Did China just take the lead in race for space-based AI computing dominance?

The Star

22-05-2025

China's launch last week of a cluster of 12 satellites that will form the foundation of the world's first orbital computing constellation signalled the country's ambition to dominate space-based computational power.
The constellation, a collaboration between aerospace innovators like Guoxing Aerospace and AI specialists at Zhejiang Lab is designed to process data in orbit rather than rely on land infrastructure.
The system aims to take advantage of the unlimited solar energy and lower cooling demands of space, as well as today's falling launch costs.
There is also the promise of speed and efficiency.
For instance, disaster response teams could use satellites to analyse real-time imagery mid-orbit, bypassing data bottlenecks on Earth and saving time. Or a network of solar-powered orbital 'data factories' could cut reliance on terrestrial grids and reduce the need for cooling.
'We believe that the fusion of AI represents the future of the space industry. In the long term, space-based computing could become a foundational infrastructure accessible to all,' said Chen Junrui, spokesman for STAR.VISION, a private aerospace company based in Hangzhou.
The company's artificial intelligence-powered satellites suggest a future where space is a hub for critical computing tasks. While global players like the European Union and US firms have similar goals, China's interconnected 12-satellite system outpaces rivals in operational scale. Space-based computing could become a foundational infrastructure accessible to all
Still, according to experts, it is just the first step in a marathon with many technical hurdles – from radiation-proof hardware to perfecting in-orbit collaboration.
Deploying computing power in the vacuum of space offers unique advantages, such as the ability to efficiently harness solar radiation to power equipment, and bypassing the cooling challenges that plague Earth-based data centres. In addition, launch costs have steadily decreased in recent years, making such ventures increasingly feasible.
STAR.VISION has already ventured into this area. Two years ago, the company launched an experimental computing satellite, and last year it partnered with Oman to launch another, equipped with 80 TOPS (trillion operations per second) of processing power and running an operating system also developed by Zhejiang Lab.
The company sees a future in which space-based computing networks are as advanced and accessible as terrestrial internet infrastructure. Once mature, such a system could support a wide range of users – including governments, enterprises, academic institutions and individuals.
'Imagine a scenario involving an earthquake or man-made disaster,' Chen said. 'With a robust space network, a satellite could capture an image, instantly relay it to another satellite for processing, and send analysed data back to Earth. This could dramatically accelerate emergency responses – identifying the safest access routes or advising how to navigate collapsed structures. Think of it as an 'Agent GPT' in space.'
In April, a paper published in the journal Chinese Space Science and Technology further explored the concept of building data centres in orbit. Researchers from Xian-based Xihai Shudao Data Technology proposed the 'Space Computing Project', which aims to deploy a 'soft supercomputing' platform powered by their proprietary high-performance real-time database, NoSQLt.
The platform would harness electricity from space-based solar power stations to process satellite data in orbit, turning energy that would otherwise be lost during transmission into valuable computing output.
'This system not only enhances the energy efficiency of space solar power stations but also reduces the need to send massive amounts of raw satellite data to Earth, easing the burden on terrestrial data centres and lowering overall energy consumption,' said the paper's lead author, Xiao Ge, who is also the company's chief technology officer.
NoSQLt is designed to be compact, hardware-agnostic, and cross-platform, making it ideal for constrained space environments. On hardware as basic as a desktop PC or a Raspberry Pi 4B, the platform has showed impressive throughput – handling 5.3 million and 1.15 million numerical data operations per second, respectively – proving its ability to deliver high performance with limited resources.
'Based on this platform, a space-based Internet of Things network could be developed to process satellite data in orbit and send only the final results to Earth,' Xiao said.
'Our goal is to build lightweight, high-performance orbital data centres that improve solar energy utilisation and provide computing support for future space exploration and scientific research,' he added, noting that more specific applications were already in development, though details remained confidential.
China's initiatives reflected a growing global interest in orbital computing. In Florida, Lonestar Data Holdings has proposed placing data centres on the moon. Meanwhile, Washington state-based Starcloud planned to launch a satellite data centre this month, with commercial operations targeted for mid-2026.
The European Union has also explored the concept of space-based data centres, but its road map includes a long wait: launch vehicles that meet the required specifications are not expected until 2035, with another 15 years projected for equipment deployment.
'In public disclosures, companies like Microsoft ... IBM and Space Compass have also expressed interest in building computing infrastructure in space,' Chen said. 'But based on available data, China's constellation of interconnected satellites appears to be leading in actual capability. No other country has demonstrated a 12-satellite interconnected system.'
Still, Chen urged caution. 'This is just an initial experiment. Many issues remain to be solved and improved upon. We should approach this rationally. Only after full deployment, testing and real-world use in orbit will we truly understand its potential.' – South China Morning Post