Google is preparing the first in-orbit test for Project Suncatcher, its research effort to explore whether machine-learning hardware could eventually operate in solar-powered satellite clusters. A prototype developed with Planet is scheduled to launch next week on SpaceX’s Transporter-18 rideshare mission, according to a new project update from Google.
The immediate goal is much smaller than an orbital data center. Google wants to learn whether its Tensor Processing Units can survive launch and function through the radiation, temperature swings and cooling constraints of low Earth orbit. The mission will return operating data that can shape later Suncatcher hardware.
First, prove the chips can survive
A rocket ride exposes components to intense vibration and sustained acceleration. Google says it shook the prototype hardware along all three axes to reproduce launch conditions, with some components experiencing forces between 50 and 100 times gravity during testing.
Radiation presents a different threat. High-energy particles can change bits or damage electronics outside Earth’s protective atmosphere. The Suncatcher team tested Trillium TPUs in a proton beam at the University of California, Davis’s Crocker Nuclear Laboratory while the chips ran AI workloads. Google says the processors withstood a total ionizing dose greater than what they would receive during five years in space.
Cooling is another unresolved engineering problem. On Earth, data centers can move heat with air and water systems. In a vacuum, there is no airflow, so the prototype relies on heat pipes and radiators. The team has tested that system in a thermal-vacuum chamber, but the orbital mission will show how it behaves under real conditions.
The satellite network comes later
Project Suncatcher’s long-term concept is a group of solar-powered satellites carrying TPUs and communicating through laser links. Google argues that satellites in low Earth orbit can receive substantially more solar energy than terrestrial systems, but a useful compute network would also need precise, high-bandwidth connections between moving spacecraft.
That network is not what is launching next week. Google says it plans to test satellite-to-satellite laser links in 2027 when it puts two satellites in orbit. Future designs could carry dozens of TPUs per satellite, but those larger clusters remain a research goal rather than a deployed system.
The first mission therefore answers a more basic question: whether Google’s current AI hardware and cooling design can make the trip, tolerate the environment and return useful measurements. Any path from that test to scalable computing in orbit will require additional launches and engineering work.
Technology
