Alibaba Cloud’s Qwen-3 has become one of the first general-purpose AI models to operate in space, completing an entire inference task from Earth to orbit and back in under two minutes. In a milestone for China’s space-tech ambitions, Adaspace Technology successfully ran the model aboard its pioneering space computing constellation, marking a new front in the global race for off-planet computing supremacy.
Imagine asking a sophisticated AI a complex question and getting an answer beamed down from space, faster than you can brew a cup of coffee. That’s the reality Adaspace Technology, a Chinese aerospace startup, demonstrated in November. According to executive vice-president Wang Yabo, the company successfully uploaded Alibaba Cloud’s powerful Qwen-3 large language model to its orbital data center.
This isn’t just a neat tech demo; it’s a strategic move in a high-stakes competition. The operation came in the same month a U.S. company, Starcloud, ran Google’s Gemma model in orbit. Why the sudden rush to the final frontier for computing? The answer lies in the unique advantages of space itself. Orbital data centers can tap into virtually unlimited solar energy, benefit from the cold vacuum of space for easier cooling, and could drastically reduce the latency and cost of transmitting vast amounts of data across the globe.
READ ALSO: https://modernmechanics24.com/post/mystery-beneath-ocean-feeding-life/
The process, as detailed by Wang Yabo in a conference reported by the Star Market Daily, is impressively swift. From the moment a query was uploaded from a ground station to the completion of on-orbit AI inference and the return of results to Earth, the entire cycle took less than two minutes. This showcases the practical viability of processing data where it’s collected, be it from Earth observation satellites or deep-space probes, rather than suffering the delays of sending it all back to terrestrial servers.
This demonstration was conducted on Adaspace’s initial space computing center, a constellation of 12 satellites launched in May 2024 that forms the first phase of its ambitious “Star-Compute Project.” The full vision is staggering: a network of 2,800 satellites in low-earth orbit, with 2,400 dedicated to inference and 400 to training AI models. The company has plans to launch its second and third orbital computing centers as soon as 2026, with the entire megaconstellation slated for completion by 2035.
WATCH ALSO: https://modernmechanics24.com/post/explore-raytheons-giant-radar-making-facility/
For Alibaba Cloud, this space-bound success strengthens the credentials of its Qwen family, which it calls the world’s largest open-source AI ecosystem. The Qwen-3 model, released in April 2024, is now proving its mettle in the most extreme environment imaginable. The collaboration underscores a significant technological head start for China in orbital data processing, leveraging falling launch costs and innovative engineering to bypass the physical and energy constraints of ground-based data centers.
The strategic implications are clear. China has designated aerospace a top national priority. With policy wind in its sails, the domestic satellite industry is projected to boom from 82.7 billion yuan in 2024 to 266.1 billion yuan by 2029. Adaspace, headquartered in Chengdu and valued at over 6.76 billion yuan, is positioning itself at the nexus of this growth, covering the entire satellite lifecycle from design to operations.
READ ALSO: https://modernmechanics24.com/post/japan-restart-worlds-biggest-nuclear-plant/
“This isn’t science fiction anymore,” one could imagine a project engineer saying. It’s a tangible step toward a future where the backbone of our global AI infrastructure isn’t in sprawling desert warehouses, but silently orbiting overhead, powered by the sun and thinking at the speed of light. The race for space-based computing is officially on, and the first laps have just been run.
