3 min read
China has successfully launched and begun assembling the first 12 satellites of its ambitious “Three-Body Computing Constellation,” marking the world’s first space-based supercomputer network.
This inaugural deployment, which marks the start of a planned 2,800-satellite AI “Star Compute” initiative, was created by ADA Space in collaboration with Zhijiang Laboratory and the Neijiang High-Tech Zone. The program is meant to offer unmatched durability and on-orbit data processing capability.
First Launch and Supercomputer Capabilities
On May 18, China’s Long March 2D rocket sent twelve Star Compute satellites into low Earth orbit. With each satellite’s 744 tera operations per second (TOPS), local data analysis can go straight in orbit rather than raw data transfer to ground stations.
Chinese engineers achieved this by housing an eight-billion-parameter artificial intelligence model. The first batch can achieve five peta operations per second collectively. These are far more than the roughly forty TOPS needed for a modern personal artificial intelligence assistant. The whole constellation is meant to run 1,000 peta operations per second.
Supercomputer Interconnect and Shared Storage
The inter-satellite communication rates of the satellites are as high as 100 Gbps, forming a laser-linked lattice that shares 30 terabytes of distributed storage. This high-speed backbone enables each node to request and receive computation offload or data from peers in real time.
Onboard scientific cargoes consist of X-ray polarization detectors for the capture of transient cosmic events such as gamma-ray bursts. Also, 3D digital-twin generation capabilities are on-board for a variety of applications, including immersive tourism and emergency response.
Strategic Advantages and Environmental Benefits
By processing data in orbit, the system bypasses bandwidth bottlenecks and ground‑station availability constraints, ensuring that more than 90 percent of collected data is analyzed rather than discarded.
Harvard astronomer Jonathan McDowell notes that space‑based data centers can leverage unlimited solar power and radiate waste heat into space, significantly reducing energy costs and carbon footprints compared to terrestrial facilities.
Future Plans
China plans to launch additional Star Compute satellites in batches, scaling the network toward its 2,800‑unit goal over the next decade. Once complete, the constellation will offer global, low‑latency AI compute services for Earth observation, climate modeling, network security, and commercial use in gaming and virtual‑reality content delivery.
Experts think the U.S. would want to nosedive into building a supercomputer in space of their own following these initiatives. U.S. currently holds the title for housing the world’s fastest supercomputer on land. IBM’s Summit at Oak Ridge National Laboratory in Tennessee uses Power9 CPUs and NVIDIA Tesla V100 GPUs. Not only that but it has 4,068 servers powered by ten petabytes of memory working concurrently to process 200,000 trillion calculations per second i.e., 200 petaflops.
