Google has laid out a 10-year strategic roadmap for its next-generation AI infrastructure, and it’s not on Earth. “Project Suncatcher” begins with “two prototype satellites by early 2027” and aims for cost-parity with terrestrial datacenters by “the middle of the 2030s.”
This “moonshot” research is Google’s answer to the $3 trillion, resource-draining AI datacentre boom. The 2027 prototypes will serve as the “first milestone,” tasked with solving the “significant engineering challenges” of thermal management, reliability, and ground communication.
If these prototypes are successful, the next phase will be to build the “compact constellations of about 80” solar-powered satellites. These satellites, carrying Google’s TPUs and connected by optical links, will form the “scalable space-based AI” network.
The economic endgame of this plan is the mid-2030s. By then, Google’s research projects that “falling cost of rocket launches” will make the total running cost of this orbital system “comparable to one on Earth.” At that point, the “unlimited” 8x-more-efficient solar power would make space the obvious choice.
This long-term, two-stage plan—proving the tech by 2027, achieving economic viability by 2035—is Google’s “working backward” strategy to eventually make space the “best place to scale AI computers.”