Zhijiang Laboratory’s space computing programme announced an ambitious timeline on the margins of a Beijing conference: 39 satellites are already in development, 10 satellites with on‑board embodied intelligence are slated for deployment in 2026, and a 100‑satellite “Three‑Body Computing Constellation” is targeted for completion in 2027. The project centres on a domestic large model called the “021 scientific foundation model” and couples it with the Nanhu computing framework and the Haina data hub to move AI models into orbit and run them on the satellites themselves.
The laboratory’s director of the space‑based centre, Li Chao, described capabilities the constellation will enable: satellite autonomy that shortens laser‑communications calibration from seven days to one, finer satellite control, multimodal data fusion and rapid model updates in orbit via a space digital‑twin system. The project is being pitched not only as a national capability but as an international public scientific product, promising flexible model deployment and real‑time updates while in orbit.
The announcement should be read against a broader global contest over where compute lives. Over the past five years governments and industry have raced to push sensors, networks and AI closer to data sources to reduce latency and overcome bandwidth limits. Moving foundation models and inference to satellites is the logical extension of that trend: on‑orbit compute can process imagery and signals before downlink, cut reaction times for monitoring and control, and reduce the volume of data that must be sent to Earth.
Technically, the programme bundles several demanding pieces: radiation‑tolerant and compact high‑performance compute nodes, robust on‑orbit software for model serving and updates, high‑bandwidth laser communications for intersatellite and ground links, and systems to manage thermal, power and fault tolerance in space. The claim that laser calibration time can be shortened so dramatically signals advances in automated calibration and link management, but achieving a 100‑satellite operational constellation within the stated timeframe will also require high launch cadence and supply‑chain resilience.
Strategically, the constellation straddles civilian and military utility. Faster, autonomous processing of multimodal data enhances weather, environmental and disaster monitoring, and it has clear applications to maritime surveillance, remote sensing and scientific research. Those civilian benefits sit alongside dual‑use implications for reconnaissance, targeting and hardened communications, and they will intensify questions about norms for on‑orbit AI, data sharing and the militarisation of space.
There are also governance and collision‑risk considerations. A rapid build‑out of hundreds of intelligent satellites raises concerns about orbital congestion and space‑traffic management. China’s framing of the system as an international public technology product is an olive branch that may meet interest from developing countries seeking low‑latency remote‑sensing services, but it will also prompt scrutiny from states wary of dual‑use applications and dependent critical infrastructure.
If Zhijiang Laboratory meets its timeline and technical goals, the constellation would mark a significant leap in how nations deploy AI at the edge of space. Success would demonstrate China’s ability to integrate domestic AI models, cloud‑edge frameworks and satellite hardware at scale. Failure or delays would point to the formidable engineering, regulatory and logistical gaps that remain in translating ambitious prototypes into resilient operational services.