A “Scalable Space-Based AI”: Google’s 2027 Milestone

Google has a new buzzword for its AI ambitions: “scalable space-based AI.” This is the ultimate goal of its “Project Suncatcher,” and the “first milestone” on this journey is a planned “two prototype” launch by early 2027.
The key word is “scalable.” The “rising demand for AI” has created a scalability crisis. The $3 trillion terrestrial datacentre model is not scaling “cleanly”—it’s causing “rising concern” over carbon emissions and resource use.
Google’s “moonshot” research concludes that space “may be the ‘best place to scale AI computers.” The “scalability” comes from two factors: “unlimited” solar power (which is 8x more productive) and freedom from Earth’s land and water constraints.
This “scalable space-based AI” would consist of “compact constellations of solar-powered satellites,” equipped with Google TPUs and connected by optical links. This is the architecture Google believes can scale to meet the demand.
But this is all theoretical. The 2027 “first milestone” is designed to test the “significant engineering challenges” (thermal, reliability, comms) that stand in the way. “Scalable” is the goal, but “survivable” is the first step.