China’s Orbital Datacenter Is Taking Shape Across Multiple Satellite Programs

China’s approach to an orbital datacenter is emerging through several satellite programs rather than one giant platform. One tests AI processing and links between satellites; others explore how to manage workloads and software or move data quickly to Earth. By September 2026, the evidence showed useful individual capabilities, but not a unified, production-scale space cloud.61146

Three-Body: AI compute linked across satellites

The Three-Body Computing Constellation launched its first 12 satellites in May 2025. After months of testing, Zhejiang Lab reported that the group had deployed 10 AI models and established inter-satellite networking. Published descriptions cite an 8-billion-parameter model and a stated 744 trillion operations per second per satellite; those figures describe reported capabilities, not a measured fleet-wide service level.113216

Developers also reported that two satellites maintained a laser link for more than eight days with 99.99% availability. That is a noteworthy link test, but it does not establish equivalent performance across the whole constellation. Plans for thousands of satellites and much greater aggregate computing power remain plans, not the capacity of the initial batch.441

Lijian-1 Y18: new nodes, with workload sharing still to prove

A September 20, 2026 Lijian-1 Y18 launch placed nine satellites in orbit, including Qinling-1 and Qinling-2. Reporting on the newly launched Chaozhisuan-1 says it had begun processing Earth-observation data in orbit, including data received from other satellites over an optical link.1345

Processing another spacecraft’s data is a step toward distributed computing. It is not, by itself, evidence that the new payloads operate as a continuously scheduled, shared computing cloud. The same launch also carried PEGA-SUS1, described as a test of combining communications, sensing, computing and intelligent processing on a satellite.344

TianSuan: managing services and testing space–ground networking

TianSuan addresses the software and operations side. Its researchers documented a 5G core-network deployment test aboard TY20, including signaling with a terrestrial private 5G network. That was a networking experiment, not a demonstration of consumer 5G coverage from orbit.17

By August 2026, TianSuan’s developer said its space-computing cloud had begun offering regular experimental in-orbit services. The reported system coordinates task submission, deployment, execution and monitoring across satellite platforms, space servers, ground stations and terrestrial datacenters. TianSuan-2’s listed test objectives include satellite inference using KubeEdge, a lightweight 5G core-network update and other edge-computing services; a list of objectives should not be mistaken for completed on-orbit results.4623

AIRSAT-02: a faster route back to Earth

The Chinese Academy of Sciences reported an AIRSAT-02 satellite-to-ground laser experiment above 100 Gbps, reaching a 120 Gbps peak with a stable link. A separate account says the higher rate followed an on-orbit software reconfiguration without changing the satellite hardware.2122

That result concerns a downlink to a ground station. It does not show that every computing satellite has a 120 Gbps connection to its peers, or that an orbital datacenter can sustain that rate from task submission through processing and delivery.2122

What remains unproven

Together, the programs illustrate the proposed architecture: process data near where it is collected, assign and manage work across spacecraft and ground systems, and return results over high-capacity links.34621 The cited reports do not establish sustained fleet-wide throughput, reliable distributed scheduling at large scale or one integrated service combining all four programs. Reported test rates and proposed constellation sizes are therefore better read as milestones and ambitions than as specifications for a datacenter already operating in orbit.44146