Meet Project Suncatcher, Google’s plan to put AI data centers in space (arstechnica.com)

Google has revealed Project Suncatcher, a moonshot to deploy clusters of AI accelerators (think TPUs) on solar-powered satellites to form distributed, orbiting data centers. The plan, outlined in a preprint, targets a dawn–dusk sun-synchronous low‑Earth orbit so satellites get near-constant sunlight—solar panels are up to eight times more efficient in orbit—helping address terrestrial power costs and site constraints as AI compute demand explodes. Google frames this as a long-term, scalable alternative to endless Earth-bound data-center expansion, building on the viability of modern satellite internet constellations. Technically, Suncatcher hinges on free-space optical links to network nodes into a single high-performance fabric: Google says intersatellite wireless must eventually operate at tens of terabits per second, though early terrestrial tests show bidirectional links of about 1.6 Tbps that could be scaled. The approach promises huge compute density and lower energy bottlenecks, but faces tough engineering hurdles—ultra‑high‑speed optical networking in orbit, thermal and radiation hardening of accelerators, deployment logistics and long multi‑year development cycles. If solved, orbiting TPU networks could reshape AI infrastructure by offloading power-hungry workloads from Earth while introducing new tradeoffs in latency, maintenance and costs.