Google Taps Solar Power with Space-Based AI Data Centers

Google has launched Project Suncatcher, an ambitious research project to develop AI data centers in space. The first satellite, about the size of a refrigerator, has been successfully placed into orbit and is equipped with Tensor Processing Units (TPUs) designed by Google for machine learning. These TPUs will run a version of Google's Gemma AI model for 15 minutes at a time due to heat management constraints.

The mission aims to test how the TPUs handle the physical stress of spaceflight and radiation in space. The satellite is in a sun-synchronous orbit, where its solar panels will always be exposed to sunlight, eliminating the need for heavy batteries or backup power. Google plans to put two more satellites into orbit next year to test their connection.

The project's senior director, Travis Beals, said that tapping into solar power is the best way to run AI compute and that space data centers would be virtually unlimited in energy supply. However, there are challenges to overcome, including heat management, which could limit the performance of computer chips in space. Brandon Lucia, a professor at Carnegie Mellon University, noted that while some technology has been proven at a research level, huge challenges remain, particularly with cooling and maintenance.