A satellite used to be a camera with a downlink. Wang Jian wants it to be a computer. The man who built Alibaba Cloud now directs Zhejiang Lab, and the project he is associated with puts AI models in orbit instead of waiting for images to reach the ground.
Key takeaways
- Wang Jian (王坚) founded Alibaba Cloud (阿里云) and was elected an academician of the Chinese Academy of Engineering (中国工程院) in 2019; he has served as director of Zhejiang Lab (之江实验室) since 2023.
- Zhejiang Lab, established in Hangzhou in 2017, is a non-profit new-type research institution organised around intelligent computing (智能计算).
- The Three-Body Computing Constellation (三体计算星座) launched its first 12 computing satellites on 14 May 2025. By July 2026, 20 AI models had been deployed in orbit, according to state media reporting.
- On 24 July 2026, a further launch added two satellites, including the first in-orbit test of the lab's embodied-intelligence (具身智能) satellite model.
- Wang has said the lab aims to work with satellite companies on a layout of more than 50 computing satellites in 2026 — a stated goal, not a completed count.
From cloud computing to computing in space
Wang's training is unusual for a computing pioneer. He studied psychology at Hangzhou University, wrote a master's thesis on human-computer interaction and multi-channel interfaces, and then moved into distributed systems. He joined Alibaba in the late 2000s, became group chief technology officer in 2012, and led development of Apsara (飞天), the distributed computing platform that became Alibaba Cloud.
That background explains the through-line. On the ground, his argument was that computing should be reorganised around data rather than machines. In orbit, the argument is similar: a satellite that can process its own observations is more useful than one that must transmit everything back to Earth, especially as distances grow.
He was elected to the Chinese Academy of Engineering in 2019 and took over Zhejiang Lab in 2023. The lab is a provincial-level research institution with a non-profit brief; its own materials say it does not manufacture satellites or compete with market players. That positioning matters for how the project should be read — it is closer to infrastructure than to a product launch.
What "intelligent computing" means here
The lab defines intelligent computing by three traits: computationally intensive, data driven and model based. Its stated motto is "Computing for Innovation". In practice this covers everything from liquid-cooled computing clusters to software that decouples model training from GPU hardware.
The same idea, applied to orbit, is captured in a phrase the lab uses: sending computing power, satellites and models into space. The constellation is named after the three-body problem, on the grounds that coordinating many independent objects creates complex interactions. That is a description of an engineering challenge, not a claim about new physics.
The constellation, stage by stage
The first 12 satellites went up on 14 May 2025. Reporting on the lab's work describes an 8-billion-parameter AI model aboard the constellation and a move from six-satellite networking toward whole-orbit interconnection across twelve.
By July 2026, according to reporting from a field visit by China Youth Net (中国青年网), 20 AI models had been deployed for applications including deep-space exploration, collaborative water-resource mapping, atmospheric monitoring and autonomous wildfire detection. The number of models and the number of satellites are different measures, and the reporting does not treat them as interchangeable.
A launch on 24 July 2026 carried two more satellites. One, the Jitianxing A-04, was reported to carry the lab's embodied-intelligence satellite model for the first in-orbit verification. The model is designed to let a satellite manage its own health, plan tasks and execute instructions, moving the constellation from single-satellite intelligence toward multi-satellite coordination.
Wang has been quoted saying that in 2026 the lab would work with satellite companies on a layout of more than 50 computing satellites. That is a target. It should not be read as a current count.
Why the idea travels
The practical case is bandwidth and latency. Weather satellites currently send data down for processing; doing part of that work in orbit could make forecasts faster and reduce the volume of data that must be transmitted. The same logic applies to disaster monitoring in places with poor ground infrastructure.
Wang has a way of framing this that is worth quoting, because it explains the bet better than a specification sheet. In remarks reported in 2026, he compared the moment to the early days of electricity — when no one could imagine washing machines or refrigerators — arguing for building the computing layer first and letting applications appear later.
For non-specialists, the useful distinction is between a satellite that takes a picture and a satellite that can decide what the picture means. The second is what this project is testing.
Honest limitations
Facts here come from Zhejiang Lab's own website (which republished People's Daily reporting), China Youth Net (中国青年网), Hangzhou Daily (杭州日报) and China Energy News (中国能源网). Satellite and model counts are reported at specific dates by the lab or state media; I have not independently verified orbital data or the performance of any model. Wang's statements are reported remarks, not audited forecasts. Zhejiang Lab is a non-profit research institution and does not publish the sort of financial disclosure a listed company would, so there is no revenue figure to check these claims against.
What readers can do now
- When you read about "space AI", ask whether inference happens on the satellite or the data is still sent to Earth — the two are often conflated.
- Track satellite count and deployed-model count as separate numbers; they measure different things and grow at different rates.
- Treat a stated target such as "more than 50 satellites" as a plan until an independent launch record confirms it.
