A new compute tier is forming in orbit

Orbital AI compute is moving from roadmap to a usable test case, and it is becoming a third tier of capacity rather than a substitute for ground data centers. Terrestrial sites are running into the same walls: cheap power, available land, and cooling. The structural answer being tested in orbit is solar generation and vacuum heat rejection, with global coverage as a byproduct.

The Chinese bet is concrete because of who is making it. ADA Space, a Chengdu-based private firm, deployed the Qwen3 large language model to its orbital computing center in November 2025, turning a corporate roadmap into a working inference node in space. The company, working with Zhejiang Lab, launched the first twelve satellites of the Three-Body Computing Constellation in May 2025 and says it plans to build out a 2,800-satellite network by 2035, with 1,000 satellites online by 2030 and 100+ ground data centers connected underneath.

The reason this lands in Sichuan, not somewhere else, is the industrial chain. Sichuan is one of the few Chinese provinces with a closed-loop aerospace sector covering more than 80% of segments, supplying roughly 30% of China's satellite payload market, with a 2025 aerospace industry value near 50 billion yuan, or about 7.3 billion U.S. dollars. Chengdu alone hosts 200+ key aerospace enterprises.

The mechanism is straightforward: when one tier of compute hits physical ceilings, the next tier goes where energy, cooling, and land are not the binding constraint. The follow-on question is which industrial clusters can build that tier at all.