China has launched the first 12 satellites in a planned 2,800-satellite orbital supercomputer network. The satellites, which the company built in collaboration with the Zhijang Laboratory and Neijang High-Tech Zone, will be able to process the collected data independently, rather than relying on ground stations, writes The Verge.
The satellites launched are part of ADA Space's Star Compute program, the first phase of the Three-Body Computing Constellation initiative. Each of the 12 satellites is equipped with an AI model with 8 billion parameters and is capable of performing 744 teraoperations per second (TOPS). Together, they provide a performance of up to 5 petaoperations per second (POPS), according to ADA Space.
Satellites have many times more power to run large language models than conventional computers, which in the case of Copilot+ computers require only 40 TOPS. The Chinese government's ultimate goal is to have a network of thousands of satellites that will reach 1,000 POPS.
The satellites exchange data with each other using laser communication at speeds of up to 100 Gbps and have a total storage capacity of 30 terabytes. All 12 satellites, launched by a Long March 2D rocket, carry scientific payloads, including X-ray polarization detectors to record short-lived cosmic phenomena such as gamma-ray bursts. They are also capable of creating 3D digital twins of data that can be used in emergency response, video games and tourism.
The benefits of a space supercomputer go beyond time savings. Unlike traditional satellite data transmission, which is slow and delivers less than 10% of the collected information to Earth, orbital processing centers can process data directly in space. They use the sun's energy and radiate excess heat into outer space, which significantly reduces energy consumption and carbon footprint. In the future, similar initiatives may appear in the United States and Europe.