Google contemplates putting giant AI installations in low-earth orbit

Google is exploring Project Suncatcher, aiming to place AI data centers in space powered by solar energy. This approach seeks to reduce the carbon footprint associated with AI cloud computing by utilizing clean, abundant energy in space. The project involves deploying Tensor Processing Units (TPUs), Google's AI chips, on solar-powered satellites in orbit. The strategic placement of these satellites offers significantly higher solar energy production compared to Earth-based systems. However, the endeavor presents significant hurdles, including radiation exposure that can damage sensitive electronics. Google has assessed the radiation tolerance of its current chips, suggesting they can withstand a five-year mission. Rapid, low-latency data links, capable of transferring tens of terabits per second, pose another challenge. Google is considering tightly clustered satellite formations to facilitate this and minimize station-keeping fuel consumption. The economic viability of launching TPUs into space is a key consideration, with Google projecting cost parity with Earth-based data centers by the mid-2030s. The company is currently engaged in preliminary research and plans initial trials to validate the concept. Google is collaborating with Planet to launch prototype satellites by 2027 for a "learning mission". This mission will assess the performance of AI models and TPU hardware in space, validating inter-satellite links for machine learning tasks.