Google Tests Space-Based AI Data Centers with Project Suncatcher
Google has launched Project Suncatcher, an ambitious effort to create AI data centers in space. The project's first step is a refrigerator-sized satellite that contains four Tensor Processing Units (TPUs), which will run a version of Google's Gemma AI model for 15 minutes at a time.
The TPUs were specifically designed by Google for machine learning and have already been deployed in data centers on Earth. However, the company notes that cooling the chips is a crucial research challenge due to the heat management constraints in space.
Project Suncatcher aims to harness the virtually unlimited free energy from the sun to power AI compute in space. Travis Beals, senior director and lead of the project, explained that the goal is to tap into the best way to use solar power for this purpose.