Google has launched its first prototype orbital compute satellite aboard a SpaceX rocket, marking the first time the tech giant has sent one of its advanced chips into space. The mission serves as a critical test for Project Suncatcher, Google’s initiative to develop large-scale compute clusters in orbit.
What Happened
Built by Planet Labs, the satellite carries a Google Tensor Processing Unit (TPU), the company’s competitor to Nvidia’s GPUs. The primary objective is to verify that the chip can function in space by supplying it with a kilowatt of continuous power, managing its cooling, and running various models to identify potential failures. Travis Beals, the Google executive managing Project Suncatcher, noted that while ground testing is useful, "there’s no test that’s completely as good as the real thing." Once commissioned, the satellite will operate its TPU in 15-minute bursts to prevent straining its power and thermal management systems. This prototype is based on a standard Planet Labs platform, but Google and Planet Labs are preparing a more advanced demo for next year that will feature two purpose-built satellites capable of running substantial workloads and collaborating via a laser communications link.
Why It Matters
Google’s involvement distinguishes Project Suncatcher from other space AI startups by focusing on long-term infrastructure rather than immediate commercial viability. The company envisions an orbital data center composed of a network of 81 satellites flying in close formation to process data in parallel. According to a peer-reviewed white paper released by Google and set to be published in Joule, the success of this vision depends heavily on launch costs. Google researchers estimate that SpaceX has achieved a price-reducing "learning curve" of approximately 20% annually since the Falcon 1 launch. To make orbital data centers cost-effective, the paper suggests SpaceX needs to reach launch prices close to $200 per kilogram by 2035. Achieving this trajectory would require Starship to fly 370,000 tons of payload into orbit, which translates to roughly 1,800 launches over the next decade, or about 180 launches per year. This volume is significantly higher than SpaceX’s current capabilities, as the vehicle has never flown more than five times in a single year.
The Bottom Line
Google’s research indicates that while its TPUs are likely resilient enough to handle large inference workloads in orbit for a five-year lifespan, the economics of space-based data centers remain a long-term challenge dependent on massive increases in SpaceX’s launch frequency. The company acknowledges that radiation errors, while low for inference, remain problematic for mega-scale training runs, and the necessary launch cadence for Starship is still a substantial hurdle.