Star.vision's Space.IDC: 1,080 spacecraft, 10 billion yuan and no orbital plan

The company calls it the Space.IDC Computing Constellation, or Space String Computing Constellation when the Chinese name is rendered literally. The pitch is straightforward: put serious compute in orbit, where solar power is abundant and waste heat can be radiated into vacuum, and sell the output back to customers on Earth or to satellites that cannot pre-process their own data.

Star.vision, also written STAR.VISION Aerospace, announced the plan at the Global Digital Trade Expo on 25 September, according to the trade outlet china-in-space.com. The company is not the first Chinese firm to propose this. It joins what that report describes as a handful of domestic enterprises with similar projects under development.

The numbers the company did share

Three spacecraft designs are on the table. The G1 verification craft is meant to deliver 40 to 60 PetaFLOPs from 11.5 kilowatts of solar generation. The G2 standard unit is quoted at 160 to 310 PetaFLOPs and 48 kilowatts. The G3 flagship is the big one: 460 to 1,000 PetaFLOPs from 140 kilowatts of panels.

Of the roughly 1,080 spacecraft in the main constellation, about 720 would handle inference tasks and 360 would be reserved for general compute. Full deployment is targeted for 2035, per china-in-space.com.

Those are the numbers Star.vision put on the record. The numbers it did not put on the record are, for anyone evaluating the project, at least as interesting. The report notes that orbits, inter-satellite links, radiator sizing for cooling and spacecraft mass were not shared. Since the company expects optical laser links to carry the data volumes implied by a compute constellation, that is a significant gap. International regulatory paperwork that would normally reveal orbital and spectrum details has not yet been filed.

Money, launches and a financing gap

The financial scaffolding is partly public. Star.vision says it has secured strategic credit agreements with the Industrial and Commercial Bank of China and the Agricultural Bank of China worth 10 billion yuan, or about 1.49 billion US dollars as of 28 September. At least 1 billion yuan, roughly 149 million dollars, is earmarked for a first phase.

Launch capacity is also claimed, if not yet scheduled. At the 25 September unveiling the company said it had booked an undisclosed number of launches on Orienspace's Gravity-1 and Galactic Energy's Ceres-2 to fly G1 spacecraft. No target month or quarter was given.

The heritage argument is stronger than the schedule. Star.vision has been putting AI systems on remote sensing satellites since at least mid-2023, one of the earlier Chinese commercial players to do so. It also claims that two satellites it built for Uzbekistan and Indonesia have been used to verify parts of the coming constellation.

Why this is not a one-company story

Orbital compute is moving from slideware to hardware across the sector, and the autonomy question is the live one. IEEE Spectrum reported on 29 September that NASA's Jet Propulsion Laboratory used Anthropic's Claude models in December to help plan two Mars drives for the Perseverance rover, with human planners checking and adjusting the route before upload. In May, NASA and IBM put a compressed AI model on the International Space Station and a satellite to identify floods and clouds from orbit, described as the first demonstration of its kind in space. In July, astronauts on the station tested a large language model against maintenance procedure questions.

Those are narrow, supervised tests. They are also the direction of travel. A constellation that sells inference cycles from orbit is a different proposition from a rover planning a drive with a human in the loop, and the engineering constraints are correspondingly harsher: power, thermal rejection, link budget and, increasingly, the question of who is accountable when a model makes a call nobody reviewed.

Star.vision mentioned that it has booked an undisclosed number of launches aboard Orienspace's Gravity-1 and Galactic Energy's Ceres-2 to deploy G1 spacecraft.

That sentence, from china-in-space.com, is the whole of the publicly available launch plan. Undisclosed number, undisclosed timing, two named vehicles.

Reading the claim against the calendar

The deployment target is 2035. That is nine years out. A first phase funded at 1 billion yuan is a rounding error against a constellation of more than a thousand compute satellites with 140-kilowatt solar arrays, even allowing for the fact that Chinese launch costs and manufacturing scale differ from Western benchmarks.

The comparison that matters is not with a Western constellation. It is with the other Chinese proposals in the same category, none of which was named in the material reviewed here. Star.vision's differentiator, on the record, is that it has flown AI payloads before and can point to two exported satellites as partial validation. Its weakness is that the hard engineering details, the ones that determine whether a watt of solar power becomes a usable FLOP in orbit, are absent.

There is also a regulatory timeline nobody has started. Radio spectrum and orbital slot filings are public documents. Until they appear, the Space.IDC constellation exists as a design study with a credit line and a partner list.

What to watch

Three things. First, the international regulatory paperwork, which would pin down orbits, spectrum and likely spacecraft mass. Second, a named launch date for G1, which would convert two vehicle bookings into a testable milestone. Third, the first independent performance figures from orbit, since PetaFLOP counts on a slide are not the same as sustained throughput behind a laser link.

Until at least one of those lands, the honest description of Space.IDC is a well-financed intention. The banks are real. The partners are named. The spacecraft are not yet flying, and the company has not said when the first one will.

For context on how crowded the near-term launch calendar is, NASASpaceflight reported on 29 September that four orbital launches were expected in a single week, all on SpaceX Falcon vehicles, including Crew-13 to the space station and a Falcon Heavy for the National Reconnaissance Office. Every one of those missions was to feature a return-to-launch-site landing. That is the manifest environment a new constellation has to compete for slots in.

The dossier also contains items that do not bear on this story and are noted only to be excluded: a tabloid report about a shape on a SpaceX livestream, a beet flavour explainer, an anonymous reasoning model preview, a wildlife photo benchmark, and a browser arcade game. None of them contributes a verifiable fact to the Star.vision announcement.