Google Sends Its First TPU Into Orbit to Test Space-Based AI Compute

Google has sent one of its advanced chips into space for the first time, with a prototype orbital compute satellite built by Planet Labs lifting off aboard a SpaceX rocket from California. The mission will test whether a Google Tensor Processing Unit can operate in orbit, including drawing a kilowatt of continuous power, staying cool and running a series of AI models.

Travis Beals, who manages Google’s Project Suncatcher, said ground testing cannot fully replicate real conditions. The satellite will run its TPU in 15-minute bursts to protect its power and thermal systems, while a follow-up demo next year will fly two purpose-built satellites linked by laser to handle heavier workloads. Beals described the effort as a long-term moonshot aimed at an orbital data center of 81 satellites flying in close formation, noting that the rockets needed to scale cost-effectively do not yet exist.

Google also released a peer-reviewed white paper, to be published in Joule, projecting that SpaceX could reach launch prices near $200 per kilogram by 2035, which would require around 1,800 Starship launches. Updated radiation tests suggest Google’s chips can handle large inference workloads in orbit over a five-year lifespan, though Beals said they remain unsuited to massive training runs.