China Prepares a National Industry to Bring AI Data Centers to Space

China wants the next wave of data centers not to be limited to land-based infrastructure. Beijing has approved the creation of the Space Computing Industry Innovation Center, an initiative designed to coordinate rocket and satellite manufacturers, semiconductor companies, AI firms, and universities around a new infrastructure: orbital computing networks capable of processing data directly in space.

The project is still in its early stages, but its strategic intent is clear. China is not just experimenting with a satellite; it is building an industrial framework aimed at accelerating space-based computing as a new layer of digital infrastructure. The idea is that part of the processing related to AI, Earth observation, communications, satellite IoT, or data analysis can be executed in orbit, reducing reliance on transmitting large volumes of data to terrestrial data centers.

An industry alliance led from Beijing

The new innovation center seeks to unify several components that have historically progressed separately: launch vehicles, satellite platforms, chips, compute payloads, AI models, control networks, and commercial services. According to official information released in China, the project identifies six major focus areas.

These include native space chips capable of withstanding heat, radiation, and high reliability standards; high-performance space computing payloads; satellite platforms and technical standards; energy-efficient AI models tailored to power constraints; integrated networks spanning land, cloud, and space; and new commercial models offering orbital computing power as a service.

Beijing University of Posts and Telecommunications is among the key players in the project, alongside aerospace and tech companies. This initiative aligns with recent developments such as the creation of a Chinese professional committee dedicated to space computing and the development of Beijing’s future Satellite Town, an industrial hub expected to be operational by the second half of 2026.

China’s approach contrasts with SpaceX. While Elon Musk’s company pursues a highly vertical integration model—covering rockets, satellites, manufacturing, networks, and potentially proprietary hardware—China appears to be constructing a national infrastructure. The goal is not just deploying satellites with chips but creating a whole industrial chain centered on orbital computation.

China’s response to SpaceX and Blue Origin

The timing is deliberate. SpaceX has proposed one of the most ambitious projects in the sector: a constellation of up to one million satellites designed as orbital data centers for AI workloads. The company has requested FCC authorization for a system of non-geostationary satellites between 500 and 2,000 kilometers altitude, featuring optical links and connection with the Starlink infrastructure.

Elon Musk has also revealed the preliminary design of the AI1 satellite—a platform intended to run AI workloads outside terrestrial power grids. According to published details, it would feature an average computational load of 120 kW, peaks of 150 kW, a deployed structure about 70 meters long, and deployable liquid radiators to dissipate heat in vacuum.

Blue Origin, meanwhile, has entered the race with Project Sunrise, proposing up to 51,600 satellites in heliosynchronous orbits between 500 and 1,800 kilometers. The company envisions a network of orbital data centers supported by optical links and its TeraWave high-capacity communication network.

The scale of these proposals warrants caution. Requesting authorization for large constellations does not guarantee full deployment. In SpaceX’s case, requesting high numbers might also be a regulatory strategy to retain flexibility, as was seen with Starlink. However, the industrial message is clear: multiple companies and governments are starting to view space as an extension of AI infrastructure.

Why put computing in orbit?

The core motivation is similar to the challenges faced by terrestrial data centers: AI requires energy, cooling, space, permits, electrical grids, water, chips, and interconnection capacity. Each large training or inference cluster demands physical infrastructure that cannot always be taken for granted.

In theory, space offers advantages. Solar energy is abundant and more consistent in selected orbits. There’s no need to occupy urban or industrial land. Some satellite-generated data, such as Earth observation images, IoT signals, or scientific data, could be processed in orbit before transmission to Earth. This could reduce downlink traffic and enable sending summaries or results instead of raw data.

For China, there’s also a geopolitical dimension. Orbital computing can enhance satellite communications, Earth observation, defense, IoT, navigation, scientific research, and future 6G networks. It can also help reduce dependence on foreign suppliers amidst ongoing semiconductor restrictions and U.S.-China technological competition.

This orbital project runs parallel to other terrestrial plans. China is developing a national AI data center network with an estimated investment of around 2 trillion yuan (about $295 billion), with a strong focus on domestic technology. Space-based computing won’t replace this infrastructure but could serve as a complementary layer for specific use cases.

The biggest challenge: cooling in a vacuum

The idea of a data center in space is often presented as an elegant energy solution. But the hard part isn’t just generating electricity; it’s dissipating heat. On Earth, data centers use air, water, liquid cooling, towers, heat exchangers, and industrial systems supported by a physically favorable environment. In space, there is no air to facilitate convection. Hardware must dissipate heat via infrared radiation.

This necessitates large radiators, redundant thermal systems, and highly efficient designs. SpaceX, for example, proposes deployable liquid radiators for AI1. Recent technical literature on space data centers emphasizes that thermal management, ground communication, latency, component reliability, and hardware lifespan are primary barriers.

Communication is also more restricted than energy. Terrestrial data centers move vast amounts of data within racks, cabling, networks, and to users. In orbit, links to ground and between satellites are limited by physical, regulatory, and economic factors. Many experts believe that practical uses will focus on processing data generated in orbit, filtering information, executing inference tasks, or serving specific workloads with tightly controlled bandwidth.

A race more promising than certain

Orbital computing is attractive because it tackles a genuine problem: terrestrial AI infrastructure faces energy and logistical constraints. But turning satellites into data centers isn’t just a matter of moving racks to space. On Earth, operators can swap servers, repair cooling systems, replace GPUs, expand capacity, and update networks. In orbit, every failure costs more, and every upgrade requires launching new hardware.

Economic uncertainties also exist. AI chips evolve rapidly—today’s satellite may lag behind the latest accelerators within a few years. The only way to compensate would be frequent launches, drastically reducing launch costs and designing modular platforms—yet both depend on manufacturing and launch capabilities still under development.

Environmental considerations remain. Although orbital solar energy can reduce some terrestrial energy demand, large constellations raise risks like orbital debris, light pollution, interference, controlled reentries, orbital slot occupation, and increased space traffic management complexity.

China’s goal: a complete supply chain

China’s distinct approach lies in its industrial strategy. Beijing isn’t relying solely on specific satellites but is instead working to establish standards, chips, platforms, AI models, and commercial services around space computing. Success could give China an advantage if it effectively coordinates universities, state enterprises, startups, semiconductor manufacturers, and space operators.

However, this strategy has limitations. A state-led approach can mobilize resources and accelerate standards but doesn’t guarantee technical efficiency or commercial viability. SpaceX’s extensive experience with mass deployment, rapid manufacturing, and constellation operation—via Starship and Starlink—sets a high bar. Blue Origin’s ambition, funding, and orbital infrastructure strategy are also significant. China offers large-scale industrial capacity, state-driven planning, and a growing commercial space market.

The race for orbital AI data centers has just begun, and it’s important not to conflate proposed projects with operational realities. Nonetheless, Beijing’s moves show that space computing is no longer a peripheral idea. The energy demands of AI, the competition among global powers, and the maturation of low Earth orbit constellations are pushing governments and companies to look beyond traditional ground-based data centers.

If this vision materializes, future digital infrastructure may extend beyond cloud regions, hyperscale campuses, or sovereign data centers. Part of it could operate overhead, powered by solar energy, connected via lasers, and designed to process data before it reaches Earth. The key questions are whether orbital physics, economics, and security will allow this promise to become reality rather than just another marketing race.

FAQs

What has China approved?
Beijing has approved the Space Computing Industry Innovation Center, a project to coordinate space companies, semiconductors, AI, and universities around space computing.

Does this mean China already has orbital AI data centers?
No. The initiative aims to develop the necessary industrial and technological chain. It’s not yet a deployed commercial network.

Why is there talk of competing with SpaceX?
Because SpaceX has requested authorization for a large orbital data center constellation and has revealed the AI1 design, intended to run AI workloads in space.

What are the main technical challenges?
Cooling in vacuum, radiation effects, launch costs, maintenance, chip lifespan, communication, latency, and space debris.