Google is rethinking satellite constellations, proposing a new model designed not for internet access, but for high-performance AI computing. “Project Suncatcher” outlines a plan for “compact constellations of about 80 solar-powered satellites” orbiting 400 miles above Earth.
These constellations would function as distributed datacenters in the sky. Each satellite would be a node in the network, equipped with Google’s powerful Tensor Processing Units (TPUs) to handle complex AI tasks. This marks a shift from communication relays to active processing hubs in orbit.
The satellites within these constellations would be “connected by free-space optical links.” This laser-based communication system would allow them to exchange data at high speeds, effectively working together as a single, powerful AI computer. The final results would then be beamed down to Earth.
The 80-satellite figure suggests a “tightly packed” formation, designed to maximize processing density and a dedicated power infrastructure. This network would be powered by solar panels that, in orbit, are eight times more productive than on Earth, providing the massive energy load required by the TPUs.
While Google’s first two prototypes are set for 2027, this constellation-based architecture is what the company believes is “a scalable space-based AI.” It is this 80-satellite hub model that must prove it can overcome the immense challenges of thermal management and on-orbit reliability.
A New Kind of Constellation: Google’s Plan for 80-Satellite AI Hubs
38