Data center in orbit: Google's Suncatcher satellite sends back first AI response

Google has reached a first milestone in its Project Suncatcher effort to build data centers in space. The prototype satellite, launched on Oct. 1, carries a data-center-class AI accelerator, a Tensor Processing Unit (TPU), and after several weeks of commissioning has delivered the first response from an AI model in orbit. Google said on Friday that the team uplinked a prompt during a ground-station pass to the open Gemma 4 model on board, asking it to write a haiku about being in space. The answer came back after a pass over Svalbard: "Blue marble below / Data streams in silent grace / Starlight feeds my core."

According to TechCrunch, the satellite was built by Planet Labs and launched on a SpaceX Falcon 9 rocket alongside more than 100 other spacecraft. It generates up to one kilowatt of power, but the TPU can run for only about 20 minutes at a time. The haiku was followed by limericks and payload status reports. Next, Google plans to test how the chips cope with thermal stress and space radiation. Radiation can flip individual bits and silently corrupt data; ground tests suggested this would barely affect running trained models but could be a problem for training, TechCrunch reported.

For 2027, Google plans two interconnected satellites with more powerful TPUs and optical laser links between them. The concept of using solar power in space directly for AI computing is not expected to become economical until the 2030s. According to Google's own analysis, that requires much lower launch costs, for example from a fully and rapidly reusable Starship rocket. "Our chips have officially earned their space wings," Travis Beals, a senior director at Google, wrote on the company blog.