China Launches First 12 Satellites in Plan for a 2,800 Satellite Orbital AI Network

A computing network takes shape above Earth

China launched 12 satellites in May 2025 from the Jiuquan Satellite Launch Center as the first phase of the Three-Body Computing Constellation, a planned network of 2,800 satellites designed to process information directly in orbit. Led by aerospace startup ADA Space and the state backed Zhijiang Laboratory, the project combines artificial intelligence, laser communications and scientific instruments in an ambitious attempt to build a supercomputer in space.

As of August 4, 2025, the first launch marks the beginning of a project expected to develop over several years. Its planned computing capacity is 1,000 peta operations per second, or one quintillion operations per second. The project presents that target as potentially exceeding the capabilities of leading computers on Earth, including the United States' El Capitan system.

The immediate change is where satellite information can be processed. Traditional satellites mainly collect data and send it to Earth for analysis. The new constellation is designed to analyse that information before transmitting it, allowing the network to send useful results instead of large quantities of unprocessed material.

Emergency response, Earth observation, national defence and scientific research are among the proposed uses. The same computing infrastructure could support commercial services, from urban planning to gaming and tourism.

China's initiative also places orbital computing within the wider competition over artificial intelligence and digital infrastructure. Its scale and ambition are presented as unprecedented, with the first 12 satellites forming an initial step toward a much larger system rather than the completed constellation.

What the satellites carry

The Three-Body Computing Constellation takes its name from Liu Cixin's science fiction novel "The Three-Body Problem", which explores complex gravitational systems and cooperation between civilisations across space. The name reflects the challenge of making many separate machines work together as a distributed computer.

Each satellite is equipped with an AI model containing 8 billion parameters. Parameters are the internal values that a model uses when processing information. The satellites are intended to use their models for tasks such as recognising images, identifying unusual patterns and making predictions from data.

The stated computing capability is up to 744 trillion operations per second, or TOPS, for an individual satellite. The initial group of 12 satellites is described as collectively reaching 5 peta operations per second, or POPS. These figures describe how many computational operations the equipment can perform in a second.

Laser links between satellites can transfer data at up to 100 gigabits per second. Each satellite also carries 30 terabytes of storage, allowing information to be held across a distributed network in low Earth orbit.

Scientific equipment includes X-ray polarization detectors for astrophysical research. Other tools support Earth observation and the creation of three dimensional terrain maps. Together, the computing hardware, communications links and instruments are designed to make the satellites active participants in analysing information, rather than simply collecting it.

Why analyse information in orbit?

Less than 10 percent of the data collected by satellites currently reaches Earth, according to the figures underlying the project. Limited transmission capacity and a shortage of ground stations mean that substantial amounts of potentially useful information are lost or delayed.

The constellation seeks to address that bottleneck through edge computing in space. Edge computing means processing information close to where it is collected. Here, the satellite would analyse data in orbit and transmit the most relevant results, reducing the volume that needs to travel to a ground station.

For emergency services, faster analysis could shorten the interval between detecting an event and receiving useful information about it. For satellite operators, sending processed results could reduce transmission costs, save energy and ease pressure on infrastructure on Earth.

The project also envisages services for regions with limited terrestrial infrastructure. A network operating above Earth could support access to satellite information across national boundaries, creating opportunities for development and international cooperation.

Jonathan McDowell, a Harvard astronomer and space historian, described the attraction of moving data processing into orbit:

Orbital data centres can use solar power and radiate their heat to space, reducing the energy needs and carbon footprint. This is a very fashionable concept right now, and we can expect similar initiatives from the United States and Europe as the technology proves viable.

How separate satellites become one computer

The constellation uses a distributed mesh network. Each satellite acts as a node, or an individual part of the larger system, sharing information and computing tasks with other satellites through laser links.

This architecture is intended to provide redundancy, scalability and resilience against failures. Redundancy gives the network alternative resources when one part is unavailable. Scalability allows more satellites to be added as the constellation grows. The objective is for many machines to function together as a single large computer.

The AI models are intended to handle image recognition, anomaly detection and predictive analysis. Those capabilities support proposed services such as rapid Earth observation, digital replicas of physical environments and the detection of events in space.

Solar panels provide power, while heat is radiated into space. The design uses the vacuum of space for passive cooling, with the aim of avoiding the large energy demands and water consumption associated with cooling data centres on Earth.

The combination of computing, communications and scientific instruments is central to the project. Satellites must exchange data and share work effectively for the constellation to support a broad range of tasks across industries and scientific disciplines.

From disaster response to digital terrain

Earth observation is one of the network's main proposed uses. By processing information in orbit, the constellation is intended to produce updated digital twins of terrain. A digital twin is a virtual replica of a physical environment that can be used for planning, simulation or immersive experiences.

Agriculture, urban planning, environmental monitoring and disaster response could all use these digital representations. The aim is to give people a more current picture of conditions on the ground without waiting for every piece of raw satellite data to reach Earth first.

Emergency services could use rapid analysis to detect and respond to earthquakes, floods and wildfires. In such situations, the practical value would come from receiving useful information sooner, allowing responders to assess affected areas and organise their work.

National defence is another proposed application. Processing intelligence in real time could give military operations and surveillance a strategic advantage. The same underlying technology therefore has both civilian and military uses, making questions about access, security and responsibility part of the project.

Commercial applications include gaming, tourism and smart city development. Three dimensional digital twins could support entertainment and immersive experiences as well as emergency management. The project envisages processed satellite data becoming a resource for businesses rather than remaining solely a tool for governments and researchers.

Scientific instruments and the AI ambition

The satellites' X-ray polarization detectors are intended to support studies of energetic cosmic events, including gamma-ray bursts. These instruments give the constellation an astrophysics role alongside its work observing Earth.

Onboard analysis and shared computing resources could support the detection and study of cosmic phenomena. Scientific research is therefore part of the network's purpose, rather than an incidental use of equipment built only for commercial services.

Wang Jian, director of Zhijiang Lab and a member of the Chinese Academy of Engineering, described the project's strategic vision:

Space is the new frontier for AI development. By bringing computational power into orbit, we can overcome the limitations of traditional systems and unlock new possibilities for science, industry, and society.

The eventual target of 1,000 POPS explains the scale of that ambition. The first 12 satellites are described as delivering 5 POPS, while the completed network is intended to support far greater computing capacity. The project seeks to rival or surpass powerful terrestrial systems through a computer distributed across orbit.

The energy argument behind orbital computing

Growing demand for AI, big data and cloud services is increasing the electricity needs of data centres on Earth. The International Energy Agency estimates that worldwide data centres could consume more than 1,000 terawatt hours of electricity annually by 2026, an amount equivalent to Japan's entire national electricity consumption.

Cooling also uses large quantities of water. Google used nearly 20 billion litres of water in 2022 just to cool its data centres, illustrating another resource demand associated with large computing facilities.

The orbital network is presented as a potential response to these pressures. Its environmental case rests on satellites harnessing solar energy continuously and dissipating heat into space, reducing reliance on electricity and water for cooling on Earth.

The project aims to lower the carbon footprint of computing and establish a more sustainable form of digital infrastructure. Solar power and passive cooling are central to that argument, alongside the energy savings sought by transmitting processed information rather than larger volumes of raw data.

Eric Schmidt, the former Google chief executive who has advocated for orbital data centres through his investment in Relativity Space, linked the concept to growing computing demand:

The demand for computational power is so large that data centers could consume as much as 96 gigawatts globally by 2030. Moving data processing into space could be a game-changer for the industry.

Jonathan McDowell, the Harvard astronomer, also pointed to the environmental attraction and the prospect of similar efforts elsewhere:

We can expect the United States and Europe to follow suit as the technology matures. Orbital data centers could significantly reduce the environmental impact of global computing.

China's technology push and international competition

The United States and Europe have conducted limited tests of orbital computing. China's deployment is described as the first operational attempt at this scale, placing the country at the forefront of an emerging competition over AI processing in space.

A large orbital computing network could allow China to help shape technical standards and influence the development of global data infrastructure. The initiative brings together advances in satellite technology, AI and communications, areas increasingly central to competition among major powers.

China's domestic technology sector is also expected to benefit. The project is intended to support development zones such as the Neijiang high-tech zone in Sichuan province, where satellite data could feed businesses and services focused on smart satellites, AI and digital innovation.

ADA Space and Zhijiang Lab have expressed an intention to build globally accessible and sustainable orbital infrastructure. China has presented international collaboration, particularly with countries in the Global South, as part of the project's direction and invited other organisations to participate.

Such participation could create new alliances around satellite services and computing resources. At the same time, the network's defence applications place it within a wider contest over control of information and technology in space.

Deployment brings engineering and governance challenges

Building and maintaining a constellation of 2,800 satellites requires reliable launch capabilities, robust engineering and coordination across a large network. The project is expected to proceed over several years, with successive phases adding capabilities. No completion date or date for the next launch was given.

Space debris and orbital congestion are among the central concerns. As more countries and companies pursue satellite networks, managing traffic and coordinating activity become more demanding. The project adds to discussion about allocating orbital resources and ensuring access to them.

Data security and privacy are also concerns because the satellites would process information used by governments, businesses and other organisations. Their potential use for both civilian services and military operations makes transparent governance and responsible use especially relevant.

International cooperation and new frameworks may be needed to manage space traffic, provide equitable access to orbital resources and prevent the weaponization of computing in space. The initiative has drawn both admiration and concern because its scale connects technical progress with questions about control and responsibility.

The milestones and dates framing the project are:

• 2022: Google used nearly 20 billion litres of water to cool its data centres.
• May 2025: China launched the constellation's first 12 satellites from Jiuquan.
• August 4, 2025: The project remained in its initial phase, with a planned total of 2,800 satellites.
• 2026: The International Energy Agency's projection places annual global data centre electricity consumption above 1,000 terawatt hours.
• 2030: Eric Schmidt's cited projection puts global data centre power demand at as much as 96 gigawatts.

The first launch demonstrates China's commitment to orbital computing. Delivering the wider vision will require the network to grow while addressing the engineering, security and coordination demands that come with operating thousands of connected machines above Earth.

Key Points

• China launched the first 12 Three-Body Computing Constellation satellites in May 2025, with 2,800 planned.
• ADA Space and Zhijiang Laboratory lead the project, which processes satellite data directly in orbit.
• Each satellite has an AI model with 8 billion parameters, up to 744 TOPS of computing capability and 30 terabytes of storage.
• The first group is described as delivering 5 POPS, with a completed network target of 1,000 POPS.
• Laser links can transfer data between satellites at up to 100 gigabits per second.
• Proposed uses include disaster response, Earth observation, defence, astrophysics and commercial digital twins.
• Solar power and heat radiation into space underpin the project's environmental ambitions.
• Deployment, space debris, data security and international governance remain central challenges.