But the carrier about to fly a commercial cell tower over a populated country, on a schedule, with a gateway already bolted to a hillside, is Japanese.

That is less a matter of engineering than of geography. By SoftBank's own account, its ground network reaches close to 100 percent of the population but only about 60 percent of Japan's land area, leaving a stubborn 40 percent, much of it mountains and islands, that towers never economically reach. Add a megaquake the country knows is coming, and a technology whose entire selling point is that you can move it to wherever the network broke has a customer here in a way it does not in most places.

Does the mobile network where you live have a disaster plan you would actually trust? And if you had to choose, would you rather have a satellite that reaches everywhere with a thin signal, or an aircraft that can be flown to you?

References