Google’s ‘Project Suncatcher’ aims to build solar-powered AI supercomputers in space

Google is setting its sights beyond Earth with Project Suncatcher, an ambitious new research initiative designed to explore the idea of running large-scale machine learning directly in space. The project envisions a future where Google’s Tensor Processing Units (TPUs) are deployed aboard satellites powered almost entirely by sunlight, creating a vast orbital compute network.

According to internal research materials, Project Suncatcher’s goal is to determine whether AI compute could eventually scale more efficiently in space than it does in terrestrial data centres. The concept relies on sun-synchronous low-Earth orbits, where satellites maintain near-continuous exposure to sunlight. In such conditions, solar panels can be up to eight times more productive than on Earth, reducing the dependency on batteries and improving energy efficiency.

These satellites would communicate through free-space optical links, distributing machine learning tasks across multiple TPUs at high bandwidth and low latency. To replicate Earth-based data centre performance, the links would need to achieve transfer speeds in the tens of terabits per second, with satellites flying in tight formations just a few kilometres—or even hundreds of metres—apart.

Radiation testing on Google’s Trillium TPU v6e chips has already yielded promising results. The chips showed no major degradation up to radiation doses roughly three times higher than expected during a five-year mission, suggesting they could operate reliably in orbit.