Particle.news
Download on the App Store

Google Outlines Space-Based AI Data Centers, Citing 1.6 Tbps Demo and Orbital Modeling

The company says large-scale viability depends on launch prices falling below $200 per kilogram.

Overview

  • Project Suncatcher proposes clusters of solar-powered satellites in sun-synchronous low Earth orbit linked by laser optical connections to operate as a distributed data center.
  • Google’s simulations at roughly 650 kilometers altitude indicate tightly clustered satellites could be maintained with modest station-keeping maneuvers.
  • A bench-scale free-space optical link achieved 1.6 terabits per second using multiplexing techniques, with future systems needing tens of terabits per second for data center-class workloads.
  • Proton-beam tests found Trillium TPU v6e chips resilient, with high-bandwidth memory showing irregularities only near 2 krad compared with an expected shielded five-year dose of about 0.75 krad.
  • Google and Planet plan a two-satellite learning mission by early 2027 to test core hardware and in-orbit operations.