Google’s Project Suncatcher moonshot to send AI chips into space for first time

Google plans to send a prototype satellite carrying four of its AI chips into low Earth orbit next week, marking the first real-world test of its long-term Project Suncatcher initiative. The mission aims to find out whether space could eventually host large-scale artificial intelligence computing infrastructure.

The prototype will launch aboard SpaceX’s Transporter-18 rideshare mission and was developed in partnership with the aerospace company Planet. Travis Beals, a senior director at Google Research, said the launch is meant to reveal what works, identify points of failure, and inform future missions rather than prove the entire concept at once.

Chasing Constant Sunlight

Project Suncatcher, first announced last year, explores whether satellites equipped with Google’s Tensor Processing Units could one day take advantage of near-constant sunlight in low Earth orbit. According to the company, satellites positioned there can generate up to eight times more solar power than equivalent panels on Earth. The eventual goal involves linking multiple satellite constellations together to handle larger AI workloads collectively.

Surviving trip up

Before any chip can compute anything useful in orbit, it first has to survive getting there. A rocket ride into low Earth orbit lasts roughly 10 minutes and subjects the spacecraft to intense vibration and sustained forces up to 10 times the pull of gravity, with individual components briefly experiencing forces as high as 50 to 100 times that pull.

Google said it shook a test satellite along all three axes to mimic those launch frequencies, and the hardware held up better than expected. Separately, the company exposed its Trillium TPU chips to a proton beam at UC Davis’s Crocker Nuclear Laboratory while running AI workloads, tracking how radiation-induced errors called bit flips affected performance. The chips reportedly survived a total radiation dose greater than what they would absorb during a five-year mission in space.

Cooling without air

Space presents a cooling problem terrestrial data centers never face. TPUs generate substantial heat in a small area, but there is no air in orbit to carry that heat away, leaving radiators as essentially the only option. Google said it is testing a combination of heat pipes and radiators, verified so far inside a thermal vacuum chamber that mimics the temperature and vacuum conditions of space.

Linking satellites by laser

Future versions of the system would pack dozens of TPU chips onto each satellite, with clusters communicating through lasers rather than radio. That requires extremely high bandwidth over very short distances, a much different demand than most existing satellite laser links are built for. Google compared the required precision to hitting a coin-sized target from miles away while both the sender and target are moving. The company plans to test two satellites equipped with this laser link in 2027.

Google isn’t alone in eyeing orbital AI infrastructure. Starcloud launched an Nvidia H100 GPU into space in November 2025, and SpaceX has made space-based AI computing a central part of its own business pitch this year.

Source: https://interestingengineering.com/space/google-project-suncatcher-ai-chips-space-launch