Back

Google puts its AI chips in orbit with its first Project Suncatcher satellite

Illustration: a computer chip with solar panel wings floating in space above a plain horizon, with sunlight streaming across it
Illustration: FrameSignal (AI-generated)

Google's first Project Suncatcher satellite, built with Planet, launched on Oct. 1 with what Planet calls the first test of Google's own AI chips in space. At launch, Google said it would spend the following weeks studying how the chips handle radiation, launch stress and extreme heat and cold, a first step in long-term research into running AI on solar-powered satellites. On Oct. 30, Google said the satellite had sent back its first answer from its AI chips, and that tests of how they handle heat and radiation come next.

Google has put its own AI chips in orbit, in what Planet calls the first test of them in space. A prototype satellite for Project Suncatcher, built with satellite maker Planet, launched on Thursday, Oct. 1, aboard SpaceX's Transporter-18 rideshare mission from Vandenberg Space Force Base in California. Google says it has confirmed contact with the satellite and that it is operating as expected. Planet, which flew 19 of its own satellites on the same launch, says it has also made initial contact and begun commissioning, the process of checking the satellite out.

The satellite carries Google's Tensor Processing Units, or TPUs, the company's own AI chips. At launch, Google said it would spend the following weeks collecting data on how they cope with the shaking of launch, radiation, and the heat and cold of orbit.

The bigger question is whether AI computing could one day move off the planet. Google calls Suncatcher a long-term research moonshot and says satellites in low Earth orbit can generate up to eight times more solar power than panels on the ground. Its plan is clusters of satellites, each carrying dozens of TPU chips, that talk to each other by laser. Google says it will test those laser links in 2027 with two satellites in orbit. Google says a peer-reviewed paper on the research is now out in the journal Joule.

So far, the results Google has shared come from its own tests on the ground. Radiation from the sun and cosmic rays can damage electronics, so Google says it ran its Trillium TPUs in a proton beam at UC Davis while they worked on AI tasks. It says initial results show they can survive more radiation than a five-year mission would deliver. Individual chips can face forces of 50 to 100 times gravity during launch, Google says, and the satellite held up when engineers shook it on the ground to mimic one. Cooling works differently in space, where there is no air to carry heat away and heat can leave only through radiators. Google says it has tested its heat pipes and radiators only in a vacuum chamber on the ground so far.

This is a single test satellite, not a data center. Google has not said how many chips are on board, how long the satellite will operate beyond the first weeks of data collection, what it cost, or when it will publish the results. Whether computing in orbit could ever compete on cost with data centers on the ground is still unknown.

Update (Nov. 14): on Oct. 30, Google said it had finished its checks of the satellite, including powering up the AI chips and running early tests, and that the satellite had sent back its first answer from an AI model running on them, a short poem about being in space. Google says it will next test how the chips perform under heat stress and radiation in orbit. It still plans two linked satellites in 2027.

More on Google