China tests AI satellite swarm for space-based computing

by Riko Seibo
Tokyo, Japan (SPX) Feb 16, 2026
China has advanced its vision for space-based computing with a satellite network that deploys 10 artificial intelligence models in orbit and demonstrates inter-satellite networking capabilities.

Developed by Zhejiang Lab in cooperation with international partners, the experimental space computing network is built around the Three-Body Computing Constellation, a planned fleet of more than 1,000 satellites designed to deliver large-scale processing power in space.

The first 12 satellites of the constellation were launched in May 2025, marking the start of an on-orbit test campaign that has now run for nearly nine months.

During this period, mission controllers verified key functions including crosslink networking between spacecraft, distributed computing across the constellation, deployment of AI models on orbit and operation of scientific payloads.

Among the payloads are two large space-based AI systems: an 8-billion-parameter remote sensing model and an 8-billion-parameter astronomical time-domain model, which rank among the most capable AI models currently operating in orbit.

In November 2025, the remote sensing model was used to carry out an infrastructure survey over an area of 189 square kilometers in northwest China, automatically identifying features such as stadiums and bridges even when they were covered by heavy snow.

For astrophysics, two satellites carry cosmic X-ray polarization detectors that work with an onboard AI model to classify gamma-ray bursts in real time, achieving an accuracy rate of 99 percent while sharply reducing the amount of data that must be transmitted to the ground and processed on Earth.

The team has also demonstrated inter-satellite links among six spacecraft in the fleet, a critical milestone for creating a functional space-based computing network capable of routing and sharing data in orbit.

According to Zhejiang Lab, once the full constellation of more than 1,000 satellites is deployed, the system is expected to provide an aggregate computing performance of around 100 quintillion operations per second.

With this architecture, part of the data collected in space could be processed directly on orbit and then delivered to end users in near real time, reducing latency and easing the burden on ground-based data centers and communications links.

Li Chao, a researcher at Zhejiang Lab, said the computing constellation concept points toward a future in which satellites handle more of the data-processing workload themselves, enabling new applications in deep space exploration, smart city management and large-scale natural resource surveys.