Can AI infrastructure eventually move beyond terrestrial data centers? Google is planning an early test of that idea through Project Suncatcher, a research moonshot exploring whether solar-powered satellites could one day or later host scalable machine learning compute. The project’s first orbital experiment will place Google Tensor Processing Units, or TPUs, aboard a prototype satellite to study how the hardware plays beneath release stress, radiation, and extreme thermal situations.
Google announced Project Suncatcher in November 2025 as a long-term effort to investigate space-based AI infrastructure. In low Earth orbit, solar panels can generate up to 8-times more power than comparable systems on Earth due to the satellites can access sunlight nearly continuously. Google’s wider concept includes clusters of satellites wearing TPUs and communicating through high-bandwidth optical links.
Hardware Survival Comes First
Before attempting distributed AI workloads in orbit, Google requires to prove its hardware can survive the trip.
The prototype satellite, developed with Planet, is scheduled to fly aboard SpaceX’s Transporter-18 rideshare mission. During release, spacecraft can go through sustained acceleration near 10 g, while individual components can confront forces of 50 to 100 g. Google examined the satellite throughout three axes to simulate release vibration and reported that the hardware survived.
Radiation gives another challenges. Engineers exposed Trillium TPUs to proton beams at UC Davis’s Crocker Nuclear Laboratory while running AI workloads. As per Google, the chips tolerated a total ionizing dose more than what they would be anticipated to encounter during a 5-year mission.
Cooling AI Hardware Without Air
Thermal management can also prove even more tough. TPUs focus enormous heat in a small area, however space gives no airflow for conventional cooling.
Google is testing heat pipes and radiators designed to transport and release warmth in a vacuum. The team has already evaluated the system inside thermal vacuum chambers, however the orbital test will offer data that ground simulations cannot fully reproduce.
Building a Distributed AI Cluster in Orbit
Project Suncatcher’s longer-term layout rely on a numerous satellites working collectively. Future spacecraft could carry dozens of TPU chips and exchange data through laser links.
That needs extraordinarily particular positioning. Unlike long-distance space communication systems optimized for lower bandwidth, Google requires very high bandwidth across especially short distances between moving satellites. The company plans to test a two-satellite configuration in 2027.
Project Suncatcher highlights how AI infrastructure research is expanding beyond model architecture. Power, thermal engineering, networking, hardware reliability, and distributed systems ought to become as crucial to future AI scaling as accelerator performance itself.
