Hundreds of kilometers above the Earth, a constellation of Low Earth Orbit (LEO) satellites is calculating the movement of fish schools for fishermen at sea. Instead of transmitting raw images of the entire sea area back to the ground, they now "compute" the imagery in orbit first, sending back only the most valuable fishery information to Earth. This capability of "compute first, transmit later" comes from a newly emerged species: the Space Computing Cloud.
Recently, the world's first Space Computing Cloud providing on-orbit testing services to the public, led by the construction of Beijing University of Posts and Telecommunications (BUPT), has officially begun to provide on-orbit computing power services on a regular basis. In simple terms, it opens up the computing power on satellites just like cloud services: institutions that need computing power no longer have to build and launch their own satellites; they only need to "one-click install" their programs in orbit, and the satellites will do the rest. If over the past two decades, satellites were merely "eyes" suspended in space, now they are beginning to grow "brains".
01. Why Computing Power Needs to Go to Space
Initially, satellites performed the job of "eyes": capturing photos, transmitting them back, and processing them on the ground. This assembly line has become increasingly strained—data downlink depends on the "whims" of ground stations, which have limited receiving capacity. A massive amount of data generated in orbit simply cannot be transmitted back, leaving it to accumulate uselessly in space. "The dynamic activities of marine fish schools are highly valuable economic data. However, in scenarios requiring rapid response and analysis, the traditional satellite data processing model faces efficiency bottlenecks," said Xing Ruolin, founder of Yiwei Space, a space operating system R&D enterprise.
Rather than moving data back to Earth for computation, it is better to send computing power into space. The Space Computing Cloud is composed of an on-orbit service platform, a ground station service platform, and an operation service platform. It integrates task orchestration, computing power scheduling, and service supply among satellite platforms, space servers, ground stations, and ground data centers, achieving end-to-end automation. For users, completing a cycle of "task application—deployment—operation—monitoring—result transmission" is like using a computer located far away in space. Professor Wang Shangguang from BUPT explained it this way: space computing power serves ground users on one end, acting as a supplement and extension to ground computing power; on the other end, it serves space and interstellar users, providing an alternative computing power supply independent of Earth.
This system is not just a theoretical concept. It operates on the "Tiansuan Constellation" initiated by BUPT in 2021—the world's first space computing constellation. To date, it has cumulatively deployed 16 LEO satellites in orbit, built 7 ground stations, supported over 60 institutions in conducting on-orbit experiments, and attracted researchers from more than 50 countries and regions to participate in ecosystem co-construction. In official terms, this means China's space computing power is advancing from "single-satellite experiments, single-time verification, and single-technology breakthroughs" to a new stage of "platform-based operation, service-oriented supply, and ecosystem development"—leaping from "whether it can compute" to "how well it is used".
02. The Distance from "Eyes" to "Brains"
Once fully operational, the value of sending computing power to space begins to materialize. Scenarios such as on-orbit big data computing, 6G communication, and distributed storage have been tested in turn, completing hundreds of space computing power calls cumulatively, serving over a hundred users, and achieving a space computing power service coverage rate of over 10%. For remote sensing, the change is nearly revolutionary: satellites compute while capturing images, completing disaster assessment and image interpretation in orbit, and transmitting only the conclusions back to the ground, thereby buying precious time for emergency rescue.
The most telling metric is the energy efficiency ratio of large model inference: 10 tokens/joule—meaning every 1 joule of energy consumed can complete the inference of approximately 10 tokens. Space computing power is constrained by on-board power supply, payload volume, and the space environment, making energy conservation far more difficult than on the ground. Wang Shangguang pointed out that this level of energy efficiency means that, under constrained computing power, the AI inference efficiency per unit of energy consumption is sufficient to support the special requirements of lightweight, long-duration, and continuous operation in space, placing it at an advanced level among similar on-orbit computing power explorations both domestically and internationally.
The significance of sending computing power to space also lies in cultivating new growth points for satellite internet, 6G, and emergency communication. Taking emergency communication as an example, when disasters strike, ground base stations may fail, while computing power in orbit remains unaffected—satellites serve as both communication links and computing hubs. Currently, relying on the continuous service capabilities formed by this system, new service boundaries and application scenarios are being expanded for various satellite systems.
In official statements, this capability can be summarized into three things: building a space operating system ecosystem to turn "one-time on-orbit experiments" into reusable and iterable platform-based services; forming a cloud-native software runtime foundation tailored for the space environment; and achieving collaborative scheduling of satellite payloads, laying the foundation for on-orbit real-time processing and autonomous decision-making. Translated into plain language, this means satellites are evolving from "single-task carriers" to "networked intelligent nodes," and the space information industry is shifting from "delivering hardware" to "continuously delivering services." Eyes are for seeing, and brains are for thinking; the generational gap between them represents the imagination space for the next leg of the entire industrial chain.
03. Installing an "Operating System" in Space
When it comes to sending computing power to space, the real challenge is not just launching a few satellites, but making them as user-friendly, schedulable, and iterable as clouds on the ground. At the bottom layer of the Space Computing Cloud is an independent, controllable, open, and compatible "space operating system": users can "one-click install" their programs on the cloud to call upon the computing power and payloads of satellites, without having to reinvent the wheel. For developers, this means the two tasks of "writing programs" and "building satellites" are completely separated—the latter has an extremely high threshold, while the former is accessible to everyone. This is highly similar to the early internet—first comes the open platform, followed by the wild growth of applications.
Around this system, the industrial chess game is being played out rapidly. In June this year, the Beijing Space Computing Power Innovation Center was unveiled and began operations in the Beijing Satellite Town. Concurrently, the country's first open-source standard system for space operating systems was released, followed by the launch of the beta version of the space computing power service—from "building the platform" to "setting standards" and then "running services," the three moves were made almost simultaneously. "Currently, various commercial aerospace enterprises are building their own infrastructure. We expect the construction of the innovation center to pool these resources into a larger-scale infrastructure," said Xing Ruolin. His expectation is also the shared expectation of this track: to truly commercialize space computing power.
Supporting this momentum is a well-formed industrial soil. Beijing has gathered over 300 commercial aerospace enterprises, accounting for more than half of the national total, with the industrial scale breaking through 100 billion CNY; from 2025 to the first quarter of this year, the disclosed financing in the domestic commercial aerospace primary market reached approximately 21.968 billion CNY, with Beijing alone accounting for 12.15 billion CNY. Amidst the roar of building rockets and launching satellites, "selling computing power" is growing into a new business.
04. The "Next Ticket" on the Same Starting Line
The window of opportunity for space computing power has opened almost simultaneously worldwide. In Wang Shangguang's view, space computing power is a crucial strategic capability for seizing the commanding heights of the next-generation aerospace information infrastructure. China is on the same starting line as pioneers like the United States, with no generational lag. "The realization of normalized services by this Space Computing Cloud holds the significance of being at the forefront." Looking further into the future, the Moon, Mars, and even deeper deep-space explorations will all require computing power to pave the way first.
Returning to the scene at the beginning. Fishermen do not care about the name of the satellite above their heads; they only care whether the movement of fish schools can be delivered to them faster and more accurately. When computing power is truly moved to space, satellites will no longer be "cameras" suspended in the air, but rather "computing workstations" that can be scheduled at any time.
Today, we may still have a long way to go before achieving "a space cloud in everyone's hands," but that door has already been pushed open just a crack.
- End -