• Home
  • News
  • AI in Space: A look into Alibaba’s Qwen-3, The First General-use AI Model to Operate in Orbit
Aerospace, Business-industry-data-aerospace, Community-youth, Defence, Global, Innovation, Space

AI in Space: A look into Alibaba’s Qwen-3, The First General-use AI Model to Operate in Orbit

  • 30 Apr 2026
  • Text by Ilhan Abdul Rahim, Temasek Polytechnic

Alibaba’s artificial intelligence (AI) model, Qwen-3, and its applications in space are marking a significant development, as China steps up efforts to become a major player in both the aerospace and computing industries.

In May 2025, Chinese multinational company Alibaba deployed Qwen-3 – its artificial intelligence (AI) model – into orbit for operations, in a joint mission with ADA Space (also known as Chengdu Guoxing Aerospace Technology Co., Ltd., or Guoxing Aerospace) and the research institute Zhejiang Lab.

Qwen-3 is touted as a high-performance, hybrid AI model from Alibaba Cloud that combines efficient inference with strong reasoning and agentic capabilities. It allows seamless switching between deep reasoning (“thinking mode”) for complex problems and fast, general-purpose inference. Combining Dense and Mixture-of-Experts (MoE) models, it is designed for high throughput, using less than 10% of active parameters while maintaining comparable performance.

Star-Compute satellite. Photo: Guoxing Aerospace (ADA Space)

The deployment forms part of ADA Space’s “Star-Compute Project”, which aims to launch 2,800 satellites by 2035 to establish a space-based supercomputing network. The inaugural launch saw the deployment of 12 computing satellites in orbit, including one hosting Qwen-3’s data centre. With in-orbit testing successfully conducted since November 2025, the Qwen-3 deployment is considered among the foremost general-use systems of its kind to operate in space.

While AI has previously been deployed in space by agencies such as NASA, these systems have traditionally been task specific. Qwen-3 represents a shift towards general-purpose AI models in orbit—closer in function to widely used large language models (LLMs) such as ChatGPT.

Qwen-3 has successfully demonstrated multiple core capabilities under real orbital conditions. This includes full processing cycles, starting with prompts sent from Earth and ending with the transmission of results back from orbit. The cycle – from query to results – took under two minutes, enabling rapid responses for time-critical applications. Beyond speed, Qwen-3 also successfully demonstrated the AI process known as end-to-end inference – processing multiple queries entirely in orbit without reliance on ground-based computing. This marks a shift from traditional architectures, where space-based systems depend heavily on Earth for processing.  Over months of testing, the system also validated key functions including networking, onboard computing, machine learning, and system monitoring, ensuring all instruments and systems aboard the constellation are functioning as expected.

Star-Compute satellite. Photo: Guoxing Aerospace (ADA Space)

The concept of an orbital data centre presents several advantages over terrestrial or ground-based data centres. These include access to near-continuous solar energy unhindered by the Earth’s day and night cycles and weather conditions as well as, lower cooling demands due to the space environment. In addition, space-based computing can significantly reduce latency and bandwidth constraints while mitigating risks such as signal interference or jamming.

Operationally, AI systems like Qwen-3 also enable greater autonomy. While traditional terrestrial systems would require human intervention for routine checks and maintenance, Qwen-3 can perform the same task and processes independently.

Beyond the AI model, the 12 satellites of the constellation are equipped with space computing hardware and software developed by ADA Space, alongside intelligent computing and inter-satellite communication system. These enable high levels of coordination and processing power across the constellation. Notably, inter-satellite communication speeds can reach up to 100 gigabits per second via laser links.

While Qwen-3’s orbital model has yet to report specific practical applications, its demonstrated capabilities suggest strong potential across military, commercial, and research domains. Possible use cases include real-time intelligence for Earth observation and disaster response, as well as scientific missions such as gamma-ray burst studies supported by onboard instruments like X-ray polarisation detector, which will work on gamma ray-burst research.

A carrier rocket carrying the computing constellation. Photo: Guoxing Aerospace (ADA Space)

The project aligns with China’s broader policy direction to maximise AI and aerospace capabilities, as outlined in its five-year plan emphasising technological self-reliance. Similar developments are also emerging globally. In the United States, companies have begun exploring orbital computing, most notably with Google’s open large language model Gemma being deployed for space in collaboration with Starcloud and Nvidia . Starcloud aims to demonstrate that outer space can serve as a viable environment for data centres, particularly as Earth-based facilities plain increasing strain on power grids, consume large volumes of water and generate significant greenhouse gas emissions. 

Initiatives such as the Star-Compute Project are expected to drive advancements not only in AI capabilities but also in the broader space economy and sustainable data infrastructure through orbital centres.

The Qwen-3 model, trained on 36 trillion tokens across a plethora of languages and dialects, reflect the scale and sophistication of current AI models. According to ADA Space, further expansions of this project are planned for 2026, with additional satellite launches expected to form the second and third space computing centres. This, to advance the long-term vision of a fully integrated orbital computing network.

You May Also LikeAAIS

Previous Article
Next Article