Google has announced “Project Suncatcher,” a research moonshot that envisions a future where artificial intelligence largely runs on infrastructure in space. The company is planning compact constellations of solar-powered satellites to handle the immense processing demands of AI.
The core concept involves about 80 satellites per constellation, orbiting 400 miles above Earth. These satellites will be equipped with Google’s Tensor Processing Units (TPUs), the specialized hardware that powers its AI models. The satellites will be interconnected using free-space optical links, essentially beaming data between them with lasers.
The primary motivation is to tap into the “unlimited, low-cost renewable energy” of space. Solar panels in orbit are dramatically more efficient—up to eight times more—than on Earth. This provides a clean, continuous power source, a stark contrast to the massive, and often carbon-intensive, energy demands of terrestrial datacenters.
This approach also aims to solve the land and water crisis associated with datacenters. By moving AI compute off-planet, Project Suncatcher would “minimize impact on terrestrial resources,” freeing up land and water currently used for cooling the massive server farms on the ground.
While Google is targeting early 2027 for its first two prototype satellites, it cautions that the path is difficult. The company highlighted “significant engineering challenges,” including thermal management in a vacuum, high-bandwidth ground communications, and long-term reliability, as key problems that still need to be solved.