Google is testing a far-out idea: running artificial intelligence computers in space. Its Project Suncatcher prototype satellite, carrying four of Google's Trillium tensor processing units, or TPUs, launched Oct. 1 aboard SpaceX's Transporter-18 rideshare mission, Google said on its blog. GCN and The Register also reported the launch.

Google says it has made contact with the satellite and that it is operating as expected. The test will look at three big questions: whether the chips survive the shaking of launch, how they hold up against radiation in orbit, and how well they can shed heat, since there is no air in space to carry it away.

Why put AI chips in orbit? Data centers on Earth use huge amounts of electricity and water. In the right orbit, solar panels can soak up sunlight almost all the time, giving a steady supply of power. Google's researchers have argued that, in the long run, clusters of satellites could run AI work powered by the sun.

There are big hurdles. Launch costs, repairs, space debris and the need to send data back and forth quickly all make space computing hard. Google plans a 2027 follow-up with two satellites to test high-speed laser links between them, a key step toward linking many satellites into one system.

Here is the background. Google first described Project Suncatcher in late 2025 as a "moonshot" research effort. Other companies, including startups and rivals such as SpaceX, have floated similar ideas as demand for AI computing soars.

The project is an early experiment, not a product. It will likely take many years before any space data center handles real work.

What we know: Four Google AI chips are in orbit and working as expected.
What we don't: Whether space computing can ever compete on cost with data centers on the ground.