Google Tests the Waters with Space-Based AI, a More Cautious Approach

Google's Project Suncatcher is taking a cautious approach to AI in space. Unlike SpaceX and Nvidia, which aim to place full AI data centers in orbit by late 2027, Google plans to launch a test satellite with Tensor Processing Units (TPUs) to see if they can work in space.

The prototype satellite will carry four TPUs, but these chips can only run for 15-minute bursts before shutting down to cool. This is because the cooling problem is quite difficult, as Brandon Lucia, a Carnegie Mellon University professor of electrical and computer engineering, pointed out: 'These computers will be basking in the sun all day.'

The satellite will endure intense vibration and sustained acceleration of up to 10 times Earth gravity during launch. Google wants to know if their TPUs can work in these conditions before investing more resources into AI in space.