The Orbital Datacenter: Google’s Plan for a New Class of AI Infrastructure

Google is proposing a new class of infrastructure to meet the demands of artificial intelligence: the orbital datacenter. Under “Project Suncatcher,” the company is researching constellations of AI-processing satellites as a long-term solution to the scaling challenges faced on Earth.
This “moonshot” initiative envisions about 80 solar-powered satellites per constellation, orbiting 400 miles high. These satellites would be equipped with Google’s custom AI processors (TPUs) and connected by laser-based optical links, forming a cohesive data processing network in space.
The primary driver is to escape the physical and environmental constraints of Earth. In space, solar power is eight times more potent, providing an unlimited, clean energy source. This approach also completely bypasses the need for the massive amounts of land and water that terrestrial datacenters consume for cooling.
This isn’t just a theoretical concept. Google is planning to launch its first two prototypes by early 2027. This follows similar announcements from Elon Musk and Nvidia, which is partnering with Starcloud to launch AI chips this month. The tech industry is clearly converging on space as a viable solution.
However, Google’s own research sounds a “cautionary note.” The company states that “significant engineering challenges remain,” specifically citing thermal management, ground communication, and on-orbit reliability. The 2027 trials will be the first test of whether these hurdles can be overcome.