Chengdu firm sets up large computing model in space
GUANGZHOU — A Chinese commercial aerospace firm has successfully deployed a general-purpose artificial intelligence model aboard its orbiting satellites, marking a major milestone for space-based computing.
GuoXing Aerospace Technology announced at a seminar on Monday that it has uplinked Alibaba's Qwen3 large language model onto its inaugural space-based computing center, enabling end-to-end reasoning tasks entirely in orbit.
"This marks the world's first deployment of a general-purpose large-scale AI model from ground control to an operational satellite constellation in orbit," said Wang Yabo, executive vice-president of the startup based in Chengdu, Sichuan province.
In May 2025, China launched a new constellation of 12 space computing satellites into orbit, the first cluster of GuoXing Aerospace's space computing project.
In the trial, the Qwen3 model completed multiple experiments, with questions transmitted from Earth to the satellite, processed on board, and results returned to ground stations — all within two minutes.
As AI fuels an insatiable appetite for computing power, a new field is emerging in the technological race as intelligent computing capabilities are pushed into space. In November, a SpaceX rocket orbited the Starcloud-1 satellite mounted with Nvidia GPUs.
Wang outlined the firm's ambitions to build a sprawling network of 2,800 specialized computing satellites by 2035.
The planned constellation includes 2,400 inference satellites and 400 training satellites, deployed across sun-synchronous, dawn-dusk and low-inclination orbits at altitudes of 500 to 1,000 kilometers.
The constellation is designed to employ laser inter-satellite links to facilitate high-speed data transfer, aiming to deliver 100,000 petaflops of inference compute and 1 million petaflops of training compute worldwide.
Its second and third satellite clusters are expected to be deployed this year, with a 1,000-satellite network completed by 2030, according to Wang.
Xinhua
