China Aerospace Science and Technology Corporation published a five-year roadmap in February 2026 for a constellation of solar-powered AI data centers in low Earth orbit. The plan is embedded in China's 15th Five-Year Plan (2026–2030), which designates the "deep integration" of space and artificial intelligence as a national strategic priority. This is not a research proposal. It is a procurement and production schedule.
The target: 2,800 satellites operational by 2030, delivering 1,000 petaflops of aggregate compute. Production launches begin in 2027. The system is designed to process AI workloads in orbit and beam results to ground stations — a cloud architecture where the data center is in space and the end-users are on Earth.
Operator: China Aerospace Science and Technology Corporation (CASC). Target constellation: 2,800 satellites. Target compute: 1,000 petaflops. Production launches: beginning 2027. Full operational footprint target: 2030. Energy source: orbital solar arrays — approximately 5x more efficient than ground-based panels due to absence of atmospheric absorption.
China's data center expansion is constrained by the same energy bottleneck affecting every other major AI market. The national grid cannot deliver power fast enough to match compute demand. Local governments are rationing power to data center operators. The resource geography for AI in China is structurally disadvantaged compared to the United States, which has more flexible grid access in high-growth markets.
Orbital solar solves this. Solar panels in LEO receive continuous, unfiltered sunlight — no clouds, no night, no atmospheric scattering. Energy output is approximately five times higher per panel than ground-based equivalents at comparable latitudes. For China, moving the energy supply to space is not a speculative exercise. It is a workaround for a real constraint that is already limiting AI deployment on the ground.
The cooling problem for orbital data centers does not look like the terrestrial cooling problem. There is no water. There are no cooling towers. There are no CDUs, no heat exchangers, no chillers. Heat rejection happens passively through radiation to deep space — a heat sink at approximately 2.7 Kelvin with unlimited capacity.
The engineering challenge is moving heat from the chips to the radiative panels efficiently. Thermal interface materials, heat pipes, and vapor chambers do the work that plumbing does on Earth. The physics are favorable. The manufacturing and integration challenges at satellite scale are not trivial, but they are solvable with existing technology.
The US has no equivalent national program. StarCloud, the US startup building orbital data centers, raised $170 million at a $1.1 billion valuation — a meaningful number by startup standards and a rounding error relative to CASC's state-backed program. The race for orbital compute is not symmetrical.
The terrestrial cooling industry should file this under long-range signal, not near-term threat. The 2030 timeline is aggressive and depends on launch cadence that China has not yet demonstrated at this scale. But the policy commitment is real, the engineering rationale is sound, and the competitive pressure it creates on the US AI infrastructure ecosystem is direct. When China's orbital constellation starts delivering petaflops, it will not be running on CDUs and cooling towers.