Google Wants to Send AI Processing to Space

A team of Google researchers have shared a plan for building a machine learning system in space. The AI processing hardware would consist of a network of satellites in low Earth orbit, potentially saving significant space, considering that Google has massive data centers for AI use. The system could see tests in 2027.

Beyond the physical footprint reduction, the satellite AI system would draw power from the Sun, rather than from Earth. Getting Project Suncatcher off the ground (both literally and figuratively) would likely take years, but it has the potential to change the way AI data centers scale up.

The space AI system would include Tensor processing units aboard "fleets of satellites equipped with solar arrays," according to the research team's preprint paper (PDF).

"We illustrate the basic approach to formation flight via a 81-satellite cluster of 1 km radius, and describe an approach for using high-precision ML ...