Google plans orbital AI data centers powered directly by sunlight

Google has announced Project Suncatcher, an ambitious research initiative aiming to develop orbital AI data centers powered directly by solar energy. The project envisions constellations of satellites equipped with Google’s Tensor Processing Units (TPUs) operating in sun-synchronous low-Earth orbits to harness nearly continuous sunlight, enabling highly scalable AI computing beyond Earth’s energy and resource constraints. These satellites would be interconnected via high-bandwidth free-space optical links, potentially reaching multi-terabit per second data transfer rates, to form a tightly clustered “AI constellation” capable of handling large-scale machine learning workloads. Key technical challenges addressed include maintaining high data transmission rates between satellites flying just hundreds of meters apart using dense wavelength-division multiplexing (DWDM) and spatial multiplexing, as well as ensuring radiation resilience of the compute hardware. Google’s TPU v6e chips have demonstrated strong resistance to radiation in proton beam tests. The project is still in early research stages, with plans to launch two prototype satellites by early 202