THE CRUNCH

Google's Project Suncatcher prototype satellite has produced the first AI response from a data-centre-class TPU in orbit. Launched on October 1, the satellite spent roughly three weeks on commissioning, including powering up its host and TPU trays. During a ground-station pass this week, engineers uplinked a prompt to the onboard Gemma 4 open model, which generated a haiku about being in space that was downlinked as the satellite passed over Svalbard, Norway. Google followed up with space-themed limericks and telemetry prompts reporting the payload's technical status.

The satellite carries four TPUs and generates up to 1 kW of electricity, but it can only run one chip at a time for about 20 minutes, a constraint that complicates space-to-ground communications. A spokesperson told TechCrunch the spacecraft is operating nominally.

The next test campaign will examine how the chips cope with thermal stress and radiation exposure. Radiation can cause bit flips, corrupting stored memory values without triggering typical monitoring tools. Google has tested TPUs in particle accelerators on Earth, where the impact appeared minimal for inference workloads and more of a problem for training, but the team will now observe those effects in orbit rather than in simulation.

If the chips prove reliable, Google plans a 2027 milestone of twin interconnected satellites with higher power generation and inter-satellite optical links. Among space data centre ventures, Suncatcher is unusual in planning high-speed links between satellites for coherent distributed workloads; SpaceX's proposed orbital data centres would also use laser links but fly too far apart to share processing effectively, while companies such as Starcloud and Cowboy Space expect to run inference on individual satellites.

WHAT HAPPENS NEXT

Google will run a series of tests evaluating how the TPUs perform under thermal stress and radiation exposure in orbit, ahead of its planned twin interconnected satellites in 2027.