SenseTime and Guoxing Aerospace plan a 1,000-satellite AI compute constellation

智东西 · wechat · 2026-07-21

A long article about the proposed “spatial computing constellation” announced by Guoxing Aerospace and SenseTime.

What they plan

The partners say they will build a space-based AI infrastructure in three stages: single-satellite validation, platform construction, and constellation networking.
The target by 2030 is a constellation of about 1,000 compute satellites with more than 10 petaflops of total spaceborne compute.
In 2026, the first “SenseTime” compute satellite is expected to enter orbit; by 2028 they aim for a hybrid space-ground commercial compute backbone.

Why it matters

The article argues space compute is not meant to replace terrestrial data centers, but to complement them where networks are weak or latency-sensitive: remote oceans, deserts, mines, disaster response, and cross-border scenarios.
It also highlights three use cases: on-orbit real-time inference in low-connectivity environments, on-orbit processing of remote-sensing data, and hybrid compute for high-value global workloads.

The technical and commercial challenge

Competition is shifting from single-satellite compute power to system-level capabilities such as inter-satellite coordination, space-ground scheduling, network stability, and long-term operations.
The harder problem is integrating onboard compute, lightweight model deployment, inter-satellite links, and unified orchestration into a repeatable engineering system.
The collaboration is framed as combining space hardware with SenseTime’s model/software stack so that orbital compute becomes a callable and schedulable resource inside a broader AI ecosystem.

More from Infra