A Beijing startup wants to put AI computing in orbit — starting with 210 satellites

There’s a hard ceiling on how much AI computing the planet can support. Power grids max out. Data centers consume entire towns’ worth of electricity. Cooling towers stretch into the sky. Deep Space Matrix, a Beijing-based startup founded by Zhang Weijie, believes the only way past that ceiling is to go through it — straight up, into orbit.

At the 2026 World Artificial Intelligence Conference on Sunday, the company unveiled the Star Ring Project, a satellite constellation purpose-built for space-based AI computing. Phase one calls for roughly 210 low-Earth orbit satellites, each operating at a standardized altitude of 450 kilometers.

Zhang laid out the logic bluntly in the company’s announcement: AI competition ultimately comes down to compute, and as large-scale AI agents go mainstream, ground-based infrastructure is running into real physical constraints — power availability, land costs, construction timelines, heat dissipation, and network capacity. Moving compute off-world, he argued, isn’t science fiction anymore. It’s an industrial necessity that the AI agent era demands be taken seriously.

That’s a bold claim, but the company isn’t starting small. The first 210 satellites will form what Deep Space Matrix calls a “communication-remote sensing-compute integrated verification constellation.” Once expanded, the constellation is designed to cover roughly 99.8 percent of the globe with a regional revisit time of five to ten minutes — fast enough for high-frequency dynamic monitoring and moving-target tracking.

All satellites in the constellation share a common orbital altitude of 450 kilometers — a deliberate design choice to maximize compatibility with existing infrastructure. The company says this altitude also allows the constellation to interoperate with the 500-to-800-kilometer orbital range used by most commercial remote-sensing satellites today.

The roadmap stretches across five phases. First comes the initial verification constellation, proving out the combination of communication, remote sensing, and on-orbit compute. From there, the company plans to link satellites across layers for distributed computing, then gradually expand the ring, adding more AI-capable nodes until the network reaches critical mass. The ultimate vision spans thousands to tens of thousands of satellites forming a coordinated computing mesh in low Earth orbit.

The project is still early. A 210-satellite constellation costs serious money and time, and Deep Space Matrix hasn’t disclosed its funding or launch timeline. But the core thesis — that AI compute will eventually need to live in space — is getting harder to dismiss. Amazon’s Project Kuiper and Starlink have already shown that large LEO constellations are technically and commercially viable. Deep Space Matrix wants to prove they can think, too.