Google launches first Project Suncatcher AI chips into orbit

Google launched a prototype satellite carrying its Tensor Processing Units into orbit Thursday, beginning the first in-space test for Project Suncatcher, its research effort aimed at developing solar-powered AI data centers in space.

The refrigerator-sized spacecraft, built with satellite company Planet, lifted off from California aboard a SpaceX Falcon 9 rocket. It carries four of Google’s TPUs, specialized chips used for machine-learning workloads in the company’s terrestrial data centers.

Once Planet commissions the satellite, Google plans to run a version of its Gemma AI model in 15-minute intervals, allowing the hardware to cool between tests. The yearlong mission is intended to assess how the chips perform under the stresses of launch and the radiation and temperature extremes of low Earth orbit.

Project Suncatcher is exploring whether groups of solar-powered satellites could eventually operate as computing clusters. Google says satellites in suitable orbits can receive nearly continuous sunlight, with solar panels potentially generating substantially more power than on Earth.

The project remains at an early experimental stage. Cooling, radiation protection, launch capacity and cost are among the major technical constraints facing orbital data centers. Google plans to launch two more purpose-built satellites next year to test more demanding computing workloads and laser communications between spacecraft. Its longer-term concept envisions closely coordinated satellite networks that process AI workloads in parallel and transmit results to Earth.