Google is planning to run some of its AI infrastructure from orbit.
The company has unveiled Project Suncatcher, a research effort aimed at placing machine learning infrastructure on satellites powered by the sun. The idea is to move compute-heavy AI workloads off the ground entirely, away from the power grids and cooling systems that data centers currently depend on. Details remain thin - this is a research proposal, not an announced product or launch date. But the ambition alone marks a shift in how seriously Google is treating the physical limits of building AI on Earth.
This matters because power, not chips, is becoming the real bottleneck on AI expansion - data center operators are already fighting utilities and local communities for electricity and water, and orbit is one of the only places with unlimited, uninterrupted sunlight and no neighbors to object. Space-based solar power has been a fantasy since the 1960s, killed every time by launch costs; Google is effectively betting that cheaper, more frequent rocket launches have finally changed that math. If it works, it would be the first time a major cloud provider treats orbit as viable data center real estate rather than just satellite territory.
Plenty of ambitious Google infrastructure projects have quietly died on the way from blog post to hardware - this one still has to survive contact with radiation, thermal cycling, and the fact that getting an AI's answer back down to Earth still requires a data center to catch it.