Google's Suncatcher Satellite Sends First AI Output From Orbit

What Suncatcher is meant to prove

The satellite is part of Google's research project Suncatcher, which examines whether space could one day host scalable infrastructure for machine learning. It carries four Tensor Processing Units (TPUs) and solar panels that deliver up to one kilowatt.

That power is also the limit. Because of the chips' energy demand, only one TPU can run at a time, and only in short intervals of about 15 minutes. Continuous operation would overload the satellite's power supply and cooling, Google says.

If the onboard TPUs prove themselves, Google plans to launch two more satellites in 2027 that can generate more power. In the longer term, several satellites are meant to form a connected orbital computing network.

A crowded race for orbital compute

Google is not alone. In January, SpaceX filed an application with the US Federal Communications Commission to put up to one million data centers into Earth orbit. US startup Starcloud launched a satellite with an Nvidia chip last winter and aims to operate orbital data centers as large as terrestrial ones by 2030.

Supporters argue that the AI boom is straining power grids and raising water demand for cooling on Earth, while excess heat in space could be released into the cold vacuum. Critics counter that the carbon footprint of orbital facilities could exceed that of ground-based ones once manufacturing, launch and disposal are counted.