China sends supercomputer satellites into space

Could supercomputers move into space soon? China has promoted the first twelve satellites of his “Three-Body Computing Constellation” to space, which are equipped with AI chips and laser communication. There are advantages over conventional data centers – but also remarkable hurdles.

China’s network for data processing in the earth orbit

The Chinese company Ada Space has transported the first twelve satellites of a comprehensive orbital network to earth orbital network in cooperation with the Zhijiang Laboratory and the Neijang Hightech Zone. These satellites are the first components of the “Star Compute” program and represent the beginning of the planned “Three-Body Computing Constellation”, which is intended to include 2800 satellites.

The technical specifications of each individual satellite contain a AI model with eight billion parameters with a computing capacity of 744 teraflops (floating point operations per second). All in all, the twelve satellites achieve a combined computing power of five petaflops. For comparison: The most powerful German supercomputer “HAWK” at the Stuttgart (HLRS) maximum performance calculation has a capacity of 26 petaflops. Although the satellite fleet cannot currently compete with the most powerful terrestrial data centers, it offers specific advantages for certain fields of application.
View on YouTube

Efficiency potential of orbital data processing

The conventional method of satellite data processing has increasing limits. According to current estimates, less than ten percent of the data collected by satellites reach the earth’s surface due to limited transmission bandwidths and the limited availability of ground stations.

The new satellite network uses laser communication with data rates of up to 100 gigabit per second and has a common storage capacity of 30 terabytes. Due to data processing directly in the orbit, the satellites can already transmit and thus substantially reduced data records to the earth.

Resource efficiency as a further aspect

According to information from South China Morning Post Conventional data centers could consume more than 1000 terawatt hours of electricity annually by 2026 – a value that, for example, corresponds to the entire electricity consumption of Japan. In 2022, Google alone needed around 19.7 billion liters of water for cooling purposes in its data centers. Jonathan McDowell from the Harvard University points out that orbital data centers can use solar energy directly and can dissolve thermal energy into space, which potentially reduces energy requirements and environmental effects.

Regardless of the advantages, space data centers also face considerable challenges. Maintenance and repair work on the hardware in orbit are associated with extreme effort or practically inconvenient. Radiation exposure can lead to technical disorders. In addition, the operating time of satellites is limited, which requires regular replacement missions.