Google is about to find out how its AI chips cope with life in orbit. The company says it will launch a prototype satellite carrying its Tensor Processing Units (TPUs) next week, the first in-orbit test for Project Suncatcher, its moonshot to build scalable AI infrastructure in space.
Google announced Project Suncatcher last year as a long-term research effort. The pitch is that satellites in low Earth orbit get near-constant sunlight and can generate up to eight times more solar power than panels on Earth. In time, Google imagines linking constellations of satellites to handle larger AI workloads in orbit.
The prototype will ride on SpaceX’s Transporter-18 rideshare mission and was developed with Planet. Its job is basic: to show whether Google’s AI hardware can survive the stresses of spaceflight, including radiation and extreme temperatures, and to gather data that will shape later launches.
In a new blog post and video series, Travis Beals, Senior Director of Paradigms of Intelligence at Google, walked through the challenges the team is working on.
Surviving launch and radiation
A rocket trip to low Earth orbit lasts about ten minutes. During it, the spacecraft feels sustained loads of up to 10 g, and individual components like the TPU chips can see 50 to 100 g. Google shook the satellite on all three axes to mimic launch vibrations, and the hardware held up better than the team expected.
Radiation is the other big risk. Google tested its Trillium TPUs in a proton beam at UC Davis’s Crocker Nuclear Laboratory while they ran AI workloads, watching how errors like bitflips affected the results. Early results suggest the chips can take a total ionizing dose beyond what they would get on a five-year space mission.
Cooling without air
Cooling may be the hardest problem. TPUs pack a lot of heat into a small area, and in a vacuum there is no airflow, so heat can only be shed through radiators. Google is working on a combination of heat pipes and radiators, and has tested it in a thermal vacuum chamber. The orbital test will show how the cooling system performs in space.
Keeping satellites connected
Future Suncatcher satellites are meant to carry dozens of TPUs each and fly in clusters, linked by lasers. Existing space laser systems are built for low bandwidth over long distances. Google needs very high bandwidth over very short ones, which Beals compares to hitting a coin-sized target from miles away while both ends are moving. That will be tested in 2027, when Google puts two satellites in orbit.
A crowded, contested race
Google isn’t alone. NVIDIA-backed Starcloud has been planning space datacenters, and Elon Musk has merged xAI and SpaceX partly to speed up the idea. Musk believes such datacenters could be a reality in under three years, while Sam Altman has been openly skeptical that they will provide meaningful compute for OpenAI in the next couple of years. China is also reportedly working on it: state aerospace giant CASC has folded space-based datacenters into its 15th Five-Year Plan.
Google is framing the launch as a first step rather than a breakthrough. Beals says the aim is to see what works, find the points of failure, and feed those lessons into future missions.